Disclosure of Invention
The application aims to at least solve one of the technical problems in the prior art, and therefore, the application provides a factory equipment fault early warning system and method based on the Internet of things, which are used for solving the technical problems that the equipment operation data cannot be accurately predicted in the prior art, and the fault early warning precision is difficult to guarantee.
In order to achieve the above object, a first aspect of the present application provides a factory equipment fault early warning system based on the internet of things, which comprises a data processing layer and a data acquisition layer connected with the data processing layer;
the data acquisition layer is used for acquiring a plurality of types of equipment operation data through an intelligent sensor arranged in a factory, judging whether the equipment operation data exceeds a standard data range, and generating an early warning signal if the equipment operation data exceeds the standard data range, wherein the equipment operation data refers to data related to the equipment operation state;
A data processing layer for inputting a plurality of device operation data into the first neural network, predicting to obtain range limit values of a plurality of predicted time points, wherein the range limit values comprise an upper limit value and a lower limit value, and
The method comprises the steps of obtaining a prediction data set by combining a plurality of range limit values of each prediction time point, inputting the prediction data set into a neural network II, and obtaining the fault information of each prediction time point by prediction.
Preferably, predicting the range limit for a number of predicted time points includes:
Determining the time sequence number of the input data of the model according to the data input dimension of the first neural network, and marking the time sequence number as the first time sequence number, wherein the first neural network is constructed based on the long-short-period memory neural network;
And the data vector I is input into the neural network I to obtain the range limit value of the running data of each type of equipment at the predicted time point.
Preferably, constructing the first neural network based on the long-term and short-term memory neural network includes:
constructing a long-term memory neural network, wherein the long-term memory neural network comprises an LSTM layer, the LSTM layer comprises a plurality of LSTM units, and each LSTM unit comprises a forgetting gate, an input gate and an output gate;
Extracting data from the stored equipment operation data according to the first time sequence number and the second time sequence number, and integrating the extracted data into model training data, wherein the second time sequence number is equal to the number of the predicted time points;
And training the long-term memory neural network through model training data to obtain a first neural network.
Preferably, the range limit value of each predicted time point is compared with the standard data range, when the range limit value is out of the standard data range of the corresponding data type, early warning is carried out, and otherwise, early warning is not carried out.
Preferably, obtaining the predicted dataset comprises:
Sequentially extracting a plurality of range limits of a predicted time point, and integrating the range limits into a target data set, wherein the number of the range limits in the target data set is equal to the type of the equipment operation data;
And combining the range limit values of each type in the target data set to obtain a prediction data set.
Preferably, combining the range limits of each type in the target data set includes:
Taking the range limit value of each type in the target data set as a limit value box;
selecting a limit value from each limit value box, integrating all the selected limit values into one piece of prediction data, and integrating all the obtained prediction data into a prediction data set.
Preferably, combining the range limits of each type in the target data set includes:
dividing the range limit value of each type in the target data set according to a set interval to obtain a plurality of numerical value points;
Selecting one numerical point from a plurality of numerical points corresponding to each type of data, integrating all the selected numerical points into one piece of prediction data, and integrating all the obtained prediction data into a prediction data set.
Preferably, predicting obtains fault information of each time point, including:
the method comprises the steps of performing tag integration on a predicted data set of each predicted time point to obtain predicted input data, wherein the tag integration refers to the step of inserting a time tag into the predicted data set, and the time tag is a positive integer;
and inputting the predicted input data into a second neural network to obtain fault information corresponding to the predicted time point, wherein the fault information comprises a fault type and fault probability.
Preferably, after the fault information is determined through the neural network two, the fault probability of each fault type in the fault information is calibrated by using a normal distribution model.
The second aspect of the application provides a factory equipment fault early warning method based on the Internet of things, which comprises the following steps:
Acquiring a plurality of types of equipment operation data through an intelligent sensor arranged in a factory, judging whether the equipment operation data exceeds a standard data range, and generating an early warning signal if the equipment operation data exceeds the standard data range, wherein the equipment operation data refers to data related to the equipment operation state;
inputting a plurality of equipment operation data into a first neural network, and predicting to obtain range limit values of a plurality of predicted time points, wherein the range limit values comprise an upper limit value and a lower limit value;
And inputting the prediction data set into a second neural network to predict and obtain fault information of each time point.
