CN112528562B - Smart tactile system and monitoring method for structural health monitoring - Google Patents
Smart tactile system and monitoring method for structural health monitoringDownload PDFInfo
- Publication number
- CN112528562B CN112528562BCN202011419882.7ACN202011419882ACN112528562BCN 112528562 BCN112528562 BCN 112528562BCN 202011419882 ACN202011419882 ACN 202011419882ACN 112528562 BCN112528562 BCN 112528562B
- Authority
- CN
- China
- Prior art keywords
- layer
- strain field
- field information
- information
- strain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
Translated fromChinese
Description
Translated fromChinese技术领域technical field
本发明涉及结构健康监测技术领域，具体涉及一种用于结构健康监测的智能触觉系统和监测方法。The invention relates to the technical field of structural health monitoring, in particular to an intelligent tactile system and monitoring method for structural health monitoring.
背景技术Background technique
使用传感器制作表面附着式或埋入式传感器分布阵列进行信号收集，是获取结构响应的主要手段，也是进行结构健康监测的第一步。保证传感器网络的稳定性与可靠性成为了健康监测系统的重中之重。The use of sensors to make surface-attached or embedded sensor distribution arrays for signal collection is the main means of obtaining structural responses and the first step in structural health monitoring. Ensuring the stability and reliability of the sensor network has become the top priority of the health monitoring system.
目前的监测系统在传感器正常工作的情况下可以得到结构的健康信息，但是在实际的工程应用中，传感器系统都面临着恶劣的使用环境。例如桥梁监测、机翼监测等，是在极端的压力、极端的温度下工作的，极易导致传感器的失效。而在传感器部分失效的情况下，监测系统无法在原有的高精度水准上正常工作。因此，在传感器信息部分丢失的情况下，如何能快速、准确的反演出结构的信息，在结构健康监测领域是亟待解决的问题之一。The current monitoring system can obtain the health information of the structure when the sensor works normally, but in the actual engineering application, the sensor system is faced with the harsh environment. For example, bridge monitoring and wing monitoring work under extreme pressure and temperature, which can easily lead to sensor failure. In the case of partial failure of the sensor, the monitoring system cannot work normally at the original high precision level. Therefore, in the case of partial loss of sensor information, how to quickly and accurately retrieve structural information is one of the problems to be solved urgently in the field of structural health monitoring.
通常需要对丢失信息的点恢复信息后再进行结构反演，现有的恢复信息的技术手段是采用“矩阵补全”方法，而“矩阵补全”方法由于算法原因，本身存在两个问题：1、计算复杂度高；2、耗时长。因此，现有的监测系统在传感器部分失效的情况下不能快速、准确的反演出结构的健康信息，并且，现有的算法对计算机硬件性能有一定的要求。Usually, it is necessary to restore the information of the points where the information is lost and then perform structural inversion. The existing technical means of restoring information is to use the "matrix completion" method, and the "matrix completion" method itself has two problems due to the algorithm: 1. High computational complexity; 2. Time-consuming. Therefore, the existing monitoring system cannot quickly and accurately retrieve the health information of the structure when the sensor part fails, and the existing algorithm has certain requirements on the performance of the computer hardware.
发明内容Contents of the invention
有鉴于此，本发明的主要目的在于提供一种用于结构健康监测的智能触觉系统和方法，以期部分地解决上述技术问题中的至少之一。In view of this, the main purpose of the present invention is to provide an intelligent haptic system and method for structural health monitoring, in order to partially solve at least one of the above technical problems.
为了实现上述目的，作为本发明的一方面，提供了一种用于结构健康监测的智能触觉监测方法，包括以下步骤：In order to achieve the above object, as an aspect of the present invention, an intelligent tactile monitoring method for structural health monitoring is provided, comprising the following steps:
获取结构的应变场信息；Obtain the strain field information of the structure;
利用生成对抗网络进行应变场信息的补充；Use the generative confrontation network to supplement the strain field information;
将应变场信息输入卷积神经网络，进行结构反演；Input the strain field information into the convolutional neural network for structural inversion;
将应变场信息与结构参数在网页端进行可视化。Visualize the strain field information and structural parameters on the web page.
