US20160371616A1 - Individualized Predictive Model & Workflow for an Asset - Google Patents

Abstract

Disclosed herein are systems, devices, and methods related to assets and predictive models and corresponding workflows that are related to the operation of assets. In particular, examples involve defining and deploying aggregate, predictive models and corresponding workflows, defining and deploying individualized, predictive models and/or corresponding workflows, and dynamically adjusting the execution of model-workflow pairs.

Description

CROSS REFERENCE TO RELATED APPLICATIONS
• This application incorporates by reference U.S. Non-Provisional patent application Ser. No. 14/732,258, filed on Jun. 5, 2015, entitled Asset Health Score in its entirety. This application is also related to the following applications filed on the same day as the present application, each of which is incorporated by reference in its entirety: U.S. Non-Provisional patent application Ser. No. ______ (Attorney Docket No. Uptake-00012), entitled Aggregate Predictive Model & Workflow for Local Execution; and U.S. Non-Provisional patent application Ser. No. ______ (Attorney Docket No. Uptake-00013), entitled Dynamic Execution of Predictive Models & Workflows.
BACKGROUND
• Today, machines (also referred to herein as “assets”) are ubiquitous in many industries. From locomotives that transfer cargo across countries to medical equipment that helps nurses and doctors to save lives, assets serve an important role in everyday life. Depending on the role that an asset serves, its complexity, and cost, may vary. For instance, some assets may include multiple subsystems that must operate in harmony for the asset to function properly (e.g., an engine, transmission, etc. of a locomotive).
• Because of the key role that assets play in everyday life, it is desirable for assets to be repairable with limited downtime. Accordingly, some have developed mechanisms to monitor and detect abnormal conditions within an asset to facilitate repairing the asset, perhaps with minimal downtime.
OVERVIEW
• The current approach for monitoring assets generally involves an on-asset computer that receives signals from various sensors and/or actuators distributed throughout an asset that monitor the operating conditions of the asset. As one representative example, if the asset is a locomotive, the sensors and/or actuators may monitor parameters such as temperatures, voltages, and speeds, among other examples. If sensor and/or actuator signals from one or more of these devices reach certain values, the on-asset computer may then generate an abnormal-condition indicator, such as a “fault code,” which is an indication that an abnormal condition has occurred within the asset.
• In general, an abnormal condition may be a defect at an asset or component thereof, which may lead to a failure of the asset and/or component. As such, an abnormal condition may be associated with a given failure, or perhaps multiple failures, in that the abnormal condition is symptomatic of the given failure or failures. In practice, a user typically defines the sensors and respective sensor values associated with each abnormal-condition indicator. That is, the user defines an asset's “normal” operating conditions (e.g., those that do not trigger fault codes) and “abnormal” operating conditions (e.g., those that trigger fault codes).
• After the on-asset computer generates an abnormal-condition indicator, the indicator and/or sensor signals may be passed to a remote location where a user may receive some indication of the abnormal condition and/or sensor signals and decide whether to take action. One action that the user might take is to assign a mechanic or the like to evaluate and potentially repair the asset. Once at the asset, the mechanic may connect a computing device to the asset and operate the computing device to cause the asset to utilize one or more local diagnostic tools to facilitate diagnosing the cause of the generated indicator.
• While current asset-monitoring systems are generally effective at triggering abnormal-condition indicators, such systems are typically reactionary. That is, by the time the asset-monitoring system triggers an indicator, a failure within the asset may have already occurred (or is about to occur), which may lead to costly downtime, among other disadvantages. Additionally, due to the simplistic nature of on-asset abnormality-detection mechanisms in such asset-monitoring systems, current asset-monitoring approaches tend to involve a remote computing system performing monitoring computations for an asset and then transmitting instructions to the asset if a problem is detected. This may be disadvantageous due to network latency and/or infeasible when the asset moves outside of coverage of a communication network. Further still, due to the nature of local diagnostic tools stored on assets, current diagnosis procedures tend to be inefficient and cumbersome because a mechanic is required to cause the asset to utilize such tools.
• The example systems, devices, and methods disclosed herein seek to help address one or more of these issues. In example implementations, a network configuration may include a communication network that facilitates communications between assets and a remote computing system. In practice, the communication network may facilitate secure communications between assets and the remote computing system (e.g., via encryption or other security measures).
• As noted above, each asset may include multiple sensors and/or actuators distributed throughout the asset that facilitate monitoring operating conditions of the asset. A number of assets may provide respective data indicative of each asset's operating conditions to the remote computing system, which may be configured to perform one or more operations based on the provided data.
• In example implementations, the remote computing system may be configured to define and deploy to assets a predictive model and corresponding workflow (referred to herein as a “model-workflow pair”) that are related to the operation of the assets. The assets may be configured to receive the model-workflow pair and utilize a local analytics device to operate in accordance with the model-workflow pair.
• Generally, a model-workflow pair may cause an asset to monitor certain operating conditions and when certain conditions exist, modify a behavior that may help facilitate preventing an occurrence of a particular event. Specifically, a predictive model may receive as inputs sensor data from a particular set of asset sensors and output a likelihood that one or more particular events could occur at the asset within a particular period of time in the future. A workflow may involve one or more operations that are performed based on the likelihood of the one or more particular events that is output by the model.
• In practice, the remote computing system may define an aggregate, predictive model and corresponding workflows, individualized, predictive models and corresponding workflows, or some combination thereof. An “aggregate” model/workflow may refer to a model/workflow that is generic for a group of assets, while an “individualized” model/workflow may refer to a model/workflow that is tailored for a single asset or subgroup of assets from the group of assets.
• In example implementations, the remote computing system may start by defining an aggregate, predictive model based on historical data for multiple assets. Utilizing data for multiple assets may facilitate defining a more accurate predictive model than utilizing operating data for a single asset.
• The historical data that forms the basis of the aggregate model may include at least operating data that indicates operating conditions of a given asset. Specifically, operating data may include abnormal-condition data identifying instances when failures occurred at assets and/or sensor data indicating one or more physical properties measured at the assets at the time of those instances. The data may also include environment data indicating environments in which assets have been operated and scheduling data indicating dates and times when assets were utilized, among other examples of asset-related data used to define the aggregate model-workflow pair.
• Based on the historical data, the remote computing system may define an aggregate model that predicts the occurrence of particular events. In a particular example implementation, an aggregate model may output a probability that a failure will occur at an asset within a particular period of time in the future. Such a model may be referred to herein as a “failure model.” Other aggregate models may predict the likelihood that an asset will complete a task within a particular period of time in the future, among other example predictive models.
• After defining the aggregate model, the remote computing system may then define an aggregate workflow that corresponds to the defined aggregate model. Generally, a workflow may include one or more operations that an asset may perform based on a corresponding model. That is, the output of the corresponding model may cause the asset to perform workflow operations. For instance, an aggregate model-workflow pair may be defined such that when the aggregate model outputs a probability within a particular range an asset will execute a particular workflow operation, such as a local diagnostic tool.
• After the aggregate model-workflow pair is defined, the remote computing system may transmit the pair to one or more assets. The one or more assets may then operate in accordance with the aggregate model-workflow pair.
• In example implementations, the remote computing system may be configured to further define an individualized predictive model and/or corresponding workflow for one or multiple assets. The remote computing system may do so based on certain characteristics of each given asset, among other considerations. In example implementations, the remote computing system may start with an aggregate model-workflow pair as a baseline and individualize one or both of the aggregate model and workflow for the given asset based on the asset's characteristics.
• In practice, the remote computing system may be configured to determine asset characteristics that are related to the aggregate model-workflow pair (e.g., characteristics of interest). Examples of such characteristics may include asset age, asset usage, asset class (e.g., brand and/or model), asset health, and environment in which an asset is operated, among other characteristics.
• Then, the remote computing system may determine characteristics of the given asset that correspond to the characteristics of interest. Based at least on some of the given asset's characteristics, the remote computing system may be configured to individualize the aggregate model and/or corresponding workflow.
• Defining an individualized model and/or workflow may involve the remote computing system making certain modifications to the aggregate model and/or workflow. For example, individualizing the aggregate model may involve changing model inputs, changing a model calculation, and/or changing a weight of a variable or output of a calculation, among other examples. Individualizing the aggregate workflow may involve changing one or more operations of the workflow and/or changing the model output value or range of values that triggers the workflow, among other examples.
• After defining an individualized model and/or workflow for the given asset, the remote computing system may then transmit the individualized model and/or workflow to the given asset. In a scenario where only one of the model or workflow is individualized, the given asset may utilize the aggregate version of the model or workflow that is not individualized. The given asset may then operate in accordance with its individualized model-workflow pair.
• In example implementations, a given asset may include a local analytics device that may be configured to cause the given asset to operate in accordance with a model-workflow pair provided by the remote computing system. The local analytics device may be configured to utilize operating data generated by the asset sensors and/or actuators (e.g., data that is typically utilized for other asset-related purposes) to run the predictive model. When the local analytics device receives certain operating data, it may execute the model and depending on the output of the model, may execute the corresponding workflow.
• Executing the corresponding workflow may help facilitate preventing an undesirable event from occurring at the given asset. In this way, the given asset may locally determine that an occurrence of a particular event is likely and may then execute a particular workflow to help prevent the occurrence of the event. This may be particularly useful if communication between the given asset and remote computing system is hindered. For example, in some situations, a failure might occur before a command to take preventative actions reaches the given asset from the remote computing system. In such situations, the local analytics device may be advantageous in that it may generate the command locally, thereby avoiding any network latency or any issues arising from the given asset being “off-line.” As such, the local analytics device executing a model-workflow pair may facilitate causing the asset to adapt to its conditions.
• While a given asset is operating in accordance with a model-workflow pair, the given asset may also continue to provide operating data to the remote computing system. Based at least on this data, the remote computing system may modify the aggregate model-workflow pair and/or one or more individualized model-workflow pairs. The remote computing system may make modifications for a number of reasons.
• In one example, the remote computing system may modify a model and/or workflow if a new event occurred at an asset that the model did not previously account for. For instance, in a failure model, the new event may be a new failure that had yet to occur at any of the assets whose data was used to define the aggregate model.
• In another example, the remote computing system may modify a model and/or workflow if an event occurred at an asset under operating conditions that typically do not cause the event to occur. For instance, returning again to a failure model, the failure model or corresponding workflow may be modified if a failure occurred under operating conditions that had yet to cause the failure to occur in the past.
• In yet another example, the remote computing system may modify a model and/or workflow if an executed workflow failed to prevent an occurrence of an event. Specifically, the remote computing system may modify the model and/or workflow if the output of the model caused an asset to execute a workflow aimed to prevent the occurrence of an event but the event occurred at the asset nonetheless. Other examples of reasons for modifying a model and/or workflow are also possible.
• The remote computing system may then distribute any modifications to the asset whose data caused the modification and/or to other assets in communication with the remote computing system. In this way, the remote computing system may dynamically modify models and/or workflows and distribute these modifications to a whole fleet of assets based on operating conditions of an individual asset.
• In some example implementations, an asset and/or the remote computing system may be configured to dynamically adjust executing a predictive model and/or workflow. In particular, the asset and/or remote computing system may be configured to detect certain events that trigger a change in responsibilities with respect to whether the asset and/or the remote computing system are executing a predictive model and/or workflow.
• For instance, in some cases, after the asset receives a model-workflow pair from the remote computing system, the asset may store the model-workflow pair in data storage but then may rely on the remote computing system to centrally execute part or all of the model-workflow pair. On the other hand, in other cases, the remote computing system may rely on the asset to locally execute part or all of the model-workflow pair. In yet other cases, the remote computing system and the asset may share in the responsibilities of executing the model-workflow pair.
• In any event, at some point in time, certain events may occur that trigger the asset and/or remote computing system to adjust the execution of the predictive model and/or workflow. For instance, the asset and/or remote computing system may detect certain characteristics of a communication network that couples the asset to the remote computing system. Based on the characteristics of the communication network, the asset may adjust whether it is locally executing a predictive model and/or workflow and the remote computing system may accordingly modify whether it is centrally executing the model and/or workflow. In this way, the asset and/or remote computing system may adapt to conditions of the asset.
• In a particular example, the asset may detect an indication that a signal strength of a communication link between the asset and the remote computing system is relatively weak (e.g., the asset may determine that is about to go “off-line”), that a network latency is relatively high, and/or that a network bandwidth is relatively low. Accordingly, the asset may be programmed to take on responsibilities for executing the model-workflow pair that were previously being handled by the remote computing system. In turn, the remote computing system may cease centrally executing some or all of the model-workflow pair. In this way, the asset may locally execute the predictive model and then, based on executing the predictive model, execute the corresponding workflow to potentially help prevent an occurrence of a failure at the asset.
• Moreover, in some implementations, the asset and/or the remote computing system may similarly adjust executing (or perhaps modify) a predictive model and/or workflow based on various other considerations. For example, based on the processing capacity of the asset, the asset may adjust locally executing a model-workflow pair and the remote computing system may accordingly adjust as well. In another example, based on the bandwidth of the communication network coupling the asset to the remote computing system, the asset may execute a modified workflow (e.g., transmitting data to the remote computing system according to a data-transmission scheme with a reduced transmission rate). Other examples are also possible.
• As discussed above, examples provided herein are related to deployment and execution of predictive models. In one aspect, a computing system is provided. The computing system comprises at least one processor, a non-transitory computer-readable medium, and program instructions stored on the non-transitory computer-readable medium that are executable by the at least one processor to cause the computing system to: (a) receive operating data for a plurality of assets, wherein the plurality of assets comprises a first asset, (b) based on the received operating data, define an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets, (c) determine one or more characteristics of the first asset, (d) based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, define at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset, and (e) transmit to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.
• In another aspect, a non-transitory computer-readable medium is provided having instructions stored thereon that are executable to cause a computing system to: (a) receive operating data for a plurality of assets, wherein the plurality of assets comprises a first asset, (b) based on the received operating data, define an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets, (c) determine one or more characteristics of the first asset, (d) based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, define at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset, and (e) transmit to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.
• In yet another aspect, a computer-implemented method is provided. The method comprises: (a) receiving operating data for a plurality of assets, wherein the plurality of assets comprises a first asset, (b) based on the received operating data, defining an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets, (c) determining one or more characteristics of the first asset, (d) based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, defining at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset, and (e) transmitting to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.