Compared with the prior art, the application has the beneficial effects that:
1. According to the application, the range limit value of each type of data at each time point in the future is predicted through the first neural network on the basis of collecting the equipment operation data, if the range limit value is within the standard data range of the corresponding data type, the equipment cannot fail, the defect that the single numerical value prediction precision is insufficient to influence the fault early warning precision can be avoided, meanwhile, the fault information of each prediction time point is predicted through the second neural network, the possible faults of the equipment can be further analyzed under the normal condition of the single type of data, and missing of the equipment faults can be avoided.
2. When the equipment is predicted to be faulty, the probability of the equipment fault under the condition that the running data of the equipment are all in the same direction extremum is considered, the probability of the equipment fault under the condition that the running data of the equipment are in the opposite direction extremum is comprehensively considered, and the accuracy of fault information can be improved.
Detailed Description
The technical solutions of the present application will be clearly and completely described in connection with the embodiments, and it is obvious that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The combination of a machine learning algorithm and the Internet of things is a mainstream scheme of fault early warning of equipment in a factory at present, data related to equipment operation is collected through the Internet of things, and then the data is mined through a pre-trained machine learning algorithm, so that the fault early warning of the equipment is realized. The scheme has higher requirements on data, but the data is influenced by various factors in the actual operation of the equipment, so that the accurate operation data of the equipment in the future is difficult to predict, and the accuracy of the equipment fault early warning result cannot be ensured.
Embodiment one:
Referring to fig. 1-2, an embodiment of the first aspect of the present application provides a factory equipment fault early warning system based on internet of things, which comprises a data processing layer and a data acquisition layer connected with the data processing layer, wherein the data acquisition layer is used for acquiring a plurality of types of equipment operation data through an intelligent sensor arranged in a factory;
The data processing layer is used for inputting a plurality of device operation data into the first neural network to predict and obtain range limit values of a plurality of prediction time points, combining the range limit values of the prediction time points to obtain a prediction data set, inputting the prediction data set into the second neural network, and predicting and obtaining fault information of the prediction time points.
The data acquisition layer is mainly used for acquiring equipment operation data through an intelligent sensor arranged on the equipment, and the equipment operation data is transmitted to the data processing layer after necessary processing. The data processing layer analyzes and mines the equipment operation data through a built-in machine learning algorithm and predicts equipment faults.
The equipment operation data collected by the intelligent sensor is related to the equipment operation state, and whether the equipment has faults or not can be judged according to the equipment operation data. The equipment operation data comprise temperature, pressure, vibration, humidity and other types of data, the intelligent sensor is matched with the type of the equipment operation data, and a correction model is built in the intelligent sensor, so that the accuracy of the acquired data can be ensured.
In this embodiment, future equipment operation data is not directly predicted by existing equipment operation data, and because the future equipment operation data is affected by many factors, it may be affected by not only the environment but also other data, and it is difficult to predict the future equipment operation data. The method predicts the range of the equipment operation data according to the existing equipment operation data, namely the future equipment operation data should change within the predicted range, the predicted range is lower than the difficulty of predicting the accurate data, and the accuracy of the predicted data range is higher under the condition that the equipment is in normal operation.
After the range limit values of the operation data of the various types of equipment are obtained through prediction, the extreme values in the range limit values can be taken as prediction bases, and the extreme values are taken as the operation data of the equipment to predict whether the equipment is likely to fail and the probability of failure. If the device does not malfunction in extreme cases, it is also not possible for the device to malfunction within the limits. If the equipment is likely to fail in the extreme case, it is possible to determine at set intervals which part of the range limits of the operation data of each equipment fails and which part does not fail.
It should be noted that, the above-mentioned operation data of the device include temperature, humidity, pressure, vibration amplitude, etc., which are relatively stable when the device is operating normally, and once some type of data is suddenly changed, the device may malfunction, which is a feature of major attention of the existing mainstream early warning scheme. However, in the actual operation of the device, even if the data of each type does not have abrupt change or the abrupt change does not reach the threshold value, the device may have a fault, for example, the data of each type deviates from the steady state, but the deviation does not exceed the threshold value, and in this case, the existing scheme cannot predict the device fault corresponding to the situation.