其中，获取外载作用下结构的应变场，将应变片贴在被测物体表面或埋入结构内部测量应变，获取结构的应变场信息。Among them, the strain field of the structure under the external load is obtained, and the strain gauge is attached to the surface of the measured object or embedded in the structure to measure the strain, and the strain field information of the structure is obtained.
其中，所述生成对抗网络包括生成器和判别器，所述生成器用于通过现有的部分的应变场信息，生成完整的应变场信息；所述判别器用于判断生成器生成的应变场信息是否合理及是否真实。Wherein, the generative confrontation network includes a generator and a discriminator, the generator is used to generate complete strain field information through the existing part of the strain field information; the discriminator is used to judge whether the strain field information generated by the generator is reasonable and true.
其中，所述生成器网络结构包括输入层、多个反卷积层、多个全连接层和自定义输出层；所述输入层用于将应变场信息输入生成器；反卷积层的卷积核大小为3；所述自定义输出层用于筛选出丢失应变信息的位置进行数据替换填充，将输入层的应变场信息更新为补充之后的应变场信息。Wherein, the generator network structure includes an input layer, multiple deconvolution layers, multiple fully connected layers, and a custom output layer; the input layer is used to input strain field information into the generator; the volume of the deconvolution layer The size of the product kernel is 3; the self-defined output layer is used to filter out the positions where the strain information is lost for data replacement and filling, and update the strain field information of the input layer to the supplemented strain field information.
其中，所述判别器网络结构包括输入层、多个卷积层、全连接层和输出层；所述输入层用于将生成器输出的应变场信息输入判别器；卷积层的卷积核大小为3；所述输出层对上一层的信息进行全连接层计算，并使用Sigmoid激活函数，所述Sigmoid激活函数将数据映射到(0，1)区间上，使输出数据具有概率的含义，判断生成器生成的应变场信息是否合理及是否真实。Wherein, the discriminator network structure includes an input layer, a plurality of convolutional layers, a fully connected layer and an output layer; the input layer is used to input the strain field information output by the generator into the discriminator; the convolution kernel of the convolutional layer The size is 3; the output layer performs full-connection layer calculation on the information of the previous layer, and uses the Sigmoid activation function, which maps the data to the (0,1) interval, so that the output data has the meaning of probability , to judge whether the strain field information generated by the generator is reasonable and true.
其中，所述卷积神经网络的网络结构包括输入层、卷积层、正则化层、全连接层、Dropout层和输出层；所述输入层用于将应变场信息输入卷积神经网络；所述卷积层用于对上一层进行卷积计算，提取信息。Wherein, the network structure of the convolutional neural network includes an input layer, a convolutional layer, a regularization layer, a fully connected layer, a Dropout layer and an output layer; the input layer is used to input the strain field information into the convolutional neural network; The above convolutional layer is used to perform convolution calculation on the upper layer to extract information.
其中，所述卷积层的卷积核大小为3；所述正则化层用于对上一层进行正则化计算，防止过拟合；所述输出层用于对上一层的信息进行全连接层计算，并使用Sigmoid激活函数，所述Sigmoid激活函数用于将数据映射到(0，1)区间上，使输出数据具有概率的含义，进而与结构正确的数字化表述方式，即0/1表述方式进行比对，最小化两者的差距。Wherein, the convolution kernel size of the convolution layer is 3; the regularization layer is used to perform regularization calculation on the upper layer to prevent over-fitting; the output layer is used to perform full The connection layer is calculated and uses the Sigmoid activation function, which is used to map the data to the (0, 1) interval, so that the output data has a probability meaning, and then has a correct digital representation with the structure, that is, 0/1 Expressions are compared to minimize the gap between the two.
其中，将应变场信息与结构参数在Flask框架搭建的网页端进行可视化的步骤中，具体包括：将训练完成的生成对抗网络的生成器以及卷积神经网络上传至服务器运行，使用Flask框架搭建API，便于实际应用中在终端进行调用，降低对设备的计算性能要求。Among them, the step of visualizing the strain field information and structural parameters on the webpage built by the Flask framework includes: uploading the trained generator of the generated confrontation network and the convolutional neural network to the server for operation, and using the Flask framework to build the API , which is convenient for invoking on the terminal in practical applications, and reduces the computing performance requirements of the device.