• One of ordinary skill in the art will appreciate these as well as numerous other aspects in reading the following disclosure.
BRIEF DESCRIPTION OF THE DRAWINGS
• FIG. 1 depicts an example network configuration in which example embodiments may be implemented.
• FIG. 2 depicts a simplified block diagram of an example asset.
• FIG. 3 depicts a conceptual illustration of example abnormal-condition indicators and triggering criteria.
• FIG. 4 depicts a simplified block diagram of an example analytics system.
• FIG. 5 depicts an example flow diagram of a definition phase that may be used for defining model-workflow pairs.
• FIG. 6A depicts a conceptual illustration of an aggregate model-workflow pair.
• FIG. 6B depicts a conceptual illustration of an individualized model-workflow pair.
• FIG. 6C depicts a conceptual illustration of another individualized model-workflow pair.
• FIG. 6D depicts a conceptual illustration of a modified model-workflow pair.
• FIG. 7 depicts an example flow diagram of a modeling phase that may be used for defining a predictive model that outputs a health metric.
• FIG. 8 depicts a conceptual illustration of data utilized to define a model.
• FIG. 9 depicts an example flow diagram of a local-execution phase that may be used for locally executing a predictive model.
• FIG. 10 depicts an example flow diagram of a modification phase that may be used for modifying model-workflow pairs.
• FIG. 11 depicts an example flow diagram of an adjustment phase that may be used for adjusting execution of model-workflow pairs.
• FIG. 12 depicts a flow diagram of an example method for defining and deploying an aggregate, predictive model and corresponding workflow
• FIG. 13 depicts a flow diagram of an example method for defining and deploying an individualized, predictive model and/or corresponding workflow
• FIG. 14 depicts a flow diagram of an example method for dynamically modifying the execution of model-workflow pairs.
DETAILED DESCRIPTION
• The following disclosure makes reference to the accompanying figures and several exemplary scenarios. One of ordinary skill in the art will understand that such references are for the purpose of explanation only and are therefore not meant to be limiting. Part or all of the disclosed systems, devices, and methods may be rearranged, combined, added to, and/or removed in a variety of manners, each of which is contemplated herein.
• I. Example Network Configuration
• Turning now to the figures,FIG. 1 depicts anexample network configuration100 in which example embodiments may be implemented. As shown, thenetwork configuration100 includes anasset102, anasset104, acommunication network106, aremote computing system108 that may take the form of an analytics system, anoutput system110, and adata source112.
• Thecommunication network106 may communicatively connect each of the components in thenetwork configuration100. For instance, theassets102 and104 may communicate with theanalytics system108 via thecommunication network106. In some cases, theassets102 and104 may communicate with one or more intermediary systems, such as an asset gateway (not pictured), that in turn communicates with theanalytics system108. Likewise, theanalytics system108 may communicate with theoutput system110 via thecommunication network106. In some cases, theanalytics system108 may communicate with one or more intermediary systems, such as a host server (not pictured), that in turn communicates with theoutput system110. Many other configurations are also possible. In example cases, thecommunication network106 may facilitate secure communications between network components (e.g., via encryption or other security measures).
• In general, theassets102 and104 may take the form of any device configured to perform one or more operations (which may be defined based on the field) and may also include equipment configured to transmit data indicative of one or more operating conditions of the given asset. In some examples, an asset may include one or more subsystems configured to perform one or more respective operations. In practice, multiple subsystems may operate in parallel or sequentially in order for an asset to operate.
• Example assets may include transportation machines (e.g., locomotives, aircraft, passenger vehicles, semi-trailer trucks, ships, etc.), industrial machines (e.g., mining equipment, construction equipment, factory automation, etc.), medical machines (e.g., medical imaging equipment, surgical equipment, medical monitoring systems, medical laboratory equipment, etc.), and utility machines (e.g., turbines, solar farms, etc.), among other examples. Those of ordinary skill in the art will appreciate that these are but a few examples of assets and that numerous others are possible and contemplated herein.
• In example implementations, theassets102 and104 may each be of the same type (e.g., a fleet of locomotives or aircrafts, a group of wind turbines, or a set of MRI machines, among other examples) and perhaps may be of the same class (e.g., same brand and/or model). In other examples, theassets102 and104 may differ by type, by brand, by model, etc. The assets are discussed in further detail below with reference toFIG. 2.
• As shown, theassets102 and104, and perhaps thedata source112, may communicate with theanalytics system108 via thecommunication network106. In general, thecommunication network106 may include one or more computing systems and network infrastructure configured to facilitate transferring data between network components. Thecommunication network106 may be or may include one or more Wide-Area Networks (WANs) and/or Local-Area Networks (LANs), which may be wired and/or wireless and support secure communication. In some examples, thecommunication network106 may include one or more cellular networks and/or the Internet, among other networks. Thecommunication network106 may operate according to one or more communication protocols, such as LTE, CDMA, GSM, LPWAN, WiFi, Bluetooth, Ethernet, HTTP/S, TCP, CoAP/DTLS and the like. Although thecommunication network106 is shown as a single network, it should be understood that thecommunication network106 may include multiple, distinct networks that are themselves communicatively linked. Thecommunication network106 could take other forms as well.
• As noted above, theanalytics system108 may be configured to receive data from theassets102 and104 and thedata source112. Broadly speaking, theanalytics system108 may include one or more computing systems, such as servers and databases, configured to receive, process, analyze, and output data. Theanalytics system108 may be configured according to a given dataflow technology, such as TPL Dataflow or NiFi, among other examples. Theanalytics system108 is discussed in further detail below with reference toFIG. 3.
• As shown, theanalytics system108 may be configured to transmit data to theassets102 and104 and/or to theoutput system110. The particular data transmitted may take various forms and will be described in further detail below.
• In general, theoutput system110 may take the form of a computing system or device configured to receive data and provide some form of output. Theoutput system110 may take various forms. In one example, theoutput system110 may be or include an output device configured to receive data and provide an audible, visual, and/or tactile output in response to the data. In general, an output device may include one or more input interfaces configured to receive user input, and the output device may be configured to transmit data through thecommunication network106 based on such user input. Examples of output devices include tablets, smartphones, laptop computers, other mobile computing devices, desktop computers, smart TVs, and the like.
• Another example of theoutput system110 may take the form of a work-order system configured to output a request for a mechanic or the like to repair an asset. Yet another example of theoutput system110 may take the form of a parts-ordering system configured to place an order for a part of an asset and output a receipt thereof. Numerous other output systems are also possible.
• Thedata source112 may be configured to communicate with theanalytics system108. In general, thedata source112 may be or include one or more computing systems configured to collect, store, and/or provide to other systems, such as theanalytics system108, data that may be relevant to the functions performed by theanalytics system108. Thedata source112 may be configured to generate and/or obtain data independently from theassets102 and104. As such, the data provided by thedata source112 may be referred to herein as “external data.” Thedata source112 may be configured to provide current and/or historical data. In practice, theanalytics system108 may receive data from thedata source112 by “subscribing” to a service provided by the data source. However, theanalytics system108 may receive data from thedata source112 in other manners as well.
• Examples of thedata source112 include environment data sources, asset-management data sources, and other data sources. In general, environment data sources provide data indicating some characteristic of the environment in which assets are operated. Examples of environment data sources include weather-data servers, global navigation satellite systems (GNSS) servers, map-data servers, and topography-data servers that provide information regarding natural and artificial features of a given area, among other examples.
• In general, asset-management data sources provide data indicating events or statuses of entities (e.g., other assets) that may affect the operation or maintenance of assets (e.g., when and where an asset may operate or receive maintenance). Examples of asset-management data sources include traffic-data servers that provide information regarding air, water, and/or ground traffic, asset-schedule servers that provide information regarding expected routes and/or locations of assets on particular dates and/or at particular times, defect detector systems (also known as “hotbox” detectors) that provide information regarding one or more operating conditions of an asset that passes in proximity to the defect detector system, part-supplier servers that provide information regarding parts that particular suppliers have in stock and prices thereof, and repair-shop servers that provide information regarding repair shop capacity and the like, among other examples.
• Examples of other data sources include power-grid servers that provide information regarding electricity consumption and external databases that store historical operating data for assets, among other examples. One of ordinary skill in the art will appreciate that these are but a few examples of data sources and that numerous others are possible.
• It should be understood that thenetwork configuration100 is one example of a network in which embodiments described herein may be implemented. Numerous other arrangements are possible and contemplated herein. For instance, other network configurations may include additional components not pictured and/or more or less of the pictured components.
• II. Example Asset
• Turning toFIG. 2, a simplified block diagram of anexample asset200 is depicted. Either or both ofassets102 and104 fromFIG. 1 may be configured like theasset200. As shown, theasset200 may include one ormore subsystems202, one ormore sensors204, one ormore actuators205, acentral processing unit206,data storage208, anetwork interface210, auser interface212, and alocal analytics device220, all of which may be communicatively linked by a system bus, network, or other connection mechanism. One of ordinary skill in the art will appreciate that theasset200 may include additional components not shown and/or more or less of the depicted components.
• Broadly speaking, theasset200 may include one or more electrical, mechanical, and/or electromechanical components configured to perform one or more operations. In some cases, one or more components may be grouped into a givensubsystem202.
• Generally, asubsystem202 may include a group of related components that are part of theasset200. Asingle subsystem202 may independently perform one or more operations or thesingle subsystem202 may operate along with one or more other subsystems to perform one or more operations. Typically, different types of assets, and even different classes of the same type of assets, may include different subsystems.
• For instance, in the context of transportation assets, examples ofsubsystems202 may include engines, transmissions, drivetrains, fuel systems, battery systems, exhaust systems, braking systems, electrical systems, signal processing systems, generators, gear boxes, rotors, and hydraulic systems, among numerous other subsystems. In the context of a medical machine, examples ofsubsystems202 may include scanning systems, motors, coil and/or magnet systems, signal processing systems, rotors, and electrical systems, among numerous other subsystems.
• As suggested above, theasset200 may be outfitted withvarious sensors204 that are configured to monitor operating conditions of theasset200 andvarious actuators205 that are configured to interact with theasset200 or a component thereof and monitor operating conditions of theasset200. In some cases, some of thesensors204 and/oractuators205 may be grouped based on aparticular subsystem202. In this way, the group ofsensors204 and/oractuators205 may be configured to monitor operating conditions of theparticular subsystem202, and the actuators from that group may be configured to interact with theparticular subsystem202 in some way that may alter the subsystem's behavior based on those operating conditions.
• In general, asensor204 may be configured to detect a physical property, which may be indicative of one or more operating conditions of theasset200, and provide an indication, such as an electrical signal, of the detected physical property. In operation, thesensors204 may be configured to obtain measurements continuously, periodically (e.g., based on a sampling frequency), and/or in response to some triggering event. In some examples, thesensors204 may be preconfigured with operating parameters for performing measurements and/or may perform measurements in accordance with operating parameters provided by the central processing unit206 (e.g., sampling signals that instruct thesensors204 to obtain measurements). In examples,different sensors204 may have different operating parameters (e.g., some sensors may sample based on a first frequency, while other sensors sample based on a second, different frequency). In any event, thesensors204 may be configured to transmit electrical signals indicative of a measured physical property to thecentral processing unit206. Thesensors204 may continuously or periodically provide such signals to thecentral processing unit206.
• For instance,sensors204 may be configured to measure physical properties such as the location and/or movement of theasset200, in which case the sensors may take the form of GNSS sensors, dead-reckoning-based sensors, accelerometers, gyroscopes, pedometers, magnetometers, or the like.
• Additionally,various sensors204 may be configured to measure other operating conditions of theasset200, examples of which may include temperatures, pressures, speeds, acceleration or deceleration rates, friction, power usages, fuel usages, fluid levels, runtimes, voltages and currents, magnetic fields, electric fields, presence or absence of objects, positions of components, and power generation, among other examples. One of ordinary skill in the art will appreciate that these are but a few example operating conditions that sensors may be configured to measure. Additional or fewer sensors may be used depending on the industrial application or specific asset.
• As suggested above, anactuator205 may be configured similar in some respects to asensor204. Specifically, anactuator205 may be configured to detect a physical property indicative of an operating condition of theasset200 and provide an indication thereof in a manner similar to thesensor204.
• Moreover, anactuator205 may be configured to interact with theasset200, one ormore subsystems202, and/or some component thereof. As such, anactuator205 may include a motor or the like that is configured to move or otherwise control a component or system. In a particular example, an actuator may be configured to measure a fuel flow and alter the fuel flow (e.g., restrict the fuel flow), or an actuator may be configured to measure a hydraulic pressure and alter the hydraulic pressure (e.g., increase or decrease the hydraulic pressure). Numerous other example interactions of an actuator are also possible and contemplated herein.
• Generally, thecentral processing unit206 may include one or more processors and/or controllers, which may take the form of a general- or special-purpose processor or controller. In particular, in example implementations, thecentral processing unit206 may be or include microprocessors, microcontrollers, application specific integrated circuits, digital signal processors, and the like. In turn, thedata storage208 may be or include one or more non-transitory computer-readable storage media, such as optical, magnetic, organic, or flash memory, among other examples.
• Thecentral processing unit206 may be configured to store, access, and execute computer-readable program instructions stored in thedata storage208 to perform the operations of an asset described herein. For instance, as suggested above, thecentral processing unit206 may be configured to receive respective sensor signals from thesensors204 and/oractuators205. Thecentral processing unit206 may be configured to store sensor and/or actuator data in and later access it from thedata storage208.
• Thecentral processing unit206 may also be configured to determine whether received sensor and/or actuator signals trigger any abnormal-condition indicators, such as fault codes. For instance, thecentral processing unit206 may be configured to store in thedata storage208 abnormal-condition rules, each of which include a given abnormal-condition indicator representing a particular abnormal condition and respective triggering criteria that trigger the abnormal-condition indicator. That is, each abnormal-condition indicator corresponds with one or more sensor and/or actuator measurement values that must be satisfied before the abnormal-condition indicator is triggered. In practice, theasset200 may be pre-programmed with the abnormal-condition rules and/or may receive new abnormal-condition rules or updates to existing rules from a computing system, such as theanalytics system108.
• In any event, thecentral processing unit206 may be configured to determine whether received sensor and/or actuator signals trigger any abnormal-condition indicators. That is, thecentral processing unit206 may determine whether received sensor and/or actuator signals satisfy any triggering criteria. When such a determination is affirmative, thecentral processing unit206 may generate abnormal-condition data and may also cause the asset'suser interface212 to output an indication of the abnormal condition, such as a visual and/or audible alert. Additionally, thecentral processing unit206 may log the occurrence of the abnormal-condition indicator being triggered in thedata storage208, perhaps with a timestamp.
• FIG. 3 depicts a conceptual illustration of example abnormal-condition indicators and respective triggering criteria for an asset. In particular,FIG. 3 depicts a conceptual illustration of example fault codes. As shown, table300 includescolumns302,304, and306 that correspond to Sensor A, Actuator B, and Sensor C, respectively, androws308,310, and312 that correspond to FaultCodes1,2, and3, respectively.Entries314 then specify sensor criteria (e.g., sensor value thresholds) that correspond to the given fault codes.
• For example,Fault Code1 will be triggered when Sensor A detects a rotational measurement greater than135 revolutions per minute (RPM) and Sensor C detects a temperature measurement greater than 65° Celsius (C),Fault Code2 will be triggered when Actuator B detects a voltage measurement greater than 1000 Volts (V) and Sensor C detects a temperature measurement less than 55° C., andFault Code3 will be triggered when Sensor A detects a rotational measurement greater than 100 RPM, Actuator B detects a voltage measurement greater than 750 V, and Sensor C detects a temperature measurement greater than 60° C. One of ordinary skill in the art will appreciate thatFIG. 3 is provided for purposes of example and explanation only and that numerous other fault codes and/or triggering criteria are possible and contemplated herein.