When the equipment operation data acquired by the intelligent sensor exceeds the standard data range, the abnormal condition exists, an early warning signal is generated, and the neural network can be utilized to analyze the equipment operation data corresponding to the equipment at present, so that fault information is determined.
One of the important tasks of this embodiment is to predict the range limit of the operation data of each type of device, so as to perform subsequent device fault early warning. The method for determining the range limit value provided in this embodiment is as follows:
The method comprises the steps of determining the time sequence number of model input data according to the data input dimension of a neural network I, marking the time sequence number as the time sequence number I, extracting the data of the time sequence number I from the operation data of various types of equipment by taking the current moment as a reference, splicing the extracted data into a data vector I according to the data type, inputting the data vector I into the neural network I, and obtaining the range limit value of the operation data of various types of equipment at a prediction time point.
The neural network is constructed based on a long-term and short-term memory neural network, and comprises the following components:
extracting data from the stored equipment operation data according to the first time sequence number and the second time sequence number, and integrating the extracted data into model training data;
And training the long-term memory neural network through model training data to obtain a first neural network.
The long-term and short-term memory neural network in the embodiment at least comprises an input layer, an LSTM layer, a full-connection layer and an output layer, wherein the LSTM layer comprises a plurality of LSTM units, and each LSTM unit comprises a forgetting gate, an input gate and an output gate.
The stored equipment operation data refers to data acquired by equipment to be subjected to fault early warning in the past operation process. Since these operational data are actually collected, the reliability is higher. Firstly, setting a first time sequence number and a second time sequence number, wherein the first time sequence number can be understood as the data length input by the model, and the second time sequence number is the data length output by the model.
The method comprises the steps of setting the first time sequence number as 60 and the second time sequence number as 10, taking a certain acquisition time corresponding to equipment operation data as a reference time, selecting 59 pieces of equipment operation data corresponding to the acquisition time from the reference time forward, integrating the 59 pieces of equipment operation data and the data corresponding to the reference time into model input data, wherein the model input data comprises data of 60 time points of each type of equipment operation data, and selecting a limit range of the 10 pieces of equipment operation data from the reference time backward, wherein the limit range can be generated through digital twin model simulation.
In other preferred embodiments, the model input data and the model output data in the model training data use device operation data, namely, the device operation data corresponding to 59 acquisition moments is selected from the reference moment forward and integrated into the model input data together with the data corresponding to the reference moment, the model input data comprises data of 60 time points of each type of device operation data, and the device operation data of 10 acquisition moments is selected from the reference moment backward and used as the model output data. Training a long-term and short-term memory neural network through model input data and model output data, and then using a confidence interval of the output data of the model as a range limit value of corresponding type data.
Of course, the long-term and short-term memory neural network can be adjusted, probability distribution parameters corresponding to the output data, such as the mean value and the standard deviation, can be calculated through the probability distribution parameters, and the range limit value can be calculated. Still other output parameters may also calculate the range limit, which is not described in detail herein.
The neural network I obtained through training is stored in a data processing layer and can be called at any time, and meanwhile, the equipment operation data is continuously updated and the neural network I is required to be updated. After the data processing layer receives the equipment operation data, the data are extracted from the equipment operation data according to the first time sequence number, and are integrated and input into the first neural network, so that range limit values corresponding to various types of data at a plurality of future prediction time points can be obtained.
After the range limit value is obtained, the range limit value is compared with the standard data range of the corresponding data type, if the range limit value is within the standard data range, the operation data of each device cannot exceed the corresponding standard data range, early warning is not needed, and otherwise an early warning signal is generated to perform early warning. The process does not conflict with subsequent equipment failure pre-warning.
It should be noted that, when the range threshold values corresponding to the various types of data are all within the standard data range, only the data cannot exceed the limit value, and the device cannot be specified to be normal at the corresponding prediction time point, so that the comprehensive evaluation result of the operation data of each device can be used to represent the device state.