其中，所述将应变场信息与结构参数在Flask框架搭建的网页端进行可视化具体使用方式为：将应变场信息的数据按照指定排列方式保存为*.TXT或*.MAT文件，上传至服务器，服务器即可将结构反演的结果返回至终端，并进行结果可视化。Wherein, the specific use method of visualizing the strain field information and structural parameters on the webpage built by the Flask framework is: save the data of the strain field information as a *.TXT or *.MAT file according to the specified arrangement, upload it to the server, The server can then return the results of the structural inversion to the terminal and visualize the results.
作为本发明的另一方面，提供了一种用于结构健康监测的智能触觉系统，包括信息采集模块、信息扩充模块、结构预测模块和云端部署系统；其中，As another aspect of the present invention, an intelligent tactile system for structural health monitoring is provided, including an information collection module, an information expansion module, a structure prediction module and a cloud deployment system; wherein,
信息采集模块，用于获取结构的应变场信息；An information collection module, used to obtain the strain field information of the structure;
信息扩充模块，用于利用生成对抗网络进行应变场信息的补充；The information expansion module is used to supplement the strain field information by using the generated confrontation network;
结构预测模块，用于将应变场信息输入卷积神经网络，进行结构反演；The structure prediction module is used to input the strain field information into the convolutional neural network for structure inversion;
云端部署系统，用于将应变场信息与结构参数在网页端进行可视化。The cloud deployment system is used to visualize the strain field information and structural parameters on the web page.
基于上述技术方案可知，本发明的用于结构健康监测的智能触觉系统和监测方法相对于现有技术至少具有如下有益效果之一或其中的一部分：Based on the above technical solutions, it can be known that the intelligent tactile system and monitoring method for structural health monitoring of the present invention have at least one or part of the following beneficial effects compared with the prior art:
1、将生成对抗网络应用于应变场信息补全，通过本发明的方法，在实践中可以补全已丢失50％信息的应变场并且将平均相对误差控制在10％以内。该算法可以使监测系统在传感器部分无效的情况下依然能正常工作；1. Applying the generative adversarial network to the strain field information completion, through the method of the present invention, the strain field that has lost 50% of the information can be completed in practice and the average relative error can be controlled within 10%. This algorithm can make the monitoring system still work normally when the sensor part is invalid;
2、本发明通过卷积神经网络建立应变场与结构参数之间的关系，在使用过程中不需要知道任何先验物理知识，减少了应变场信息补全所需要的已知条件，同时，提高了应变场信息补全的效率，减少了时间消耗；2. The present invention establishes the relationship between the strain field and the structural parameters through the convolutional neural network, and does not need to know any prior physical knowledge during use, which reduces the known conditions required for the completion of the strain field information, and at the same time, improves Improve the efficiency of strain field information completion and reduce time consumption;
3、本发明创新性地将数字超材料的概念与One-Hot编码相结合，在实践中通过应变场信息反演结构参数的准确率能够达到98％以上，保证了结构监测系统的准确性与可靠性；3. The present invention innovatively combines the concept of digital metamaterials with One-Hot coding. In practice, the accuracy rate of structural parameter inversion through strain field information can reach more than 98%, ensuring the accuracy and accuracy of the structural monitoring system. reliability;
4、本发明的主要算法均基于Google机器学习开源框架TensorFlow编写完成，由此带来的优点是所有的算法均可部署于云端，所有的计算都是在云端进行，降低了对结构健康监测终端的计算机硬件性能要求。4. The main algorithms of the present invention are all written based on the Google machine learning open source framework TensorFlow, which has the advantage that all algorithms can be deployed on the cloud, and all calculations are performed on the cloud, reducing the need for structural health monitoring terminals. computer hardware performance requirements.