• Referring back toFIG. 2, thecentral processing unit206 may be configured to carry out various additional functions for managing and/or controlling operations of theasset200 as well. For example, thecentral processing unit206 may be configured to provide instruction signals to thesubsystems202 and/or theactuators205 that cause thesubsystems202 and/or theactuators205 to perform some operation, such as modifying a throttle position. Additionally, thecentral processing unit206 may be configured to modify the rate at which it processes data from thesensors204 and/or theactuators205, or thecentral processing unit206 may be configured to provide instruction signals to thesensors204 and/oractuators205 that cause thesensors204 and/oractuators205 to, for example, modify a sampling rate. Moreover, thecentral processing unit206 may be configured to receive signals from thesubsystems202, thesensors204, theactuators205, the network interfaces210, and/or theuser interfaces212 and based on such signals, cause an operation to occur. Further still, thecentral processing unit206 may be configured to receive signals from a computing device, such as a diagnostic device, that cause thecentral processing unit206 to execute one or more diagnostic tools in accordance with diagnostic rules stored in thedata storage208. Other functionalities of thecentral processing unit206 are discussed below.
• Thenetwork interface210 may be configured to provide for communication between theasset200 and various network components connected tocommunication network106. For example, thenetwork interface210 may be configured to facilitate wireless communications to and from thecommunication network106 and may thus take the form of an antenna structure and associated equipment for transmitting and receiving various over-the-air signals. Other examples are possible as well. In practice, thenetwork interface210 may be configured according to a communication protocol, such as but not limited to any of those described above.
• Theuser interface212 may be configured to facilitate user interaction with theasset200 and may also be configured to facilitate causing theasset200 to perform an operation in response to user interaction. Examples ofuser interfaces212 include touch-sensitive interfaces, mechanical interfaces (e.g., levers, buttons, wheels, dials, keyboards, etc.), and other input interfaces (e.g., microphones), among other examples. In some cases, theuser interface212 may include or provide connectivity to output components, such as display screens, speakers, headphone jacks, and the like.
• Thelocal analytics device220 may generally be configured to receive and analyze data and based on such analysis, cause one or more operations to occur at theasset200. In particular, thelocal analytics device220 may receive data from thesensors204 and/oractuators205 and based on such data, may provide instructions to thecentral processing unit206 that cause theasset200 to perform an operation.
• In practice, thelocal analytics device220 may enable theasset200 to locally perform advanced analytics and associated operations, such as executing a predictive model and corresponding workflow, that may otherwise not be able to be performed with the other on-asset components. As such, thelocal analytics device220 may help provide additional processing power and/or intelligence to theasset200.
• As shown, thelocal analytics device220 may include aprocessing unit222, adata storage224, and anetwork interface226, all of which may be communicatively linked by a system bus, network, or other connection mechanism. Theprocessing unit222 may include any of the components discussed above with respect to thecentral processing unit206. In turn, thedata storage224 may be or include one or more non-transitory computer-readable storage media, which may take any of the forms of computer-readable storage media discussed above.
• Theprocessing unit222 may be configured to store, access, and execute computer-readable program instructions stored in thedata storage224 to perform the operations of a local analytics device described herein. For instance, theprocessing unit222 may be configured to receive respective sensor and/or actuator signals from thesensors204 and/oractuators205 and execute a predictive model-workflow pair based on such signals. Other functions are described below.
• Thenetwork interface226 may be the same or similar to the network interfaces described above. In practice, thenetwork interface226 may facilitate communication between theasset200 and theanalytics system108.
• In some example implementations, thelocal analytics device220 may include and/or communicate with a user interface that may be similar to theuser interface212. In practice, the user interface may be located remotely from the local analytics device220 (and the asset200). Other examples are also possible.
• One of ordinary skill in the art will appreciate that theasset200 shown inFIG. 2 is but one example of a simplified representation of an asset and that numerous others are also possible. For instance, other assets may include additional components not pictured and/or more or less of the pictured components. Moreover, a given asset may include multiple, individual assets that are operated in concert to perform operations of the given asset. Other examples are also possible.
• III. Example Analytics System
• Referring now toFIG. 4, a simplified block diagram of anexample analytics system400 is depicted. As suggested above, theanalytics system400 may include one or more computing systems communicatively linked and arranged to carry out various operations described herein. Specifically, as shown, theanalytics system400 may include adata intake system402, adata science system404, and one ormore databases406. These system components may be communicatively coupled via one or more wireless and/or wired connections, which may be configured to facilitate secure communications.
• Thedata intake system402 may generally function to receive and process data and output data to thedata science system404. As such, thedata intake system402 may include one or more network interfaces configured to receive data from various network components of thenetwork configuration100, such as theassets102 and104, theoutput system110, and/or thedata source112. Specifically, thedata intake system402 may be configured to receive analog signals, data streams, and/or network packets, among other examples. As such, the network interfaces may include one or more wired network interfaces, such as a port or the like, and/or wireless network interfaces, similar to those described above. In some examples, thedata intake system402 may be or include components configured according to a given dataflow technology, such as a NiFi receiver or the like.
• Thedata intake system402 may include one or more processing components configured to perform one or more operations. Example operations may include compression and/or decompression, encryption and/or de-encryption, analog-to-digital and/or digital-to-analog conversion, filtration, and amplification, among other operations. Moreover, thedata intake system402 may be configured to parse, sort, organize, and/or route data based on data type and/or characteristics of the data. In some examples, thedata intake system402 may be configured to format, package, and/or route data based on one or more characteristics or operating parameters of thedata science system404.
• In general, the data received by thedata intake system402 may take various forms. For example, the payload of the data may include a single sensor or actuator measurement, multiple sensor and/or actuator measurements and/or one or more abnormal-condition data. Other examples are also possible.
• Moreover, the received data may include certain characteristics, such as a source identifier and a timestamp (e.g., a date and/or time at which the information was obtained). For instance, a unique identifier (e.g., a computer generated alphabetic, numeric, alphanumeric, or the like identifier) may be assigned to each asset, and perhaps to each sensor and actuator. Such identifiers may be operable to identify the asset, sensor, or actuator from which data originates. In some cases, another characteristic may include the location (e.g., GPS coordinates) at which the information was obtained. Data characteristics may come in the form of signal signatures or metadata, among other examples.
• Thedata science system404 may generally function to receive (e.g., from the data intake system402) and analyze data and based on such analysis, cause one or more operations to occur. As such, thedata science system404 may include one ormore network interfaces408, aprocessing unit410, anddata storage412, all of which may be communicatively linked by a system bus, network, or other connection mechanism. In some cases, thedata science system404 may be configured to store and/or access one or more application program interfaces (APIs) that facilitate carrying out some of the functionality disclosed herein.
• The network interfaces408 may be the same or similar to any network interface described above. In practice, the network interfaces408 may facilitate communication (e.g., with some level of security) between thedata science system404 and various other entities, such as thedata intake system402, thedatabases406, theassets102, theoutput system110, etc.
• Theprocessing unit410 may include one or more processors, which may take any of the processor forms described above. In turn, thedata storage412 may be or include one or more non-transitory computer-readable storage media, which may take any of the forms of computer-readable storage media discussed above. Theprocessing unit410 may be configured to store, access, and execute computer-readable program instructions stored in thedata storage412 to perform the operations of an analytics system described herein.
• In general, theprocessing unit410 may be configured to perform analytics on data received from thedata intake system402. To that end, theprocessing unit410 may be configured to execute one or more modules, which may each take the form of one or more sets of program instructions that are stored in thedata storage412. The modules may be configured to facilitate causing an outcome to occur based on the execution of the respective program instructions. An example outcome from a given module may include outputting data into another module, updating the program instructions of the given module and/or of another module, and outputting data to anetwork interface408 for transmission to an asset and/or theoutput system110, among other examples.
• Thedatabases406 may generally function to receive (e.g., from the data science system404) and store data. As such, eachdatabase406 may include one or more non-transitory computer-readable storage media, such as any of the examples provided above. In practice, thedatabases406 may be separate from or integrated with thedata storage412.
• Thedatabases406 may be configured to store numerous types of data, some of which is discussed below. In practice, some of the data stored in thedatabases406 may include a timestamp indicating a date and time at which the data was generated or added to the database. Moreover, data may be stored in a number of manners in thedatabases406. For instance, data may be stored in time sequence, in a tabular manner, and/or organized based on data source type (e.g., based on asset, asset type, sensor, sensor type, actuator, or actuator type) or abnormal-condition indicator, among other examples.
• IV. Example Operations
• The operations of theexample network configuration100 depicted inFIG. 1 will now be discussed in further detail below. To help describe some of these operations, flow diagrams may be referenced to describe combinations of operations that may be performed. In some cases, each block may represent a module or portion of program code that includes instructions that are executable by a processor to implement specific logical functions or steps in a process. The program code may be stored on any type of computer-readable medium, such as non-transitory computer-readable media. In other cases, each block may represent circuitry that is wired to perform specific logical functions or steps in a process. Moreover, the blocks shown in the flow diagrams may be rearranged into different orders, combined into fewer blocks, separated into additional blocks, and/or removed based upon the particular embodiment.
• The following description may reference examples where a single data source, such as theasset102, provides data to theanalytics system108 that then performs one or more functions. It should be understood that this is done merely for sake of clarity and explanation and is not meant to be limiting. In practice, theanalytics system108 generally receives data from multiple sources, perhaps simultaneously, and performs operations based on such aggregate received data.
• A. Collection of Operating Data
• As mentioned above, therepresentative asset102 may take various forms and may be configured to perform a number of operations. In a non-limiting example, theasset102 may take the form of a locomotive that is operable to transfer cargo across the United States. While in transit, the sensors and/or actuators of theasset102 may obtain data that reflects one or more operating conditions of theasset102. The sensors and/or actuators may transmit the data to a processing unit of theasset102.
• The processing unit may be configured to receive the data from the sensors and/or actuators. In practice, the processing unit may receive sensor data from multiple sensors and/or actuator data from multiple actuators simultaneously or sequentially. As discussed above, while receiving this data, the processing unit may also be configured to determine whether the data satisfies triggering criteria that trigger any abnormal-condition indicators, such as fault codes. In the event the processing unit determines that one or more abnormal-condition indicators are triggered, the processing unit may be configured to perform one or more local operations, such as outputting an indication of the triggered indicator via a user interface.
• Theasset102 may then transmit operating data to theanalytics system108 via a network interface of theasset102 and thecommunication network106. In operation, theasset102 may transmit operating data to theanalytics system108 continuously, periodically, and/or in response to triggering events (e.g., abnormal conditions). Specifically, theasset102 may transmit operating data periodically based on a particular frequency (e.g., daily, hourly, every fifteen minutes, once per minute, once per second, etc.), or theasset102 may be configured to transmit a continuous, real-time feed of operating data. Additionally or alternatively, theasset102 may be configured to transmit operating data based on certain triggers, such as when sensor and/or actuator measurements satisfy triggering criteria for any abnormal-condition indicators. Theasset102 may transmit operating data in other manners as well.
• In practice, operating data for theasset102 may include sensor data, actuator data, and/or abnormal-condition data. In some implementations, theasset102 may be configured to provide the operating data in a single data stream, while in other implementations theasset102 may be configured to provide the operating data in multiple, distinct data streams. For example, theasset102 may provide to the analytics system108 a first data stream of sensor and/or actuator data and a second data stream of abnormal-condition data. Other possibilities also exist.
• Sensor and actuator data may take various forms. For example, at times, sensor data (or actuator data) may include measurements obtained by each of the sensors (or actuators) of theasset102. While at other times, sensor data (or actuator data) may include measurements obtained by a subset of the sensors (or actuators) of theasset102.
• Specifically, the sensor and/or actuator data may include measurements obtained by the sensors and/or actuators associated with a given triggered abnormal-condition indicator. For example, if a triggered fault code isFault Code1 fromFIG. 3, then sensor data may include raw measurements obtained by Sensors A and C. Additionally or alternatively, the data may include measurements obtained by one or more sensors or actuators not directly associated with the triggered fault code. Continuing off the last example, the data may additionally include measurements obtained by Actuator B and/or other sensors or actuators. In some examples, theasset102 may include particular sensor data in the operating data based on a fault-code rule or instruction provided by theanalytics system108, which may have, for example, determined that there is a correlation between that which Actuator B is measuring and that which caused theFault Code1 to be triggered in the first place. Other examples are also possible.
• Further still, the data may include one or more sensor and/or actuator measurements from each sensor and/or actuator of interest based on a particular time of interest, which may be selected based on a number of factors. In some examples, the particular time of interest may be based on a sampling rate. In other examples, the particular time of interest may be based on the time at which an abnormal-condition indicator is triggered.
• In particular, based on the time at which an abnormal-condition indicator is triggered, the data may include one or more respective sensor and/or actuator measurements from each sensor and/or actuator of interest (e.g., sensors and/or actuators directly and indirectly associated with the triggered indicator). The one or more measurements may be based on a particular number of measurements or particular duration of time around the time of the triggered abnormal-condition indicator.
• For example, if a triggered fault code isFault Code2 fromFIG. 3, the sensors and actuators of interest might include Actuator B and Sensor C. The one or more measurements may include the most recent respective measurements obtained by Actuator B and Sensor C prior to the triggering of the fault code (e.g., triggering measurements) or a respective set of measurements before, after, or about the triggering measurements. For example, a set of five measurements may include the five measurements before or after the triggering measurement (e.g., excluding the triggering measurement), the four measurements before or after the triggering measurement and the triggering measurement, or the two measurements before and the two after as well as the triggering measurement, among other possibilities.
• Similar to sensor and actuator data, the abnormal-condition data may take various forms. In general, the abnormal-condition data may include or take the form of an indicator that is operable to uniquely identify a particular abnormal condition that occurred at theasset102 from all other abnormal conditions that may occur at theasset102. The abnormal-condition indicator may take the form of an alphabetic, numeric, or alphanumeric identifier, among other examples. Moreover, the abnormal-condition indicator may take the form of a string of words that is descriptive of the abnormal condition, such as “Overheated Engine” or “Out of Fuel”, among other examples.
• Theanalytics system108, and in particular, the data intake system of theanalytics system108, may be configured to receive operating data from one or more assets and/or data sources. The data intake system may be configured to perform one or more operations to the received data and then relay the data to the data science system of theanalytics system108. In turn, the data science system may analyze the received data and based on such analysis, perform one or more operations.
• B. Defining Predictive Models & Workflows
• As one example, theanalytics system108 may be configured to define predictive models and corresponding workflows based on received operating data for one or more assets and/or received external data related to the one or more assets. Theanalytics system108 may define model-workflow pairs based on various other data as well.
• In general, a model-workflow pair may include a set of program instructions that cause an asset to monitor certain operating conditions and carry out certain operations that help facilitate preventing the occurrence of a particular event suggested by the monitored operating conditions. Specifically, a predictive model may include one or more algorithms whose inputs are sensor and/or actuator data from one or more sensors and/or actuators of an asset and whose outputs are utilized to determine a probability that a particular event may occur at the asset within a particular period of time in the future. In turn, a workflow may include one or more triggers (e.g., model output values) and corresponding operations that the asset carries out based on the triggers.