When the range limit value of each type of data at a plurality of future prediction time points is obtained through the first prediction of the neural network, the predicted range limit value is compared with the standard data range of the corresponding type of data, if the range of the standard data is exceeded, an early warning signal is generated for early warning, otherwise, the predicted range of the range limit value is used for early warning of equipment faults.
And each prediction time point is associated with a range limit value of various types of data, and the range limit value of each prediction time point needs to be analyzed through a neural network II to judge the equipment fault probability. Before the determination, the prediction data set needs to be determined according to the range limit value of each prediction time point, and the following steps can be referred to:
Sequentially extracting a plurality of range limits of a predicted time point, and integrating the range limits into a target data set, wherein the number of the range limits in the target data set is equal to the type of the equipment operation data;
And combining the range limit values of each type in the target data set to obtain a prediction data set.
The data type or type data in the present embodiment is to distinguish to which device operation data the range limit corresponds, and thus the data type or type data refers to one of temperature, humidity, pressure, vibration. Each prediction time point corresponds to a range limit value (the range limit value comprises an upper limit value and a lower limit value) of various types of data, and the range limit value corresponding to each prediction time point is integrated into a target data set.
Referring to fig. 3, the range limits of each type in the target data set are next set as a limit box, and each limit box includes a pair of data, i.e., an upper limit and a lower limit of one type of data. And then extracting one piece of data from each limit value box in the random to complete combination, generating one piece of prediction data, obtaining a plurality of pieces of prediction data through combination, and integrating the plurality of pieces of prediction data to generate a prediction data set. It should be noted that, the running data of the device does not have a sequence, if the limit values of the various types of data in the two pieces of predicted data are the same and the sequence is different, the two pieces of predicted data are treated as one piece of predicted data, so that the predicted data do not have repeated predicted data in the predicted data set.
It is noted that after combination the prediction dataset contains only target data of the same directional extremum as well as target data of different directional extremum. The same-direction extreme value refers to that the equipment operation data all takes an upper limit value or a lower limit value, and the different-direction extreme value refers to that the equipment operation data part takes an upper limit value and the equipment operation data part takes a lower limit value.
The prediction data sets are in one-to-one correspondence with the prediction time points, and the prediction data sets of the prediction time points are input into the neural network II to obtain fault information corresponding to the prediction time points, namely fault types and fault probabilities. Of course, if the equipment does not have a fault at the predicted time point, the output fault information is a fault-free identification.
The fault probability in the second output fault information of the neural network refers to a probability value corresponding to the case that the equipment operation data is the prediction data set corresponding to the prediction time point, namely the equipment operation data of the equipment at the prediction time point is the prediction data set, and then the equipment is likely to have faults and the probability thereof. However, whether the device operation data of the device at the predicted time point is a predicted data set is not determined, but a certain probability exists as the predicted data set, and the probability is superimposed with the fault probability in the fault information to obtain the predicted fault occurrence probability.
Next, the technical scheme provided by the present embodiment is briefly described by way of example.
And the data acquisition layer is used for acquiring equipment operation data by installing intelligent sensors such as a temperature sensor, a humidity sensor, a vibration sensor, a pressure sensor and the like on equipment to be monitored. And correction models are built in the intelligent sensors, so that collected data can be automatically corrected, and the accuracy of the data is improved.
And the data processing layer is internally provided with a first neural network, a second neural network and other data processing algorithms, and is used for analyzing and processing the corrected equipment operation data so as to perform equipment fault and early warning.
During specific work, the intelligent sensor collects equipment operation data according to a set time interval, and performs pretreatment such as filtering and denoising on the data to ensure the data quality. And comparing the processed equipment operation data with a standard data range according to the data type, and generating an early warning signal to early warn if the data is abnormal. And if the early warning signal is not generated, the equipment operation data is sent to the data processing layer. Of course, in other preferred embodiments, the data acquisition layer may not have an early warning operation, and may directly perform subsequent fault prediction.