附图说明Description of drawings
图1是本发明实施例提供的用于结构健康监测的智能触觉监测方法的流程图；Fig. 1 is a flowchart of an intelligent tactile monitoring method for structural health monitoring provided by an embodiment of the present invention;
图2是本发明实施例提供的用于结构健康监测的智能触觉系统的框架布局图；Fig. 2 is a frame layout diagram of an intelligent tactile system for structural health monitoring provided by an embodiment of the present invention;
图3是本发明实施例提供的核心算法流程图及云上部署方式；Fig. 3 is the flow chart of the core algorithm provided by the embodiment of the present invention and the deployment method on the cloud;
图4是本发明实施例提供的采集到的应变场信息；Fig. 4 is the collected strain field information provided by the embodiment of the present invention;
图5是本发明实施例提供的经过生成对抗网络补充之后的应变场信息；Fig. 5 is the strain field information provided by the embodiment of the present invention after being supplemented by the generated confrontation network;
图6是本发明实施例提供的经过卷积神经网络预测的结构参数图；Fig. 6 is a structural parameter diagram predicted by a convolutional neural network provided by an embodiment of the present invention;
图7是本发明实施例提供的Flask API接口示意图。Fig. 7 is a schematic diagram of the Flask API interface provided by the embodiment of the present invention.
具体实施方式Detailed ways
本发明通过使用生成对抗网络，补充应变场丢失的数据，丰富其携带的结构信息，为后续的结构预测反问题提供了必要条件The invention supplements the lost data of the strain field by using the generative confrontation network, enriches the structural information it carries, and provides the necessary conditions for the subsequent structure prediction inverse problem
此外，本发明提出了将卷积神经网络应用于结构反演。本发明引入数字超材料，在表示结构组成时可使用数字0或1来离散地表示，通过改变0/1的排布方式去表示不同的结构属性，由于独立的变量只有0/1，因此也降低了结构反演的难度。根据力学相关知识，应变场与结构之间存在着定性关系：应变大的地方材料较软，应变小的地方材料较硬。简单来说，可以将0/1理解为软/硬材料，因此，在宏观上通过观测结构在外载作用下应变场的值就可以推算出该处的材料属于软材料还是硬材料，也即该处结构的数字化表述方式为0还是1。同时，使用卷积神经网络建立应变场与结构之间的定量关系，如下述步骤三，将应变场作为输入，将结构的数字化表述方式作为输出，通过训练神经网络，进一步提高了计算精确度与效率。Furthermore, the present invention proposes the application of convolutional neural networks to structure inversion. The present invention introduces digital metamaterials, which can be discretely represented by
本发明还引入了One-Hot编码，One-Hot编码是一种在机器学习领域非常常用的编码手段，One-Hot编码是将类别变量转换为机器学习算法易于利用的一种形式的过程。The present invention also introduces One-Hot coding, which is a very common coding method in the field of machine learning. One-Hot coding is a process of converting category variables into a form that is easy to use by machine learning algorithms.
更重要的，本发明充分利用TensorFlow框架的优点，结合Flask框架搭建RESTAPI，直接将本发明涉及到的所有算法部署到云端，便于调用。More importantly, the present invention makes full use of the advantages of the TensorFlow framework, combines the Flask framework to build a REST API, and directly deploys all the algorithms involved in the present invention to the cloud for easy invocation.
需要说明的是，触觉系统在这里更像是一种拟人化的名称。本发明使用应变传感器获取结构的应变，进而使用神经网络对应变信息进行扩充进而预测结构。神经网络使传感器的功能不再单一，而是具有了获取信息、处理信息、反馈信息的功能，因此，将整套系统类比于人类的触觉，是一套智能的系统。It should be noted that the haptic system is more of an anthropomorphic name here. The invention uses a strain sensor to obtain the strain of the structure, and then uses a neural network to expand the strain information to predict the structure. The neural network makes the function of the sensor no longer single, but has the functions of acquiring information, processing information, and feeding back information. Therefore, it is an intelligent system that compares the entire system to the human sense of touch.
为使本发明的目的、技术方案和优点更加清楚明白，以下结合具体实施例，并参照附图，对本发明作进一步的详细说明。In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.
如图1所示，本发明公开了一种用于结构健康监测的智能触觉监测方法，包括以下步骤：As shown in Figure 1, the present invention discloses an intelligent tactile monitoring method for structural health monitoring, including the following steps:
步骤1、获取结构的应变场信息；其中，获取外载作用下结构的应变场，将应变片贴在被测物体表面或埋入结构内部测量应变，获取结构的应变场信息。Step 1. Obtain the strain field information of the structure; wherein, obtain the strain field of the structure under the action of external load, stick the strain gauge on the surface of the object to be measured or embed it in the structure to measure the strain, and obtain the strain field information of the structure.