• As suggested above, theanalytics system108 may be configured to define aggregate and/or individualized predictive models and/or workflows. An “aggregate” model/workflow may refer to a model/workflow that is generic for a group of assets and defined without taking into consideration particular characteristics of the assets to which the model/workflow is deployed. On the other hand, an “individualized” model/workflow may refer to a model/workflow that is specifically tailored for a single asset or a subgroup of assets from the group of assets and defined based on particular characteristics of the single asset or subgroup of assets to which the model/workflow is deployed. These different types of models/workflows and the operations performed by theanalytics system108 to define them are discussed in further detail below.
• 1. Aggregate Models & Workflows
• In example implementations, theanalytics system108 may be configured to define an aggregate model-workflow pair based on aggregated data for a plurality of assets. Defining aggregate model-workflow pairs may be performed in a variety of manners.
• FIG. 5 is a flow diagram500 depicting one possible example of a definition phase that may be used for defining model-workflow pairs. For purposes of illustration, the example definition phase is described as being carried out by theanalytics system108, but this definition phase may be carried out by other systems as well. One of ordinary skill in the art will appreciate that the flow diagram500 is provided for sake of clarity and explanation and that numerous other combinations of operations may be utilized to define a model-workflow pair.
• As shown inFIG. 5, atblock502, theanalytics system108 may begin by defining a set of data that forms the basis for a given predictive model (e.g., the data of interest). The data of interest may derive from a number of sources, such as theassets102 and104 and thedata source112, and may be stored in a database of theanalytics system108.
• The data of interest may include historical data for a particular set of assets from a group of assets or all of the assets from a group of assets (e.g., the assets of interest). Moreover, the data of interest may include measurements from a particular set of sensors and/or actuators from each of the assets of interest or from all of the sensors and/or actuators from each of the assets of interest. Further still, the data of interest may include data from a particular period of time in the past, such as two week's worth of historical data.
• The data of interest may include a variety of types data, which may depend on the given predictive model. In some instances, the data of interest may include at least operating data indicating operating conditions of assets, where the operating data is as discussed above in the Collection of Operating Data section. Additionally, the data of interest may include environment data indicating environments in which assets are typically operated and/or scheduling data indicating planned dates and times during which assets are to carry out certain tasks. Other types of data may also be included in the data of interest.
• In practice, the data of interest may be defined in a number of manners. In one example, the data of interest may be user-defined. In particular, a user may operate anoutput system110 that receives user inputs indicating a selection of certain data of interest, and theoutput system110 may provide to theanalytics system108 data indicating such selections. Based on the received data, theanalytics system108 may then define the data of interest.
• In another example, the data of interest may be machine-defined. In particular, theanalytics system108 may perform various operations, such as simulations, to determine the data of interest that generates the most accurate predictive model. Other examples are also possible.
• Returning toFIG. 5, atblock504, theanalytics system108 may be configured to, based on the data of interest, define an aggregate, predictive model that is related to the operation of assets. In general, an aggregate, predictive model may define a relationship between operating conditions of assets and a likelihood of an event occurring at the assets. Specifically, an aggregate, predictive model may receive as inputs sensor data from sensors of an asset and/or actuator data from actuators of the asset and output a probability that an event will occur at the asset within a certain amount of time into the future.
• The event that the predictive model predicts may vary depending on the particular implementation. For example, the event may be a failure and so, the predictive model may be a failure model that predicts whether a failure will occur within a certain period of time in the future (failure models are discussed in detail below in the Health-Score Models & Workflows section). In another example, the event may be an asset completing a task and so, the predictive model may predict the likelihood that an asset will complete a task on time. In other examples, the event may be a fluid or component replacement, and so, the predictive model may predict an amount of time before a particular asset fluid or component needs to be replaced. In yet other examples, the event may be a change in asset productivity, and so, the predictive model may predict the productivity of an asset during a particular period of time in the future. In one other example, the event may be the occurrence of a “leading indicator” event, which may indicate an asset behavior that differs from expected asset behaviors, and so, the predictive model may predict the likelihood of one or more leading indicator events occurring in the future. Other examples of predictive models are also possible.
• In any event, theanalytics system108 may define the aggregate, predictive model in a variety of manners. In general, this operation may involve utilizing one or more modeling techniques to generate a model that returns a probability between zero and one, such as a random forest technique, logistic regression technique, or other regression technique, among other modeling techniques. In a particular example implementation, theanalytics system108 may define the aggregate, predictive model in line with the below discussion referencingFIG. 7. Theanalytics system108 may define the aggregate model in other manners as well.
• Atblock506, theanalytics system108 may be configured to define an aggregate workflow that corresponds to the defined model fromblock504. In general, a workflow may take the form of an action that is carried out based on a particular output of a predictive model. In example implementations, a workflow may include one or more operations that an asset performs based on the output of the defined predictive model. Examples of operations that may be part of a workflow include an asset acquiring data according to a particular data-acquisition scheme, transmitting data to theanalytics system108 according to a particular data-transmission scheme, executing a local diagnostic tool, and/or modifying an operating condition of the asset, among other example workflow operations.
• A particular data-acquisition scheme may indicate how an asset acquires data. In particular, a data-acquisition scheme may indicate certain sensors and/or actuators from which the asset obtains data, such as a subset of sensors and/or actuators of the asset's plurality of sensors and actuators (e.g., sensors/actuators of interest). Further, a data-acquisition scheme may indicate an amount of data that the asset obtains from the sensors/actuators of interest and/or a sampling frequency at which the asset acquires such data. Data-acquisition schemes may include various other attributes as well. In a particular example implementation, a particular data-acquisition scheme may correspond to a predictive model for asset health and may be adjusted to acquire more data and/or particular data (e.g., from particular sensors) based on a decreasing asset health. Or a particular data-acquisition scheme may correspond to a leading-indicators predictive model and may be adjusted to a modify data acquired by asset sensors and/or actuators based on an increased likelihood of an occurrence of a leading indicator event that may signal that a subsystem failure might occur.
• A particular data-transmission scheme may indicate how an asset transmits data to theanalytics system108. Specifically, a data-transmission scheme may indicate a type of data (and may also indicate a format and/or structure of the data) that the asset should transmit, such as data from certain sensors or actuators, a number of data samples that the asset should transmit, a transmission frequency, and/or a priority-scheme for the data that the asset should include in its data transmission. In some cases, a particular data-acquisition scheme may include a data-transmission scheme or a data-acquisition scheme may be paired with a data-transmission scheme. In some example implementations, a particular data-transmission scheme may correspond to a predictive model for asset health and may be adjusted to transmit data less frequently based on an asset health that is above a threshold value. Other examples are also possible.
• As suggested above, a local diagnostic tool may be a set of procedures or the like that are stored locally at an asset. The local diagnostic tool may generally facilitate diagnosing a cause of a fault or failure at an asset. In some cases, when executed, a local diagnostic tool may pass test inputs into a subsystem of an asset or a portion thereof to obtain test results, which may facilitate diagnosing the cause of a fault or failure. These local diagnostic tools are typically dormant on an asset and will not be executed unless the asset receives particular diagnostic instructions. Other local diagnostic tools are also possible. In one example implementation, a particular local diagnostic tool may correspond to a predictive model for health of a subsystem of an asset and may be executed based on a subsystem health that is at or below a threshold value.
• Lastly, a workflow may involve modifying an operating condition of an asset. For instance, one or more actuators of an asset may be controlled to facilitate modifying an operating condition of the asset. Various operating conditions may be modified, such as a speed, temperature, pressure, fluid level, current draw, and power distribution, among other examples. In a particular example implementation, an operating-condition modification workflow may correspond to a predictive model for predicting whether an asset will complete a task on time and may cause the asset to increase its speed of travel based on a predicted completion percentage that is below a threshold value.
• In any event, the aggregate workflow may be defined in a variety of manners. In one example, the aggregate workflow may be user defined. Specifically, a user may operate a computing device that receives user inputs indicating selection of certain workflow operations, and the computing device may provide to theanalytics system108 data indicating such selections. Based on this data, theanalytics system108 may then define the aggregate workflow.
• In another example, the aggregate workflow may be machine-defined. In particular, theanalytics system108 may perform various operations, such as simulations, to determine a workflow that may facilitate determining a cause of the probability output by the predictive model and/or preventing an occurrence of an event predicted by the model. Other examples of defining the aggregate workflow are also possible.
• In defining the workflow corresponding to the predictive model, theanalytics system108 may define the triggers of the workflow. In example implementations, a workflow trigger may be a value of the probability output by the predictive model or a range of values output by the predictive model. In some cases, a workflow may have multiple triggers, each of which may cause a different operation or operations to occur.
• To illustrate,FIG. 6A is a conceptual illustration of an aggregate model-workflow pair600. As shown, the aggregate model-workflow pair illustration600 includes a column formodel inputs602,model calculations604, model output ranges606, andcorresponding workflow operations608. In this example, the predictive model has a single input, data from Sensor A, and has two calculations, Calculations I and II. The output of this predictive model affects the workflow operation that is performed. If the output probability is less than or equal to 80%, thenworkflow Operation1 is performed. Otherwise, theworkflow Operation2 is performed. Other example model-workflow pairs are possible and contemplated herein.
• 2. Individualized Models & Workflows
• In another aspect, theanalytics system108 may be configured to define individualized predictive models and/or workflows for assets, which may involve utilizing the aggregate model-workflow pair as a baseline. The individualization may be based on certain characteristics of assets. In this way, theanalytics system108 may provide a given asset a more accurate and robust model-workflow pair compared to the aggregate model-workflow pair.
• In particular, returning toFIG. 5, atblock508, theanalytics system108 may be configured to decide whether to individualize the aggregate model defined atblock504 for a given asset, such as theasset102. Theanalytics system108 may carry out this decision in a number of manners.
• In some cases, theanalytics system108 may be configured to define individualized predictive models by default. In other cases, theanalytics system108 may be configured to decide whether to define an individualized predictive model based on certain characteristics of theasset102. For example, in some cases, only assets of certain types or classes, or operated in certain environments, or that have certain health scores may receive an individualized predictive model. In yet other cases, a user may define whether an individualized model is defined for theasset102. Other examples are also possible.
• In any event, if theanalytics system108 decides to define an individualized predictive model for theasset102, theanalytics system108 may do so atblock510. Otherwise, theanalytics system108 may proceed to block512.
• Atblock510, theanalytics system108 may be configured to define an individualized predictive model in a number of manners. In example implementations, theanalytics system108 may define an individualized predictive model based at least in part on one or more characteristics of theasset102.
• Before defining the individualized predictive model for theasset102, theanalytics system108 may have determined one or more asset characteristics of interest that form the basis of individualized models. In practice, different predictive models may have different corresponding characteristics of interest.
• In general, the characteristics of interest may be characteristics that are related to the aggregate model-workflow pair. For instance, the characteristics of interest may be characteristics that theanalytics system108 has determined influence the accuracy of the aggregate model-workflow pair. Examples of such characteristics may include asset age, asset usage, asset capacity, asset load, asset health (perhaps indicated by an asset health metric, discussed below), asset class (e.g., brand and/or model), and environment in which an asset is operated, among other characteristics.
• Theanalytics system108 may have determined the characteristics of interest in a number of manners. In one example, theanalytics system108 may have done so by performing one or more modeling simulations that facilitate identifying the characteristics of interest. In another example, the characteristics of interest may have been predefined and stored in the data storage of theanalytics system108. In yet another example, characteristics of interest may have been define by a user and provided to theanalytics system108 via theoutput system110. Other examples are also possible.
• In any event, after determining the characteristics of interest, theanalytics system108 may determine characteristics of theasset102 that correspond to the determined characteristics of interest. That is, theanalytics system108 may determine a type, value, existence or lack thereof, etc. of theasset102′s characteristics that correspond to the characteristics of interest. Theanalytics system108 may perform this operation in a number of manners.
• For examples, theanalytics system108 may be configured to perform this operation based on data originating from theasset102 and/or thedata source112. In particular, theanalytics system108 may utilize operating data for theasset102 and/or external data from thedata source112 to determine one or more characteristics of theasset102. Other examples are also possible.
• Based on the determined one or more characteristics of theasset102, theanalytics system108 may define an individualized, predictive model by modifying the aggregate model. The aggregate model may be modified in a number of manners. For example, the aggregate model may be modified by changing (e.g., adding, removing, re-ordering, etc.) one or more model inputs, changing one or more sensor and/or actuator measurement ranges that correspond to asset-operating limits (e.g., changing operating limits that correspond to “leading indicator” events), changing one or more model calculations, weighting (or changing a weight of) a variable or output of a calculation, utilizing a modeling technique that differs from that which was utilized to define the aggregate model, and/or utilizing a response variable that differs from that which was utilized to define the aggregate model, among other examples.
• To illustrate,FIG. 6B is a conceptual illustration of an individualized model-workflow pair610. Specifically, the individualized model-workflow pair illustration610 is a modified version of the aggregate model-workflow pair fromFIG. 6A. As shown, the individualized model-workflow pair illustration610 includes a modified column formodel inputs612 andmodel calculations614 and includes the original columns for model output ranges606 andworkflow operations608 fromFIG. 6A. In this example, the individualized model has two inputs, data from Sensor A and Actuator B, and has two calculations, Calculations II and III. The output ranges and corresponding workflow operations are the same as those ofFIG. 6A. Theanalytics system108 may have defined the individualized model in this way based on determining that theasset102 is, for example, relatively old and has relatively poor health, among other reasons.
• In practice, individualizing the aggregate model may depend on the one or more characteristics of the given asset. In particular, certain characteristics may affect the modification of the aggregate model differently than other characteristics. Further, the type, value, existence, or the like of a characteristic may affect the modification as well. For example, the asset age may affect a first part of the aggregate model, while an asset class may affect a second, different part of the aggregate model. And an asset age within a first range of ages may affect the first part of the aggregate model in a first manner, while an asset age within a second range of ages, different from the first range, may affect the first part of the aggregate model in a second, different manner. Other examples are also possible.
• In some implementations, individualizing the aggregate model may depend on considerations in addition to or alternatively to asset characteristics. For instance, the aggregate model may be individualized based on sensor and/or actuator readings of an asset when the asset is known to be in a relatively good operating state (e.g., as defined by a mechanic or the like). More particularly, in an example of a leading-indicator predictive model, theanalytics system108 may be configured to receive an indication that the asset is in a good operating state (e.g., from a computing device operated by a mechanic) along with operating data from the asset. Based at least on the operating data, theanalytics system108 may then individualize the leading-indicator predictive model for the asset by modifying respective operating limits corresponding to “leading indicator” events. Other examples are also possible.