Training of neural network one:
1. Model training data. A number of time stamped equipment operational data, including temperature, pressure amplitude, vibration, humidity, etc., are collected from the historical operational data of the plant equipment. A first number of timings (e.g., 60) and a second number of timings (e.g., 10) of the model input data are determined. Based on the current time, the data point of the first time sequence number is selected forward as input, and the data point of the second time sequence number is selected backward as output (the output is the range limit value obtained by statistical analysis or simulation). The data is preprocessed, including operations such as normalization, denoising, abnormal value elimination and the like, so that the quality and consistency of the data are ensured. The data is divided into training, validation and test sets, typically in proportions of 70%, 15% and 15%, resulting in model training data.
2. And (5) constructing a model. A structure based on long-short-term memory (LSTM) neural network is constructed, and comprises an input layer, an LSTM layer, a full-connection layer and an output layer. The LSTM layer is composed of a plurality of LSTM cells, each cell including a forget gate, an input gate, and an output gate to capture long-term dependencies in time series data. Setting super parameters of the model, such as hidden layer size, learning rate, training period number and the like.
3. And training the constructed long-term and short-term memory neural network through simulation training data, and storing the trained neural network in a data processing layer after test and evaluation.
The construction and training of the second neural network can refer to the construction and training of the first neural network, and the difference is that the model structures used by the first neural network and the second neural network are different and the input data and the output data are different. The second neural network is mainly to identify whether the input data is similar to the equipment operation data corresponding to a certain fault type, and the essence of the second neural network is a classification problem, which can be realized by a support vector machine. And the input of the second neural network is the predicted limit value of the equipment operation data, the output is fault information, and the training is carried out through corresponding data during training.
Based on the first embodiment, the second embodiment calibrates the fault probability in the fault information according to the change probability of the equipment operation data within the limit value range.
The upper limit and the lower limit of the data change correspond to the limit range, and theoretically, the probability that the equipment operation data is the upper limit or the lower limit is relatively low, and the probability that the intermediate value of the upper limit and the lower limit is relatively high, so that the probability that the data contained in the upper limit and the lower limit corresponds to the normal distribution is met. If the upper limit value and the lower limit value are calculated according to the 95% confidence interval, the data representing 95% is between the upper limit value and the lower limit value, and the probability distribution between the upper limit value and the lower limit value can be obtained by combining the normal distribution standard model.
The output layer of the first neural network can be modified to enable the output device operation data to correspond to the mean value and standard deviation of the predicted value, a normal distribution model is constructed by utilizing the parameters, and the fault probability is calibrated through the normal distribution model. If the equipment operation data are mutually affected, the fault probability in the fault information can be calibrated by calculating the conditional probability.
Assuming that the input of the neural network is the upper limit value of temperature, humidity, pressure and vibration, probability values corresponding to the upper limit values can be determined through the constructed normal distribution, and the probability values are multiplied and then multiplied with the predicted fault probability, so that the occurrence probability of the fault type can be obtained.
The third embodiment is different from the third embodiment in that not only the upper limit value and the lower limit value of the equipment operation data are utilized to predict fault information, but also interpolation processing is carried out between the upper limit value and the lower limit value, so that more equipment operation data are obtained to predict the fault information under different data combinations, and more perfect fault early warning is carried out.
The main difference from the first embodiment is that the predicted data set is obtained in the following manner, referring to fig. 4:
Dividing range limit values of all types in a target data set according to set intervals to obtain a plurality of numerical value points, associating the numerical value points with corresponding data types, selecting one numerical value point from the numerical value points corresponding to each type of data, integrating all the selected numerical value points into one piece of prediction data, and integrating all the obtained prediction data into a prediction data set.
An embodiment of the second aspect of the present application provides a factory equipment fault early warning method based on the internet of things, including:
Acquiring a plurality of types of equipment operation data through an intelligent sensor arranged in a factory, judging whether the equipment operation data exceeds a standard data range, and generating an early warning signal if the equipment operation data exceeds the standard data range, wherein the equipment operation data refers to data related to the equipment operation state;
inputting a plurality of equipment operation data into a first neural network, and predicting to obtain range limit values of a plurality of predicted time points, wherein the range limit values comprise an upper limit value and a lower limit value;
And inputting the prediction data set into a second neural network to predict and obtain fault information of each time point.
The above embodiments are only for illustrating the technical method of the present application and not for limiting the same, and it should be understood by those skilled in the art that the technical method of the present application may be modified or substituted without departing from the spirit and scope of the technical method of the present application.