步骤2、利用生成对抗网络进行应变场信息的补充；其中，所述生成对抗网络包括生成器和判别器，所述生成器用于通过现有的部分的应变场信息，生成完整的应变场信息；所述判别器用于判断生成器生成的应变场信息是否合理及是否真实。所述生成器网络结构包括输入层、多个反卷积层、多个全连接层和自定义输出层；所述输入层用于将应变场信息输入生成器；反卷积层的卷积核大小为3；所述自定义输出层用于筛选出丢失应变信息的位置进行数据替换填充，将输入层的应变场信息更新为补充之后的应变场信息。所述判别器网络结构包括输入层、多个卷积层、全连接层和输出层；所述输入层用于将生成器输出的应变场信息输入判别器；卷积层的卷积核大小为3；所述输出层对上一层的信息进行全连接层计算，并使用Sigmoid激活函数，所述Sigmoid激活函数将数据映射到(0，1)区间上，使输出数据具有概率的含义，判断生成器生成的应变场信息是否合理及是否真实。
步骤3、将应变场信息输入卷积神经网络，进行结构反演；其中，所述卷积神经网络的网络结构包括输入层、卷积层、正则化层、全连接层、Dropout层和输出层；所述输入层用于将应变场信息输入卷积神经网络；所述卷积层用于对上一层进行卷积计算，提取信息。所述卷积层的卷积核大小为3；所述正则化层用于对上一层进行正则化计算，防止过拟合；所述输出层用于对上一层的信息进行全连接层计算，并使用Sigmoid激活函数，所述Sigmoid激活函数用于将数据映射到(0，1)区间上，使输出数据具有概率的含义，进而与结构正确的数字化表述方式，即0/1表述方式进行比对，最小化两者的差距。Step 3, input the strain field information into the convolutional neural network, and perform structural inversion; wherein, the network structure of the convolutional neural network includes an input layer, a convolutional layer, a regularization layer, a fully connected layer, a dropout layer and an output layer ; The input layer is used to input the strain field information into the convolutional neural network; the convolutional layer is used to perform convolution calculation on the upper layer to extract information. The convolution kernel size of the convolutional layer is 3; the regularization layer is used to perform regularized calculations on the previous layer to prevent overfitting; the output layer is used to perform a fully connected layer on the information of the previous layer Calculate and use the Sigmoid activation function, the Sigmoid activation function is used to map the data to the (0, 1) interval, so that the output data has the meaning of probability, and then with the correct digital representation, that is, the 0/1 representation Compare and minimize the gap between the two.
步骤4、将应变场信息与结构参数在网页端进行可视化。其中，将应变场信息与结构参数在Flask框架搭建的网页端进行可视化的步骤中，具体包括：将训练完成的生成对抗网络的生成器以及卷积神经网络上传至服务器运行，使用Flask框架搭建API，便于实际应用中在终端进行调用，降低对设备的计算性能要求。所述将应变场信息与结构参数在Flask框架搭建的网页端进行可视化具体使用方式为：将应变场信息的数据按照指定排列方式保存为*.TXT或*.MAT文件，上传至服务器，服务器即可将结构反演的结果返回至终端，并进行结果可视化。
本发明还公开了一种用于结构健康监测的智能触觉系统，包括信息采集模块、信息扩充模块、结构预测模块和云端部署系统；其中，The invention also discloses an intelligent tactile system for structural health monitoring, including an information collection module, an information expansion module, a structure prediction module and a cloud deployment system; wherein,
信息采集模块，用于获取结构的应变场信息；An information collection module, used to obtain the strain field information of the structure;
信息扩充模块，用于利用生成对抗网络进行应变场信息的补充；The information expansion module is used to supplement the strain field information by using the generated confrontation network;
结构预测模块，用于将应变场信息输入卷积神经网络，进行结构反演；The structure prediction module is used to input the strain field information into the convolutional neural network for structure inversion;
云端部署系统，用于将应变场信息与结构参数在网页端进行可视化。The cloud deployment system is used to visualize the strain field information and structural parameters on the web page.