• Returning toFIG. 5, atblock512, theanalytics system108 may also be configured to decide whether to individualize a workflow for theasset102. Theanalytics system108 may carry out this decision in a number of manners. In some implementations, theanalytics system108 may perform this operation in line withblock508. In other implementations, theanalytics system108 may decide whether to define an individualized workflow based on the individualized predictive model. In yet another implementation, theanalytics system108 may decide to define an individualized workflow if an individualized predictive model was defined. Other examples are also possible.
• In any event, if theanalytics system108 decides to define an individualized workflow for theasset102, theanalytics system108 may do so atblock514. Otherwise, theanalytics system108 may end the definition phase.
• Atblock514, theanalytics system108 may be configured to define an individualized workflow in a number of manners. In example implementations, theanalytics system108 may define an individualized workflow based at least in part on one or more characteristics of theasset102.
• Before defining the individualized workflow for theasset102, similar to defining the individualized predictive model, theanalytics system108 may have determined one or more asset characteristics of interest that form the basis of an individualized workflow, which may have been determined in line with the discussion ofblock510. In general, these characteristics of interest may be characteristics that affect the efficacy of the aggregate workflow. Such characteristics may include any of the example characteristics discussed above. Other characteristics are possible as well.
• Similar again to block510, theanalytics system108 may determine characteristics of theasset102 that correspond to the determined characteristics of interest for an individualized workflow. In example implementations, theanalytics system108 may determine characteristics of theasset102 in a manner similar to the characteristic determination discussed with reference to block510 and in fact, may utilize some or all of that determination.
• In any event, based on the determined one or more characteristics of theasset102, theanalytics system108 may individualize a workflow for theasset102 by modifying the aggregate workflow. The aggregate workflow may be modified in a number of manners. For example, the aggregate workflow may be modified by changing (e.g., adding, removing, re-ordering, replacing, etc.) one or more workflow operations (e.g., changing from a first data-acquisition scheme to a second scheme or changing from a particular data-acquisition scheme to a particular local diagnostic tool) and/or changing (e.g., increasing, decreasing, adding to, removing from, etc.) the corresponding model output value or range of values that triggers particular workflow operations, among other examples. In practice, modification to the aggregate workflow may depend on the one or more characteristics of theasset102 in a manner similar to the modification to the aggregate model.
• To illustrate,FIG. 6C is a conceptual illustration of an individualized model-workflow pair620. Specifically, the individualized model-workflow pair illustration620 is a modified version of the aggregate model-workflow pair fromFIG. 6A. As shown, the individualized model-workflow pair illustration620 includes the original columns formodel inputs602,model calculations604, and model output ranges606 fromFIG. 6A, but includes a modified column forworkflow operations628. In this example, the individualized model-workflow pair is similar to the aggregate model-workflow pair fromFIG. 6A, except that when the output of the aggregate model is greater than 80% workflow Operation3 is triggered instead ofOperation1. Theanalytics system108 may have defined this individual workflow based on determining that theasset102, for example, operates in an environment that historically increases the occurrence of asset failures, among other reasons.
• After defining the individualized workflow, theanalytics system108 may end the definition phase. At that point, theanalytics system108 may then have an individualized model-workflow pair for theasset102.
• In some example implementations, theanalytics system108 may be configured to define an individualized predictive model and/or corresponding workflow for a given asset without first defining an aggregate predictive model and/or corresponding workflow. Other examples are also possible.
• 3. Health-Score Models & Workflows
• In a particular implementation, as mentioned above, theanalytics system108 may be configured to define predictive models and corresponding workflows associated with the health of assets. In example implementations, one or more predictive models for monitoring the health of an asset may be utilized to output a health metric (e.g., “health score”) for an asset, which is a single, aggregated metric that indicates whether a failure will occur at a given asset within a given timeframe into the future (e.g., the next two weeks). In particular, a health metric may indicate a likelihood that no failures from a group of failures will occur at an asset within a given timeframe into the future, or a health metric may indicate a likelihood that at least one failure from a group of failures will occur at an asset within a given timeframe into the future.
• In practice, the predictive models utilized to output a health metric and the corresponding workflows may be defined as aggregate or individualized models and/or workflows, in line with the above discussion.
• Moreover, depending on the desired granularity of the health metric, theanalytics system108 may be configured to define different predictive models that output different levels of health metrics and to define different corresponding workflows. For example, theanalytics system108 may define a predictive model that outputs a health metric for the asset as a whole (i.e., an asset-level health metric). As another example, theanalytics system108 may define a respective predictive model that outputs a respective health metric for one or more subsystems of the asset (i.e., subsystem-level health metrics). In some cases, the outputs of each subsystem-level predictive model may be combined to generate an asset-level health metric. Other examples are also possible.
• In general, defining a predictive model that outputs a health metric may be performed in a variety of manners.FIG. 7 is a flow diagram700 depicting one possible example of a modeling phase that may be used for defining a model that outputs a health metric. For purposes of illustration, the example modeling phase is described as being carried out by theanalytics system108, but this modeling phase may be carried out by other systems as well. One of ordinary skill in the art will appreciate that the flow diagram700 is provided for sake of clarity and explanation and that numerous other combinations of operations may be utilized to determine a health metric.
• As shown inFIG. 7, atblock702, theanalytics system108 may begin by defining a set of the one or more failures that form the basis for the health metric (i.e., the failures of interest). In practice, the one or more failures may be those failures that could render an asset (or a subsystem thereof) inoperable if they were to occur. Based on the defined set of failures, theanalytics system108 may take steps to define a model for predicting a likelihood of any of the failures occurring within a given timeframe in the future (e.g., the next two weeks).
• In particular, atblock704, theanalytics system108 may analyze historical operating data for a group of one or more assets to identify past occurrences of a given failure from the set of failures. Atblock706, theanalytics system108 may identify a respective set of operating data that is associated with each identified past occurrence of the given failure (e.g., sensor and/or actuator data from a given timeframe prior to the occurrence of the given failure). Atblock708, theanalytics system108 may analyze the identified sets of operating data associated with past occurrences of the given failure to define a relationship (e.g., a failure model) between (1) the values for a given set of operating metrics and (2) the likelihood of the given failure occurring within a given timeframe in the future (e.g., the next two weeks). Lastly, atblock710, the defined relationship for each failure in the defined set (e.g., the individual failure models) may then be combined into a model for predicting the overall likelihood of a failure occurring.
• As theanalytics system108 continues to receive updated operating data for the group of one or more assets, theanalytics system108 may also continue to refine the predictive model for the defined set of one or more failures by repeating steps704-710 on the updated operating data.
• The functions of the example modeling phase illustrated inFIG. 7 will now be described in further detail. Starting withblock702, as noted above, theanalytics system108 may begin by defining a set of the one or more failures that form the basis for the health metric. Theanalytics system108 may perform this function in various manners.
• In one example, the set of the one or more failures may be based on one or more user inputs. Specifically, theanalytics system108 may receive from a computing system operated by a user, such as theoutput system110, input data indicating a user selection of the one or more failures. As such, the set of one or more failures may be user-defined.
• In other examples, the set of the one or more failures may be based on a determination made by the analytics system108 (e.g., machine-defined). In particular, theanalytics system108 may be configured to define the set of one or more failures, which may occur in a number of manners.
• For instance, theanalytics system108 may be configured to define the set of failures based on one or more characteristics of theasset102. That is, certain failures may correspond to certain characteristics, such as asset type, class, etc., of an asset. For example, each type and/or class of asset may have respective failures of interest.
• In another instance, theanalytics system108 may be configured to define the set of failures based on historical data stored in the databases of theanalytics system108 and/or external data provided by thedata source112. For example, theanalytics system108 may utilize such data to determine which failures result in the longest repair-time and/or which failures are historically followed by additional failures, among other examples.
• In yet other examples, the set of one or more failures may be defined based on a combination of user inputs and determinations made by theanalytics system108. Other examples are also possible.
• Atblock704, for each of the failures from the set of failures, theanalytics system108 may analyze historical operating data for a group of one or more assets (e.g., abnormal-behavior data) to identify past occurrences of a given failure. The group of the one or more assets may include a single asset, such asasset102, or multiple assets of a same or similar type, such as fleet of assets that includes theassets102 and104. Theanalytics system108 may analyze a particular amount of historical operating data, such as a certain amount of time's worth of data (e.g., a month's worth) or a certain number of data-points (e.g., the most recent thousand data-points), among other examples.
• In practice, identifying past occurrences of the given failure may involve theanalytics system108 identifying the type of operating data, such as abnormal-condition data, that indicates the given failure. In general, a given failure may be associated with one or multiple abnormal-condition indicators, such as fault codes. That is, when the given failure occurs, one or multiple abnormal-condition indicators may be triggered. As such, abnormal-condition indicators may be reflective of an underlying symptom of a given failure.
• After identifying the type of operating data that indicates the given failure, theanalytics system108 may identify the past occurrences of the given failure in a number of manners. For instance, theanalytics system108 may locate, from historical operating data stored in the databases of theanalytics system108, abnormal-condition data corresponding to the abnormal-condition indicators associated with the given failure. Each located abnormal-condition data would indicate an occurrence of the given failure. Based on this located abnormal-condition data, theanalytics system108 may identify a time at which a past failure occurred.
• Atblock706, theanalytics system108 may identify a respective set of operating data that is associated with each identified past occurrence of the given failure. In particular, theanalytics system108 may identify a set of sensor and/or actuator data from a certain timeframe around the time of the given occurrence of the given failure. For example, the set of data may be from a particular timeframe (e.g., two weeks) before, after, or around the given occurrence of the failure. In other cases, the set of data may be identified from a certain number of data-points before, after, or around the given occurrence of the failure.
• In example implementations, the set of operating data may include sensor and/or actuator data from some or all of the sensors and actuators of theasset102. For example, the set of operating data may include data from sensors and/or actuators associated with an abnormal-condition indicator corresponding to the given failure.
• To illustrate,FIG. 8 depicts a conceptual illustration of historical operating data that theanalytics system108 may analyze to facilitate defining a model. Plot800 may correspond to a segment of historical data that originated from some (e.g., Sensor A and Actuator B) or all of the sensors and actuators of theasset102. As shown, theplot800 includes time on thex-axis802, measurement values on the y-axis804, andsensor data806 corresponding to Sensor A andactuator data808 corresponding to Actuator B, each of which includes various data-points representing measurements at particular points in time, T_i. Moreover, theplot800 includes an indication of an occurrence of afailure810 that occurred at a past time, T_f(e.g., “time of failure”), and an indication of an amount oftime812 before the occurrence of the failure, ΔT, from which sets of operating data are identified. As such, T_f-ΔT defines atimeframe814 of data-points of interest.
• Returning toFIG. 7, after theanalytics system108 identifies the set of operating data for the given occurrence of the given failure (e.g., the occurrence at T_f), theanalytics system108 may determine whether there are any remaining occurrences for which a set of operating data should be identified. In the event that there is a remaining occurrence, block706 would be repeated for each remaining occurrence.
• Thereafter, atblock708, theanalytics system108 may analyze the identified sets of operating data associated with the past occurrences of the given failure to define a relationship (e.g., a failure model) between (1) a given set of operating metrics (e.g., a given set of sensor and/or actuator measurements) and (2) the likelihood of the given failure occurring within a given timeframe in the future (e.g., the next two weeks). That is, a given failure model may take as inputs sensor and/or actuator measurements from one or more sensors and/or actuators and output a probability that the given failure will occur within the given timeframe in the future.
• In general, a failure model may define a relationship between operating conditions of theasset102 and the likelihood of a failure occurring. In some implementations, in addition to raw data signals from sensors and/or actuators of theasset102, a failure model may receive a number of other data inputs, also known as features, which are derived from the sensor and/or actuator signals. Such features may include an average or range of values that were historically measured when a failure occurred, an average or range of value gradients (e.g., a rate of change in measurements) that were historically measured prior to an occurrence of a failure, a duration of time between failures (e.g., an amount of time or number of data-points between a first occurrence of a failure and a second occurrence of a failure), and/or one or more failure patterns indicating sensor and/or actuator measurement trends around the occurrence of a failure. One of ordinary skill in the art will appreciate that these are but a few example features that can be derived from sensor and/or actuator signals and that numerous other features are possible.
• In practice, a failure model may be defined in a number of manners. In example implementations, theanalytics system108 may define a failure model by utilizing one or more modeling techniques that return a probability between zero and one, which may take the form of any modeling techniques described above.
• In a particular example, defining a failure model may involve theanalytics system108 generating a response variable based on the historical operating data identified atblock706. Specifically, theanalytics system108 may determine an associated response variable for each set of sensor and/or actuator measurements received at a particular point in time. As such, the response variable may take the form of a data set associated with the failure model.
• The response variable may indicate whether the given set of measurements is within any of the timeframes determined atblock706. That is, a response variable may reflect whether a given set of data is from a time of interest about the occurrence of a failure. The response variable may be a binary-valued response variable such that, if the given set of measurements is within any of determined timeframes, the associated response variable is assigned a value of one, and otherwise, the associated response variable is assigned a value of zero.
• Returning toFIG. 8, a conceptual illustration of a response variable vector, Y_resis shown on theplot800. As shown, response variables associated with sets of measurements that are within thetimeframe814 have a value of one (e.g., Y_resat times T_i+3-T_i+8), while response variables associated with sets of measurements outside thetimeframe814 have a value of zero (e.g., Y_resat times T_i-T_i+2and T_i+9-T_i+10). Other response variables are also possible.
• Continuing in the particular example of defining a failure model based on a response variable, theanalytics system108 may train the failure model with the historical operating data identified atblock706 and the generated response variable. Based on this training process, theanalytics system108 may then define the failure model that receives as inputs various sensor and/or actuator data and outputs a probability between zero and one that a failure will occur within a period of time equivalent to the timeframe used to generate the response variable.
• In some cases, training with the historical operating data identified atblock706 and the generated response variable may result in variable importance statistics for each sensor and/or actuator. A given variable importance statistic may indicate the sensor's or actuator's relative effect on the probability that a given failure will occur within the period of time into the future.
• Additionally or alternatively, theanalytics system108 may be configured to define a failure model based on one or more survival analysis techniques, such as a Cox proportional hazard technique. Theanalytics system108 may utilize a survival analysis technique in a manner similar in some respects to the above-discussed modeling technique, but theanalytics system108 may determine a survival time-response variable that indicates an amount of time from the last failure to a next expected event. A next expected event may be either reception of senor and/or actuator measurements or an occurrence of a failure, whichever occurs first. This response variable may include a pair of values that are associated with each of the particular points in time at which measurements are received. The response variable may then be utilized to determine a probability that a failure will occur within the given timeframe in the future.
• In some example implementations, the failure model may be defined based in part on external data, such as weather data, and “hotbox” data, among other data. For instance, based on such data, the failure model may increase or decrease an output failure probability.
• In practice, external data may be observed at points in time that do not coincide with times at which asset sensors and/or actuators obtain measurements. For example, the times at which “hotbox” data is collected (e.g., times at which a locomotive passes along a section of railroad track that is outfitted with hot box sensors) may be in disagreement with sensor and/or actuator measurement times. In such cases, theanalytics system108 may be configured to perform one or more operations to determine external data observations that would have been observed at times that correspond to the sensor measurement times.
• Specifically, theanalytics system108 may utilize the times of the external data observations and times of the measurements to interpolate the external data observations to produce external data values for times corresponding to the measurement times. Interpolation of the external data may allow external data observations or features derived therefrom to be included as inputs into the failure model. In practice, various techniques may be used to interpolate the external data with the sensor and/or actuator data, such as nearest-neighbor interpolation, linear interpolation, polynomial interpolation, and spline interpolation, among other examples.
• Returning toFIG. 7, after theanalytics system108 determines a failure model for a given failure from the set of failures defined atblock702, theanalytics system108 may determine whether there are any remaining failures for which a failure model should be determined. In the event that there remains a failure for which a failure model should be determined, theanalytics system108 may repeat the loop of blocks704-708. In some implementations, theanalytics system108 may determine a single failure model that encompasses all of the failures defined atblock702. In other implementations, theanalytics system108 may determine a failure model for each subsystem of theasset102, which may then be utilized to determine an asset-level failure model. Other examples are also possible.
• Lastly, atblock710, the defined relationship for each failure in the defined set (e.g., the individual failure models) may then be combined into the model (e.g., the health-metric model) for predicting the overall likelihood of a failure occurring within the given timeframe in the future (e.g., the next two weeks). That is, the model receives as inputs sensor and/or actuator measurements from one or more sensors and/or actuators and outputs a single probability that at least one failure from the set of failures will occur within the given timeframe in the future.
• Theanalytics system108 may define the health-metric model in a number of manners, which may depend on the desired granularity of the health metric. That is, in instances where there are multiple failure models, the outcomes of the failure models may be utilized in a number of manners to obtain the output of the health-metric model. For example, theanalytics system108 may determine a maximum, median, or average from the multiple failure models and utilize that determined value as the output of the health-metric model.
• In other examples, determining the health-metric model may involve theanalytics system108 attributing a weight to individual probabilities output by the individual failure models. For instance, each failure from the set of failures may be considered equally undesirable, and so each probability may likewise be weighted the same in determining the health-metric model. In other instances, some failures may be considered more undesirable than others (e.g., more catastrophic or require longer repair time, etc.), and so those corresponding probabilities may be weighted more than others.
• In yet other examples, determining the health-metric model may involve theanalytics system108 utilizing one or more modeling techniques, such as a regression technique. An aggregate response variable may take the form of the logical disjunction (logical OR) of the response variables (e.g., Y_resinFIG. 8) from each of the individual failure models. For example, aggregate response variables associated with any set of measurements that occur within any timeframe determined at block706 (e.g., thetimeframe814 ofFIG. 8) may have a value of one, while aggregate response variables associated with sets of measurements that occur outside any of the timeframes may have a value of zero. Other manners of defining the health-metric model are also possible.
• In some implementations, block710 may be unnecessary. For example, as discussed above, theanalytics system108 may determine a single failure model, in which case the health-metric model may be the single failure model.
• In practice, theanalytics system108 may be configured to update the individual failure models and/or the overall health-metric model. Theanalytics system108 may update a model daily, weekly, monthly, etc. and may do so based on a new portion of historical operating data from theasset102 or from other assets (e.g., from other assets in the same fleet as the asset102). Other examples are also possible.
• C. Deploying Models & Workflows
• After theanalytics system108 defines a model-workflow pair, theanalytics system108 may deploy the defined model-workflow pair to one or more assets. Specifically, theanalytics system108 may transmit the defined predictive model and/or corresponding workflow to at least one asset, such as theasset102. Theanalytics system108 may transmit model-workflow pairs periodically or based on triggering events, such as any modifications or updates to a given model-workflow pair.
• In some cases, theanalytics system108 may transmit only one of an individualized model or an individualized workflow. For example, in scenarios where theanalytics system108 defined only an individualized model or workflow, theanalytics system108 may transmit an aggregate version of the workflow or model along with the individualized model or workflow, or theanalytics system108 may not need to transmit an aggregate version if theasset102 already has the aggregate version stored in data storage. In sum, theanalytics system108 may transmit (1) an individualized model and/or individualized workflow, (2) an individualized model and the aggregate workflow, (3) the aggregate model and an individualized workflow, or (4) the aggregate model and the aggregate workflow.
• In practice, theanalytics system108 may have carried out some or all of the operations of blocks702-710 ofFIG. 7 for multiple assets to define model-workflow pairs for each asset. For example, theanalytics system108 may have additionally defined a model-workflow pair for theasset104. Theanalytics system108 may be configured to transmit respective model-workflow pairs to theassets102 and104 simultaneously or sequentially.
• D. Local Execution by Asset
• A given asset, such as theasset102, may be configured to receive a model-workflow pair or a portion thereof and operate in accordance with the received model-workflow pair. That is, theasset102 may store in data storage the model-workflow pair and input into the predictive model data obtained by sensors and/or actuators of theasset102 and at times, execute the corresponding workflow based on the output of the predictive model.
• In practice, various components of theasset102 may execute the predictive model and/or corresponding workflow. For example, as discussed above, each asset may include a local analytics device configured to store and run model-workflow pairs provided by theanalytics system108. When the local analytics device receives particular sensor and/or actuator data, it may input the received data into the predictive model and depending on the output of the model, may execute one or more operations of the corresponding workflow.
• In another example, a central processing unit of theasset102 that is separate from the local analytics device may execute the predictive model and/or corresponding workflow. In yet other examples, the local analytics system and central processing unit of theasset102 may collaboratively execute the model-workflow pair. For instance, the local analytics system may execute the predictive model and the central processing unit may execute the workflow or vice versa.
• In general, an asset executing a predictive model and based on the resulting output, executing operations of the workflow may facilitate determining a cause or causes of the likelihood of a particular event occurring that is output by the model and/or may facilitate preventing a particular event from occurring in the future. In executing a workflow, an asset may locally determine and take actions to help prevent an event from occurring, which may be beneficial in situations when reliance on theanalytics system108 to make such determinations and provide recommended actions is not efficient or feasible (e.g., when there is network latency, when network connection is poor, when the asset moves out of coverage of thecommunication network106, etc.).
• In practice, an asset may execute a predictive model in a variety of manners, which may be dependent on the particular predictive model.FIG. 9 is a flow diagram900 depicting one possible example of a local-execution phase that may be used for locally executing a predictive model. The example local-execution phase will be discussed in the context of a health-metric model that outputs a health metric of an asset, but it should be understood that a same or similar local-execution phase may be utilized for other types of predictive models. Moreover, for purposes of illustration, the example local-execution phase is described as being carried out by a local analytics device of theasset102, but this phase may be carried out by other devices and/or systems as well. One of ordinary skill in the art will appreciate that the flow diagram900 is provided for sake of clarity and explanation and that numerous other combinations of operations and functions may be utilized to locally execute a predictive model.
• As shown inFIG. 9, atblock902, the local analytics device may receive data that reflects the current operating conditions of theasset102. Atblock904, the local analytics device may identify, from the received data, the set of operating data that is to be input into the model provided by theanalytics system108. Atblock906, the local analytics device may then input the identified set of operating data into the model and run the model to obtain a health metric for theasset102.
• As the local analytics device continues to receive updated operating data for theasset102, the local analytics device may also continue to update the health metric for theasset102 by repeating the operations of blocks902-906 based on the updated operating data. In some cases, the operations of blocks902-906 may be repeated each time the local analytics device receives new data from sensors and/or actuators of theasset102 or periodically (e.g., hourly, daily, weekly, monthly, etc.). In this way, local analytics devices may be configured to dynamically update health metrics, perhaps in real-time, as assets are used in operation.
• The functions of the example local-execution phase illustrated inFIG. 9 will now be described in further detail. Atblock902, the local analytics device may receive data that reflects the current operating conditions of theasset102. Such data may include sensor data from one or more of the sensors of theasset102, actuator data from one or more actuators of theasset102, and/or it may include abnormal-condition data, among other types of data.
• Atblock904, the local analytics device may identify, from the received data, the set of operating data that is to be input into the health-metric model provided by theanalytics system108. This operation may be performed in a number of manners.
• In one example, the local analytics device may identify the set of operating data inputs (e.g., data from particular sensors and/or actuators of interest) for the model based on a characteristic of theasset102, such as asset type or asset class, for which the health metric is being determined. In some cases, the identified set of operating data inputs may be sensor data from some or all of the sensors of theasset102 and/or actuator data from some of all of the actuators of theasset102.
• In another example, the local analytics device may identify the set of operating data inputs based on the predictive model provided by theanalytics system108. That is, theanalytics system108 may provide some indication to the asset102 (e.g., either in the predictive model or in a separate data transmission) of the particular inputs for the model. Other examples of identifying the set of operating data inputs are also possible.
• Atblock906, the local analytics device may then run the health-metric model. Specifically, the local analytics device may input the identified set of operating data into the model, which in turn determines and outputs an overall likelihood of at least one failure occurring within the given timeframe in the future (e.g., the next two weeks).
• In some implementations, this operation may involve the local analytics device inputting particular operating data (e.g., sensor and/or actuator data) into one or more individual failure models of the health-metric model, which each may output an individual probability. The local analytics device may then use these individual probabilities, perhaps weighting some more than others in accordance with the health-metric model, to determine the overall likelihood of a failure occurring within the given timeframe in the future.
• After determining the overall likelihood of a failure occurring, the local analytics device may convert the probability of a failure occurring into the health metric that may take the form of a single, aggregated parameter that reflects the likelihood that no failures will occur at theasset102 within the give timeframe in the future (e.g., two weeks). In example implementations, converting the failure probability into the health metric may involve the local analytics device determining the complement of the failure probability. Specifically, the overall failure probability may take the form of a value ranging from zero to one; the health metric may be determined by subtracting one by that number. Other examples of converting the failure probability into the health metric are also possible.
• After an asset locally executes a predictive model, the asset may then execute a corresponding workflow based on the resulting output of the executed predictive model. As mentioned above, workflows may take various forms and so, workflows may be executed in a variety of manners.
• For example, theasset102 may internally execute one or more operations that modify some behavior of theasset102, such as modifying a data-acquisition and/or -transmission scheme, executing a local diagnostic tool, modifying an operating condition of the asset102 (e.g., modifying a velocity, acceleration, fan speed, propeller angle, air intake, etc. via one or more actuators of the asset102), or outputting an indication, perhaps of a relatively low health metric or of recommended preventative actions that should be executed in relation to theasset102, at a user interface of theasset102 or to an external computing system.
• In another example, theasset102 may transmit to a system on thecommunication network106, such as theoutput system110, an instruction to cause the system to carry out an operation, such as generating a work-order or ordering a particular part for a repair of theasset102. Other examples of theasset102 locally executing a workflow are also possible.
• E. Model/Workflow Modification Phase
• In another aspect, theanalytics system108 may carry out a modification phase during which theanalytics system108 modifies a deployed model and/or workflow based on new asset data. This phase may be performed for both aggregate and individualized models and workflows.
• In particular, as a given asset (e.g., the asset102) operates in accordance with a model-workflow pair, theasset102 may provide operating data to theanalytics system108 and/or thedata source112 may provide to theanalytics system108 external data related to theasset102. Based at least on this data, theanalytics system108 may modify the model and/or workflow for theasset102 and/or the model and/or workflow for other assets, such as theasset104. In modifying models and/or workflows for other assets, theanalytics system108 may share information learned from the behavior of theasset102.
• In practice, theanalytics system108 may make modifications in a number of manners.FIG. 10 is a flow diagram1000 depicting one possible example of a modification phase that may be used for modifying model-workflow pairs. For purposes of illustration, the example modification phase is described as being carried out by theanalytics system108, but this modification phase may be carried out by other systems as well. One of ordinary skill in the art will appreciate that the flow diagram1000 is provided for sake of clarity and explanation and that numerous other combinations of operations may be utilized to modify model-workflow pairs.
• As shown inFIG. 10, atblock1002, theanalytics system108 may receive data from which theanalytics system108 identifies an occurrence of a particular event. The data may be operating data originating from theasset102 or external data related to theasset102 from thedata source112, among other data. The event may take the form of any of the events discussed above, such as a failure at theasset102.
• In other example implementations, the event may take the form of a new component or subsystem being added to theasset102. Another event may take the form of a “leading indicator” event, which may involve sensors and/or actuators of theasset102 generating data that differs, perhaps by a threshold differential, from the data identified atblock706 ofFIG. 7 during the model-definition phase. This difference may indicate that theasset102 has operating conditions that are above or below normal operating conditions for assets similar to theasset102. Yet another event may take the form of an event that is followed by one or more leading indicator events.
• Based on the identified occurrence of the particular event and/or the underlying data (e.g., operating data and/or external data related to the asset102), theanalytics system108 may then modify the aggregate, predictive model and/or workflow and/or one or more individualized predictive models and/or workflows. In particular, atblock1004, theanalytics system108 may determine whether to modify the aggregate, predictive model. Theanalytics system108 may determine to modify the aggregate, predictive model for a number of reasons.
• For example, theanalytics system108 may modify the aggregate, predictive model if the identified occurrence of the particular event was the first occurrence of this particular event for a plurality of assets including theasset102, such as the first time a particular failure occurred at an asset from a fleet of assets or the first time a particular new component was added to an asset from a fleet of assets.
• In another example, theanalytics system108 may make a modification if data associated with the identified occurrence of the particular event is different from data that was utilized to originally define the aggregate model. For instance, the identified occurrence of the particular event may have occurred under operating conditions that had not previously been associated with an occurrence of the particular event (e.g., a particular failure might have occurred with associated sensor values not previously measured before with the particular failure). Other reasons for modifying the aggregate model are also possible.
• If theanalytics system108 determines to modify the aggregate, predictive model, theanalytics system108 may do so atblock1006. Otherwise, theanalytics system108 may proceed to block1008.
• Atblock1006, theanalytics system108 may modify the aggregate model based at least in part on the data related to theasset102 that was received atblock1002. In example implementations, the aggregate model may be modified in various manners, such as any manner discussed above with reference to block510 ofFIG. 5. In other implementations, the aggregate model may be modified in other manners as well.
• Atblock1008, theanalytics system108 may then determine whether to modify the aggregate workflow. Theanalytics system108 may modify the aggregate workflow for a number of reasons.
• For example, theanalytics system108 may modify the aggregate workflow based on whether the aggregate model was modified atblock1004 and/or if there was some other change at theanalytics system108. In other examples, theanalytics system108 may modify the aggregate workflow if the identified occurrence of the event atblock1002 occurred despite theasset102 executing the aggregate workflow. For instance, if the workflow was aimed to help facilitate preventing the occurrence of the event (e.g., a failure) and the workflow was executed properly but the event still occurred nonetheless, then theanalytics system108 may modify the aggregate workflow. Other reasons for modifying the aggregate workflow are also possible.