如图2所示，为用于结构健康监测的智能触觉系统的框架布局图；包括：As shown in Figure 2, it is a frame layout diagram of an intelligent tactile system for structural health monitoring; including:
1、信息采集模块。1. Information collection module.
获取外载作用下结构的应变场，将应变片贴在被测物体表面或埋入结构内部测量应变，获取结构的应变场信息；Obtain the strain field of the structure under external load, stick the strain gauge on the surface of the measured object or embed it inside the structure to measure the strain, and obtain the strain field information of the structure;
2、生成对抗网络，用于补充应变场信息。2. Generate an adversarial network to supplement the strain field information.
由于传感器损坏导致信息丢失，为了继续保持高精度的计算，需要将丢失的信息通过算法“补充”回来。是一种通过软件技术来弥补硬件缺陷的思路。Due to the loss of information due to sensor damage, in order to continue to maintain high-precision calculations, it is necessary to "supplement" the lost information through algorithms. It is a way of making up for hardware defects through software technology.
在使用有限元计算出的应变场进行测试时，本发明可以将丢失至多50％数据量的应变场还原，且误差不高于10％。50％为理想情况下的极限值。When the strain field calculated by the finite element is used for testing, the present invention can restore the strain field that loses at most 50% of the data, and the error is not higher than 10%. 50% is an ideal limit.
生成对抗网络包括两大结构：生成器和判别器。生成器的作用为通过现有的部分的应变场信息，生成一个完整的应变场信息。判别器的作用为判断生成器生成的应变场信息是否合理、是否真实。Generative confrontation network consists of two structures: generator and discriminator. The role of the generator is to generate a complete strain field information through the existing partial strain field information. The role of the discriminator is to judge whether the strain field information generated by the generator is reasonable and true.
其中，生成器网络结构为：Among them, the generator network structure is:
判别器网络结构为：The discriminator network structure is:
特别说明：生成器和判别器的卷积核大小均为3*3，该参数过大会降低神经网络的训练速度，过小会陷入局部特征导致网络失效。除特殊说明外均使用ReLU激活函数，该激活函数在实践中被证明是收敛速度较快并且能有效防止神经网络训练过拟合的一种激活函数。判别器最后一层使用Sigmoid激活函数，将数据映射到(0，1)区间上，使输出数据具有概率的含义，才能判断生成器生成的应变场信息是否合理、是否真实。Special Note: The convolution kernel size of the generator and the discriminator are both 3*3. If this parameter is too large, it will reduce the training speed of the neural network. If it is too small, it will fall into local features and cause the network to fail. Unless otherwise specified, the ReLU activation function is used. In practice, this activation function has been proved to be an activation function that has a faster convergence speed and can effectively prevent neural network training from overfitting. The last layer of the discriminator uses the Sigmoid activation function to map the data to the (0, 1) interval, so that the output data has the meaning of probability, in order to judge whether the strain field information generated by the generator is reasonable and true.
3、卷积神经网络，用于从应变场反演结构信息。3. Convolutional neural network for inversion of structural information from strain fields.
本发明中的卷积神经网络可以根据输入的应变场反演出结构的信息。The convolutional neural network in the present invention can invert the information of the structure according to the input strain field.
其网络结构为：Its network structure is:
卷积核大小均为3*3，使用ReLU激活函数，其中最后一层使用Sigmoid激活函数，将数据映射到(0，1)区间上，使输出数据具有概率的含义，进而与结构正确的数字化表述方式，即0/1表述方式进行比对，最小化两者的差距。The size of the convolution kernel is 3*3, and the ReLU activation function is used. The last layer uses the Sigmoid activation function to map the data to the (0, 1) interval, so that the output data has a probabilistic meaning, and then digitized with the correct structure The expression method, that is, the 0/1 expression method is compared to minimize the gap between the two.
4、云端部署系统。4. Cloud deployment system.
如图3所示，为本发明采用的核心算法及云上部署方式；As shown in Figure 3, it is the core algorithm adopted by the present invention and the deployment method on the cloud;
一般来说，神经网络的模型需要在配置好相关环境的PC上运行。本发明本着方便、快捷的想法，打造一站式软件解决方案，在任意可以访问网页的终端(包括但不限于手机)，均可以使用本发明的所有功能。将训练完成的生成对抗网络的生成器以及卷积神经网络上传至服务器运行，使用Flask框架搭建API，便于实际应用中在终端进行调用，降低对设备的计算性能要求。具体使用方式为：将应变场信息的数据按照指定排列方式保存为*.TXT或*.MAT文件，上传至服务器，服务器即可将结构反演的结果返回至终端，并进行结果可视化。Generally speaking, the model of the neural network needs to be run on a PC with the relevant environment configured. Based on the idea of convenience and quickness, the present invention creates a one-stop software solution, and all the functions of the present invention can be used on any terminal (including but not limited to mobile phones) that can access webpages. Upload the trained GAN generator and convolutional neural network to the server for operation, and use the Flask framework to build the API, which is convenient for calling on the terminal in practical applications and reduces the computing performance requirements for the device. The specific usage method is: save the data of the strain field information as a *.TXT or *.MAT file according to the specified arrangement, upload it to the server, and the server can return the result of the structural inversion to the terminal and visualize the result.
实施例Example
将应变场对应的*.MAT文件或*.TXT文件通过Flask API接口上传至服务器，服务器调用训练好的卷积神经网络对应变场信息进行处理，即可返回补充信息之后的应变场并对其进行结构参数的预测并将结果在网页端进行可视化。Upload the *.MAT file or *.TXT file corresponding to the strain field to the server through the Flask API interface, and the server calls the trained convolutional neural network to process the strain field information, and returns the strain field after the supplementary information and compares it Predict structural parameters and visualize the results on the web page.
通过数据采集系统的传感器获取到如图4所示的应变场，图中，白色区域表示由于传感器失效导致的丢失应变信息的区域；如图5所示为经过生成对抗网络补充之后的应变场；如图6所示为经过卷积神经网络预测的结构参数图，其中深灰色代表硬材料(1)，浅灰色代表软材料(0)；如图7所示为Flask框架搭建API接口的示意图。The strain field shown in Figure 4 is obtained through the sensor of the data acquisition system. In the figure, the white area indicates the area where the strain information is lost due to sensor failure; Figure 5 shows the strain field after supplementation by the generated confrontation network; Figure 6 shows the structural parameter map predicted by the convolutional neural network, where dark gray represents hard material (1), and light gray represents soft material (0); Figure 7 shows a schematic diagram of the API interface built by the Flask framework.
以上所述的具体实施例，对本发明的目的、技术方案和有益效果进行了进一步详细说明，应理解的是，以上所述仅为本发明的具体实施例而已，并不用于限制本发明，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention, and are not intended to limit the present invention. Within the spirit and principles of the present invention, any modifications, equivalent replacements, improvements, etc., shall be included in the protection scope of the present invention.