• If theanalytics system108 determines to modify the aggregate workflow, theanalytics system108 may do so atblock1010. Otherwise, theanalytics system108 may proceed to block1012.
• Atblock1010, theanalytics system108 may modify the aggregate workflow based at least in part on the data related to theasset102 that was received atblock1002. In example implementations, the aggregate workflow may be modified in various manners, such as any manner discussed above with reference to block514 ofFIG. 5. In other implementations, the aggregate model may be modified in other manners as well.
• Atblocks1012 throughblocks1018, theanalytics system108 may be configured to modify one or more individualized models (e.g., for each ofassets102 and104) and/or one or more individualized workflows (e.g., for one ofasset102 or asset104) based at least in part on the data related to theasset102 that was received atblock1002. Theanalytics system108 may do so in a manner similar to blocks1004-1010.
• However, the reasons for modifying an individualized model or workflow may differ from the reasons for the aggregate case. For instance, theanalytics system108 may further consider the underlying asset characteristics that were utilized to define the individualized model and/or workflow in the first place. In a particular example, theanalytics system108 may modify an individualized model and/or workflow if the identified occurrence of the particular event was the first occurrence of this particular event for assets with asset characteristics of theasset102. Other reasons for modifying an individualized model and/or workflow are also possible.
• To illustrate,FIG. 6D is a conceptual illustration of a modified model-workflow pair630. Specifically, the model-workflow pair illustration630 is a modified version of the aggregate model-workflow pair fromFIG. 6A. As shown, the modified model-workflow pair illustration630 includes the original column formodel inputs602 fromFIG. 6A and includes modified columns formodel calculations634, model output ranges636, andworkflow operations638. In this example, the modified predictive model has a single input, data from Sensor A, and has two calculations, Calculations I and III. If the output probability of the modified model is less than 75%, thenworkflow Operation1 is performed. If the output probability is between 75% and 85%, thenworkflow Operation2 is performed. And if the output probability is greater than 85%, thenworkflow Operation3 is performed. Other example modified model-workflow pairs are possible and contemplated herein.
• Returning toFIG. 10, atblock1020, theanalytics system108 may then transmit any model and/or workflow modifications to one or more assets. For example, theanalytics system108 may transmit a modified individualized model-workflow pair to the asset102 (e.g., the asset whose data caused the modification) and a modified aggregate model to theasset104. In this way, theanalytics system108 may dynamically modify models and/or workflows based on data associated with the operation of theasset102 and distribute such modifications to multiple assets, such as the fleet to which theasset102 belongs. Accordingly, other assets may benefit from the data originating from theasset102 in that the other assets' local model-workflow pairs may be refined based on such data, thereby helping to create more accurate and robust model-workflow pairs
• F. Dynamic Execution of Model/Workflow
• In another aspect, theasset102 and/or theanalytics system108 may be configured to dynamically adjust executing a model-workflow pair. In particular, theasset102 and/or theanalytics system108 may be configured to detect certain events that trigger a change in responsibilities with respect to whether theasset102 and/or theanalytics system108 should be executing the predictive model and/or workflow.
• In operation, both theasset102 and theanalytics system108 may execute all or a part of a model-workflow pair on behalf of theasset102. For example, after theasset102 receives a model-workflow pair from theanalytics system108, theasset102 may store the model-workflow pair in data storage but then may rely on theanalytics system108 to centrally execute part or all of the model-workflow pair. In particular, theasset102 may provide at least sensor and/or actuator data to theanalytics system108, which may then use such data to centrally execute a predictive model for theasset102. Based on the output of the model, theanalytics system108 may then execute the corresponding workflow or theanalytics system108 may transmit to theasset102 the output of the model or an instruction for theasset102 to locally execute the workflow.
• In other examples, theanalytics system108 may rely on theasset102 to locally execute part or all of the model-workflow pair. Specifically, theasset102 may locally execute part or all of the predictive model and transmit results to theanalytics system108, which may then cause theanalytics system108 to centrally execute the corresponding workflow. Or theasset102 may also locally execute the corresponding workflow.
• In yet other examples, theanalytics system108 and theasset102 may share in the responsibilities of executing the model-workflow pair. For instance, theanalytics system108 may centrally execute portions of the model and/or workflow, while theasset102 locally executes the other portions of the model and/or workflow. Theasset102 andanalytics system108 may transmit results from their respective executed responsibilities. Other examples are also possible.
• At some point in time, theasset102 and/or theanalytics system108 may determine that the execution of the model-workflow pair should be adjusted. That is, one or both may determine that the execution responsibilities should be modified. This operation may occur in a variety of manners.
• FIG. 11 is a flow diagram1100 depicting one possible example of an adjustment phase that may be used for adjusting execution of a model-workflow pair. For purposes of illustration, the example adjustment phase is described as being carried out by theasset102 and/or theanalytics system108, but this modification phase may be carried out by other systems as well. One of ordinary skill in the art will appreciate that the flow diagram1100 is provided for sake of clarity and explanation and that numerous other combinations of operations may be utilized to adjust the execution of a model-workflow pair.
• Atblock1102, theasset102 and/or theanalytics system108 may detect an adjustment factor (or potentially multiple) that indicates conditions that require an adjustment to the execution of the model-workflow pair. Examples of such conditions include network conditions of thecommunication network106 or processing conditions of theasset102 and/oranalytics system108, among other examples. Example network conditions may include network latency, network bandwidth, signal strength of a link between theasset102 and thecommunication network106, or some other indication of network performance, among other examples. Example processing conditions may include processing capacity (e.g., available processing power), processing usage (e.g., amount of processing power being consumed) or some other indication of processing capabilities, among other examples.
• In practice, detecting an adjustment factor may be performed in a variety of manners. For example, this operation may involve determining whether network (or processing) conditions reach one or more threshold values or whether conditions have changed in a certain manner. Other examples of detecting an adjustment factor are also possible.
• In particular, in some cases, detecting an adjustment factor may involve theasset102 and/or theanalytics system108 detecting an indication that a signal strength of a communication link between theasset102 and theanalytics system108 is below a threshold signal strength or has been decreasing at a certain rate of change. In this example, the adjustment factor may indicate that theasset102 is about to go “off-line.”
• In another case, detecting an adjustment factor may additionally or alternatively involve theasset102 and/or theanalytics system108 detecting an indication that network latency is above a threshold latency or has been increasing at a certain rate of change. Or the indication may be that a network bandwidth is below a threshold bandwidth or has been decreasing at a certain rate of change. In these examples, the adjustment factor may indicate that thecommunication network106 is lagging.
• In yet other cases, detecting an adjustment factor may additionally or alternatively involve theasset102 and/or theanalytics system108 detecting an indication that processing capacity is below a particular threshold or has been decreasing at a certain rate of change and/or that processing usage is above a threshold value or increasing at a certain rate of change. In such examples, the adjustment factor may indicate that processing capabilities of the asset102 (and/or the analytics system108) are low. Other examples of detecting an adjustment factor are also possible.
• Atblock1104, based on the detected adjustment factor, the local execution responsibilities may be adjusted, which may occur in a number of manners. For example, theasset102 may have detected the adjustment factor and then determined to locally execute the model-workflow pair or a portion thereof. In some cases, theasset102 may then transmit to the analytics system108 a notification that theasset102 is locally executing the predictive model and/or workflow.
• In another example, theanalytics system108 may have detected the adjustment factor and then transmitted an instruction to theasset102 to cause theasset102 to locally execute the model-workflow pair or a portion thereof. Based on the instruction, theasset102 may then locally execute the model-workflow pair.
• Atblock1106, the central execution responsibilities may be adjusted, which may occur in a number of manners. For example, the central execution responsibilities may be adjusted based on theanalytics system108 detecting an indication that theasset102 is locally executing the predictive model and/or the workflow. Theanalytics system108 may detect such an indication in a variety of manners.
• In some examples, theanalytics system108 may detect the indication by receiving from the asset102 a notification that theasset102 is locally executing the predictive model and/or workflow. The notification may take various forms, such as binary or textual, and may identify the particular predictive model and/or workflow that the asset is locally executing.
• In other examples, theanalytics system108 may detect the indication based on received operating data for theasset102. Specifically, detecting the indication may involve theanalytics system108 receiving operating data for theasset102 and then detecting one or more characteristics of the received data. From the one or more detected characteristics of the received data, theanalytics system108 may infer that theasset102 is locally executing the predictive model and/or workflow.
• In practice, detecting the one or more characteristics of the received data may be performed in a variety of manners. For instance, theanalytics system108 may detect a type of the received data. In particular, theanalytics system108 may detect a source of the data, such as a particular sensor or actuator that generated sensor or actuator data. Based on the type of the received data, theanalytics system108 may infer that theasset102 is locally executing the predictive model and/or workflow. For example, based on detecting a sensor-identifier of a particular sensor, theanalytics system108 may infer the thatasset102 is locally executing a predictive model and corresponding workflow that causes theasset102 to acquire data from the particular sensor and transmit that data to theanalytics system108.
• In another instance, theanalytics system108 may detect an amount of the received data. Theanalytics system108 may compare that amount to a certain threshold amount of data. Based on the amount reaching the threshold amount, theanalytics system108 may infer that theasset102 is locally executing a predictive model and/or workflow that causes theasset102 to acquire an amount of data equivalent to or greater than the threshold amount. Other examples are also possible.
• In example implementations, detecting the one or more characteristics of the received data may involve theanalytics system108 detecting a certain change in one or more characteristics of the received data, such as a change in the type of the received data, a change in the amount of data that is received, or change in the frequency at which data is received. In a particular example, a change in the type of the received data may involve theanalytics system108 detecting a change in the source of sensor data that it is receiving (e.g., a change in sensors and/or actuators that are generating the data provided to the analytics system108).
• In some cases, detecting a change in the received data may involve theanalytics system108 comparing recently received data to data received in the past (e.g., an hour, day, week, etc. before a present time). In any event, based on detecting the change in the one or more characteristics of the received data, theanalytics system108 may infer that theasset102 is locally executing a predictive model and/or workflow that causes such a change to the data provided by theasset102 to theanalytics system108.
• Moreover, theanalytics system108 may detect an indication that theasset102 is locally executing the predictive model and/or the workflow based on detecting the adjustment factor atblock1102. For example, in the event that theanalytics system108 detects the adjustment factor atblock1102, theanalytics system108 may then transmit to theasset102 instructions that cause theasset102 to adjust its local execution responsibilities and accordingly, theanalytics system108 may adjust its own central execution responsibilities. Other examples of detecting the indication are also possible.
• In example implementations, the central execution responsibilities may be adjusted in accordance with the adjustment to the local execution responsibilities. For instance, if theasset102 is now locally executing the predictive model, then theanalytics system108 may accordingly cease centrally executing the predictive model (and may or may not cease centrally executing the corresponding workflow). Further, if theasset102 is locally executing the corresponding workflow, then theanalytics system108 may accordingly cease executing the workflow (and may or may not cease centrally executing the predictive model). Other examples are also possible.
• In practice, theasset102 and/or theanalytics system108 may continuously perform the operations of blocks1102-1106. And at times, the local and central execution responsibilities may be adjusted to facilitate optimizing the execution of model-workflow pairs.
• Moreover, in some implementations, theasset102 and/or theanalytics system108 may perform other operations based on detecting an adjustment factor. For example, based on a condition of the communication network106 (e.g., bandwidth, latency, signal strength, or another indication of network quality), theasset102 may locally execute a particular workflow. The particular workflow may be provided by theanalytics system108 based on theanalytics system108 detecting the condition of the communication network, may be already stored on theasset102, or may be a modified version of a workflow already stored on the asset102 (e.g., theasset102 may locally modify a workflow). In some cases, the particular workflow may include a data-acquisition scheme that increases or decreases a sampling rate and/or a data-transmission scheme that increases or decreases a transmission rate or amount of data transmitted to theanalytics system108, among other possible workflow operations.
• In a particular example, theasset102 may determine that one or more detected conditions of the communication network have reached respective thresholds (e.g., indicating poor network quality). Based on such a determination, theasset102 may locally execute a workflow that includes transmitting data according to a data-transmission scheme that reduces the amount and/or frequency of data theasset102 transmits to theanalytics system108. Other examples are also possible.
• V. Example Methods
• Turning now toFIG. 12, a flow diagram is depicted illustrating anexample method1200 for defining and deploying an aggregate, predictive model and corresponding workflow that may be performed by theanalytics system108. For themethod1200 and the other methods discussed below, the operations illustrated by the blocks in the flow diagrams may be performed in line with the above discussion. Moreover, one or more operations discussed above may be added to a given flow diagram.
• Atblock1202, themethod1200 may involve theanalytics system108 receiving respective operating data for a plurality of assets (e.g., theassets102 and104). Atblock1204, themethod1200 may involve theanalytics system108, based on the received operating data, defining a predictive model and a corresponding workflow (e.g., a failure model and corresponding workflow) that are related to the operation of the plurality of assets. Atblock1206, themethod1200 may involve theanalytics system108 transmitting to at least one asset of the plurality of assets (e.g., the asset102) the predictive model and the corresponding workflow for local execution by the at least one asset.
• FIG. 13 depicts a flow diagram of anexample method1300 for defining and deploying an individualized, predictive model and/or corresponding workflow that may be performed by theanalytics system108. Atblock1302, themethod1300 may involve theanalytics system108 receiving operating data for a plurality of assets, where the plurality of assets includes at least a first asset (e.g., the asset102). Atblock1304, themethod1300 may involve theanalytics system108, based on the received operating data, defining an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets. Atblock1306, themethod1300 may involve theanalytics system108 determining one or more characteristics of the first asset. Atblock1308, themethod1300 may involve theanalytics system108, based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, defining at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset. Atblock1310, themethod1300 may involve theanalytics system108 transmitting to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.
• FIG. 14 depicts a flow diagram of anexample method1400 for dynamically modifying the execution of model-workflow pairs that may be performed by theanalytics system108. Atblock1402, themethod1400 may involve theanalytics system108 transmitting to an asset (e.g., the asset102) a predictive model and corresponding workflow that are related to the operation of the asset for local execution by the asset. Atblock1404, themethod1400 may involve theanalytics system108 detecting an indication that the asset is locally executing at least one of the predictive model or the corresponding workflow. Atblock1406, themethod1400 may involve theanalytics system108, based on the detected indication, modifying central execution by the computing system of at least one of the predictive model or the corresponding workflow.
• Similar tomethod1400, another method for dynamically modifying the execution of model-workflow pairs may be performed by an asset (e.g., the asset102). For instance, such a method may involve theasset102 receiving from a central computing system (e.g., the analytics system108) a predictive model and corresponding workflow that are related to the operation of theasset102. The method may also involve theasset102 detecting an adjustment factor indicating one or more conditions associated with adjusting execution of the predictive model and the corresponding workflow. The method may involve, based on the detected adjustment factor, (i) modifying local execution by theasset102 of at least one of the predictive model or the corresponding workflow and (ii) transmitting to the central computing system an indication that theasset102 is locally executing the at least one of the predictive model or the corresponding workflow to facilitate causing the central computing system to modify central execution by the computing system of at least one of the predictive model or the corresponding workflow.
• To the extent that examples described herein involve operations performed or initiated by actors, such as “humans”, “operators”, “users” or other entities, this is for purposes of example and explanation only. The claims should not be construed as requiring action by such actors unless explicitly recited in the claim language.

Claims (20)

1. A computing system comprising:

at least one processor;

a non-transitory computer-readable medium; and

program instructions stored on the non-transitory computer-readable medium that are executable by the at least one processor to cause the computing system to:

receive operating data for a plurality of assets, wherein the plurality of assets comprises a first asset;

based on the received operating data, define an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets;

determine one or more characteristics of the first asset;

based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, define at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset; and

transmit to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.

2. The computing system ofclaim 1, wherein the one or more characteristics of the first asset comprises at least one of an asset age or an asset health.

3. The computing system ofclaim 1, wherein determining the one or more characteristics of the first asset comprises determining the one or more characteristics of the first asset based on received operating data for the first asset.

4. The computing system ofclaim 1, wherein defining at least one of an individualized predictive model or an individualized corresponding workflow comprises defining the individualized predictive model and the individualized corresponding workflow, and wherein transmitting the at least one individualized predictive model or individualized corresponding workflow comprises transmitting the individualized predictive model and the individualized corresponding workflow.

5. The computing system ofclaim 1, wherein defining at least one of an individualized predictive model or an individualized corresponding workflow comprises defining the individualized corresponding workflow, and wherein transmitting the at least one individualized predictive model or individualized corresponding workflow comprises transmitting the aggregate predictive model and the individualized corresponding workflow.

6. The computing system ofclaim 5, wherein the aggregate corresponding workflow comprises a first operation, and wherein the individualized corresponding workflow comprises a second operation that differs from the first operation.

7. The computing system ofclaim 6, wherein the first operation comprises acquiring data according to a first acquisition scheme, and wherein the second operation comprises acquiring data according to a second acquisition scheme.

8. The computing system ofclaim 6, wherein the first operation comprises acquiring data according to an acquisition scheme, and wherein the second operation comprises executing one or more diagnostic tools.

9. The computing system ofclaim 1, wherein the plurality of assets further comprises a second asset, and wherein the program instructions further comprise instructions that are executable to cause the computing system to:

after transmitting the at least one individualized predictive model or individualized corresponding workflow, receive operating data for the second asset indicating an occurrence of an event at the second asset;

based on the received operating data for the second asset, modify the at least one individualized predictive model or individualized corresponding workflow; and

transmit to the first asset the modified at least one individualized predictive model or individualized corresponding workflow.

10. A non-transitory computer-readable medium having instructions stored thereon that are executable to cause a computing system to:

receive operating data for a plurality of assets, wherein the plurality of assets comprises a first asset;

based on the received operating data, define an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets;

determine one or more characteristics of the first asset;

based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, define at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset; and

transmit to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.

11. The non-transitory computer-readable medium ofclaim 10, wherein defining at least one of an individualized predictive model or an individualized corresponding workflow comprises defining the individualized predictive model and the individualized corresponding workflow, and wherein transmitting the at least one individualized predictive model or individualized corresponding workflow comprises transmitting the individualized predictive model and the individualized corresponding workflow.

12. The non-transitory computer-readable medium ofclaim 10, wherein defining at least one of an individualized predictive model or an individualized corresponding workflow comprises defining the individualized corresponding workflow, and wherein transmitting the at least one individualized predictive model or individualized corresponding workflow comprises transmitting the aggregate predictive model and the individualized corresponding workflow.

13. The non-transitory computer-readable medium ofclaim 12, wherein the aggregate corresponding workflow comprises a first operation, and wherein the individualized corresponding workflow comprises a second operation that differs from the first operation.

14. The non-transitory computer-readable medium ofclaim 13, wherein the first operation comprises acquiring data according to a first acquisition scheme, and wherein the second operation comprises acquiring data according to a second acquisition scheme.

15. The non-transitory computer-readable medium ofclaim 13, wherein the first operation comprises acquiring data according to an acquisition scheme, and wherein the second operation comprises executing one or more diagnostic tools.

16. The non-transitory computer-readable medium ofclaim 10, wherein the plurality of assets further comprises a second asset, and wherein the program instructions further comprise instructions that are executable to cause the computing system to:

after transmitting the at least one individualized predictive model or individualized corresponding workflow, receive operating data for the second asset indicating an occurrence of an event at the second asset;

based on the received operating data for the second asset, modify the at least one individualized predictive model or individualized corresponding workflow; and

transmit to the first asset the modified at least one individualized predictive model or individualized corresponding workflow.

17. A computer-implemented method comprising:

receiving operating data for a plurality of assets, wherein the plurality of assets comprises a first asset;

based on the received operating data, defining an aggregate predictive model and an aggregate corresponding workflow that are related to the operation of the plurality of assets;

determining one or more characteristics of the first asset;

based on the one or more characteristics of the first asset and the aggregate predictive model and the aggregate corresponding workflow, defining at least one of an individualized predictive model or an individualized corresponding workflow that is related to the operation of the first asset; and

transmitting to the first asset the defined at least one individualized predictive model or individualized corresponding workflow for local execution by the first asset.

18. The computer-implemented method ofclaim 17, wherein defining at least one of an individualized predictive model or an individualized corresponding workflow comprises defining the individualized corresponding workflow, and wherein transmitting the at least one individualized predictive model or individualized corresponding workflow comprises transmitting the aggregate predictive model and the individualized corresponding workflow.

19. The computer-implemented method ofclaim 18, wherein the aggregate corresponding workflow comprises a first operation, and wherein the individualized corresponding workflow comprises a second operation that differs from the first operation.

20. The computer-implemented method ofclaim 19, wherein one of the first operation or the second operation comprises executing one or more diagnostic tools.

Images