Claims (10)
Translated fromChinesePriority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011419882.7ACN112528562B (en) | 2020-12-07 | 2020-12-07 | Smart tactile system and monitoring method for structural health monitoring |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011419882.7ACN112528562B (en) | 2020-12-07 | 2020-12-07 | Smart tactile system and monitoring method for structural health monitoring |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112528562A CN112528562A (en) | 2021-03-19 |
| CN112528562Btrue CN112528562B (en) | 2022-11-15 |
Family
ID=74997997
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011419882.7AActiveCN112528562B (en) | 2020-12-07 | 2020-12-07 | Smart tactile system and monitoring method for structural health monitoring |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112528562B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918831A (en)* | 2019-03-21 | 2019-06-21 | 大连理工大学 | A Global Prediction Method of Local Strain Based on Prediction Model |
| CN111241725A (en)* | 2019-12-30 | 2020-06-05 | 浙江大学 | A Conditional Generative Adversarial Network-Based Structural Response Reconstruction Method |
| CN111241614A (en)* | 2019-12-30 | 2020-06-05 | 浙江大学 | Engineering structure load inversion method based on condition generation confrontation network model |
| CN111679811A (en)* | 2020-05-26 | 2020-09-18 | 中国工商银行股份有限公司 | Web service construction method and device |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7571058B2 (en)* | 2006-05-09 | 2009-08-04 | Lockheed Martin Corporation | System to monitor the health of a structure, program product and related methods |
| CA2828576A1 (en)* | 2011-02-28 | 2012-09-07 | Critical Materials, Lda. | Structural health management system and method based on combined physical and simulated data |
| CN109446187B (en)* | 2018-10-16 | 2021-01-15 | 浙江大学 | Health status monitoring method of complex equipment based on attention mechanism and neural network |
| CN110728654B (en)* | 2019-09-06 | 2023-01-10 | 台州学院 | Automatic pipeline detection and classification method based on deep residual error neural network |
- 2020
- 2020-12-07CNCN202011419882.7Apatent/CN112528562B/enactiveActive
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918831A (en)* | 2019-03-21 | 2019-06-21 | 大连理工大学 | A Global Prediction Method of Local Strain Based on Prediction Model |
| CN111241725A (en)* | 2019-12-30 | 2020-06-05 | 浙江大学 | A Conditional Generative Adversarial Network-Based Structural Response Reconstruction Method |
| CN111241614A (en)* | 2019-12-30 | 2020-06-05 | 浙江大学 | Engineering structure load inversion method based on condition generation confrontation network model |
| CN111679811A (en)* | 2020-05-26 | 2020-09-18 | 中国工商银行股份有限公司 | Web service construction method and device |
Non-Patent Citations (2)
| Title |
|---|
| 卷积神经网络在桥梁结构健康监测系统中的应用;王珏辉;《电子测试》;20200131;52-53* |
| 面向飞行器结构的健康监控技术研究现状;张卫方 等;《航空制造技术》;20171001(第19期);38-47* |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112528562A (en) | 2021-03-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110442936A (en) | Equipment fault diagnosis method, apparatus and system based on the twin model of number | |
| CN111079333B (en) | Deep learning sensing method of flexible touch sensor | |
| CN116450486B (en) | Modeling method, device, equipment and medium for nodes in multi-element heterogeneous computing system | |
| CN115617634A (en) | 5G-based nuclear power equipment abnormality diagnosis method, device and computer equipment | |
| CN111339724B (en) | Method, apparatus and storage medium for generating data processing model and layout | |
| CN118052114B (en) | Missile digital twin modeling method based on calculation and measurement fusion | |
| US20220318591A1 (en) | Inference method and information processing apparatus | |
| CN117194163A (en) | Computer equipment, fault detection system, method and readable storage medium | |
| CN114638048A (en) | Three-dimensional spray pipe flow field rapid prediction and sensitivity parameter analysis method and device | |
| CN115380214A (en) | Fault warning system | |
| CN104898089B (en) | A kind of passive type localization method based on spatial migration compressed sensing | |
| CN112861374B (en) | Multi-physical coupling simulation processing method, device and equipment based on pre-controller | |
| CN116453633A (en) | Stress or strain field reconstruction method and device based on physical perception neural network | |
| CN112528562B (en) | Smart tactile system and monitoring method for structural health monitoring | |
| CN116205155B (en) | Push plate kiln flow field data prediction method and device based on digital twin | |
| CN119886266A (en) | Ultrasonic multi-mode pre-training method and device, computer equipment and storage medium | |
| CN116714437A (en) | Hydrogen fuel cell automobile safety monitoring system and monitoring method based on big data | |
| Zhang et al. | On-Device Crack Segmentation for Edge Structural Health Monitoring | |
| CN118171464B (en) | A management method for digital twin feature layers | |
| CN114494109A (en) | Techniques for generating subjective style comparison metrics for B-REP of 3D CAD objects | |
| CN117011805B (en) | Data exception evaluation method, device, equipment and readable storage medium | |
| CN114299001A (en) | Method and device for detecting different-source image change based on structural cycle consistency | |
| Liu et al. | Resolution enhancement of tongue tactile image based on deconvolution neural network | |
| US20230409780A1 (en) | Method and System for Determining Ply-by-Ply Damage in a Composite Structure | |
| CN118711781B (en) | Graded nursing management system and method for intensive care unit patients |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |