Method and system for evaluating health state of automatic verification system
Through convolutional neural networks and multi-layer long and short-term neural networks, the information loss problem caused by manual selection of features is solved, and the accurate health status evaluation and rapid response of the fault of the automated verification system are realized, and maintenance efficiency is improved.
Patent Information
- Application Number
- CN202411176411.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-12
AI Technical Summary
In the health status evaluation of automated verification systems, manual selection of features leads to loss or distortion of information, making it difficult to fully capture the feature information of complex nonlinear systems, affecting the accuracy of evaluation, and leading to insufficient or excessive maintenance.
Convolutional neural network and multi-layer long and short-term neural network are used to extract and analyze sensor data characteristics through convolutional layer, downsampling layer, full-connection layer and long-term memory network to generate health values to judge health status, helping operation and maintenance personnel in advance Lock the equipment situation that needs to be monitored to achieve rapid response to failures.
Through intelligent feature extraction and analysis, the accuracy of the health status evaluation of the automated verification system is improved, helping operation and maintenance personnel to identify potential problems in advance, and achieving more accurate fault response and maintenance optimization.
Smart Images

Figure CN118940123A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of equipment reliability analysis, and in particular relates to a method and system for evaluating the health status of an automated verification system. Background Art
[0002] The production line is an automated verification system, and each link of the equipment is a mechanical application with its own life cycle, application response mode, working production environment, etc. Its service life is often linked to factors such as the manufacturer, production batch, working environment, usage method, and work intensity. By collecting the corresponding information of each device in the automated verification system, integrating the production intensity, temperature, humidity, power consumption, and installation configuration and quality of the corresponding spare parts during the work process, production management can be optimized, and the operation status of the automated verification system can be monitored and evaluated in real time. Automated verification systems have been widely used in industrial production and power grid monitoring, so the performance evaluation of automated verification systems is crucial to ensure their accuracy and reliability.
[0003] Automated calibration system performance evaluation refers to the evaluation of the functions, accuracy, reliability, stability and other relevant indicators of the automated calibration system to obtain its health status. The evaluation of system performance can help calibration personnel understand the advantages and disadvantages of the system and provide a basis for system improvement and optimization.
[0004] With the development of science and technology, automated calibration systems are becoming increasingly intelligent and complex. In order to ensure the normal operation of automated calibration systems, traditional methods use various signal processing techniques (Fourier transform, wavelet transform, etc.) to extract features from a large amount of collected sensor data to evaluate the health status of the calibration system.
[0005] However, the above method relies on manual feature selection and may cause information loss or distortion during the feature extraction process. For complex nonlinear systems, it is unable to fully capture all feature information in a large amount of sensor data, thus affecting the evaluation of the health status of the automated calibration system and causing problems of insufficient or excessive daily maintenance, which makes it difficult to meet actual needs. Summary of the invention
[0006] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a method and system for evaluating the health status of an automated calibration system, which solves the accuracy problem caused by manual feature selection in the prior art, helps operation and maintenance personnel to lock in the equipment conditions that need to be monitored in advance, and accurately realizes the ability to respond quickly to faults.
[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:
[0008] A first aspect of the present invention provides a method for evaluating the health status of an automated verification system.
[0009] A method for evaluating the health status of an automated verification system, comprising:
[0010] Obtain the real-time data collected by each sensor of the automated verification system to be evaluated, and pre-process it into a data matrix;
[0011] Inputting the data matrix into the trained health assessment model to obtain the health status of the automated verification system;
[0012] Among them, the health assessment model uses a convolutional neural network to extract data features, inputs the extracted data features into a multi-layer long-short time neural network, obtains a health value, and makes a health status judgment based on the health value.
[0013] Furthermore, the preprocessing is specifically as follows:
[0014] Normalize the real-time data collected by each sensor to form a vector of 1*n dimension, where n is the number of sensors;
[0015] Use a sliding window of length L to resample the vector to obtain several sequences;
[0016] Combine the sequences into a (n-L+1)*L dimensional data matrix.
[0017] Furthermore, the convolutional neural network includes a convolutional layer Conv+ReLu, a downsampling layer MaxPooling, a fully connected layer FullyConnected+ReLU and a Softmax layer;
[0018] The convolution layer Conv+ReLu is used to increase the dimension;
[0019] The downsampling layer MaxPooling is used for downsampling;
[0020] The fully connected layer FullyConnected+ReLU is used to reduce the dimension;
[0021] The Softmax layer is used for normalization, and finally obtains the vector [x1 x2 … x T ], where T is the vector dimension.
[0022] Furthermore, the multi-layer long short-time neural network is composed of a plurality of VLSTMs and a fully connected layer FullyConnected+ReLU;
[0023] The VLSTM processes each dimension of the data features based on the mean and covariance to generate corresponding variables;
[0024] The fully connected layer FullyConnected+ReLU is used to fuse the variables corresponding to all dimensions to obtain the final health value.
[0025] Furthermore, the VLSTM is composed of a long short-term memory network LSTM and a VUE, and each VLSTM includes two parts in the time domain: encoding inference and decoding generation.
[0026] Furthermore, the coding inference estimates the mean and covariance for each dimension of the data feature and calculates the posterior probability distribution;
[0027] The decoding generation is to decode the posterior probability distribution, map it back to the original data space, and generate a variable for each dimensional data feature.
[0028] Furthermore, the health status judgment based on the health value is specifically as follows:
[0029] Damage status, health value RUL range is 0 <RUL<=0.2;
[0030] Sub-health state, the health value RUL range is 0.2 <RUL<=0.75;
[0031] Health status, the range of health value RUL is 0.75 <RUL<=1.0。
[0032] A second aspect of the present invention provides a system for evaluating the health status of an automated verification system.
[0033] A system for evaluating the health status of an automated verification system, comprising a data acquisition module and a health evaluation module:
[0034] The data acquisition module is configured to: acquire the real-time data currently collected by each sensor of the automated verification system to be evaluated, and pre-process it into a data matrix;
[0035] The health evaluation module is configured to: input the data matrix into the trained health evaluation model to obtain the health status of the automated verification system;
[0036] Among them, the health assessment model uses a convolutional neural network to extract data features, inputs the extracted data features into a multi-layer long-short time neural network, obtains a health value, and makes a health status judgment based on the health value.
[0037] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in a method for evaluating the health status of an automated verification system as described in the first aspect of the present invention.
[0038] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a method for evaluating the health status of an automated inspection system as described in the first aspect of the present invention are implemented.
[0039] One or more of the above technical solutions have the following beneficial effects:
[0040] The present invention adopts convolutional neural network to mine and learn sensor data, so as to intelligently extract feature information and solve the accuracy problem caused by manual feature selection in the prior art; and further analyzes and classifies the feature information through a multi-layer long short-term memory network LSTM and VUE, so as to help operation and maintenance personnel to lock in the equipment conditions that need to be monitored in advance and accurately realize the rapid response capability of faults.
[0041] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0043] Figure 1 This is a flow chart of the method of the first embodiment.
[0044] Figure 2 This is a structural diagram of the CNN of the first embodiment;
[0045] Figure 3 It is a structural diagram of the first embodiment MLSTM;
[0046] Figure 4 This is a structural diagram of the VLSTM of the first embodiment. DETAILED DESCRIPTION
[0047] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0048] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0049] Embodiment 1
[0050] In one embodiment of the present disclosure, a method for evaluating the health status of an automated verification system is provided. Figure 1 As shown, the following steps are included:
[0051] Step S1: obtaining the real-time data collected by each sensor of the automated verification system to be evaluated, and preprocessing it into a data matrix;
[0052] Furthermore, the preprocessing is specifically as follows:
[0053] Normalize the real-time data collected by each sensor to form a vector of 1*n dimension, where n is the number of sensors;
[0054] Use a sliding window of length L to resample the vector to obtain several sequences;
[0055] Combine the sequences into a (n-L+1)*L dimensional data matrix.
[0056] Step S2: inputting the data matrix into the trained health assessment model to obtain the health status of the automated verification system;
[0057] Among them, the health assessment model uses a convolutional neural network to extract data features, inputs the extracted data features into a multi-layer long-short time neural network, obtains a health value, and makes a health status judgment based on the health value.
[0058] Furthermore, the convolutional neural network includes a convolutional layer Conv+ReLu, a downsampling layer MaxPooling, a fully connected layer FullyConnected+ReLU and a Softmax layer;
[0059] The convolution layer Conv+ReLu is used to increase the dimension;
[0060] The downsampling layer MaxPooling is used for downsampling;
[0061] The fully connected layer FullyConnected+ReLU is used to reduce the dimension;
[0062] The Softmax layer is used for normalization, and finally obtains the vector [x1 x2 … x T ], where T is the vector dimension.
[0063] Furthermore, the multi-layer long short-time neural network is composed of a plurality of VLSTMs and a fully connected layer FullyConnected+ReLU;
[0064] The VLSTM processes each dimension of the data features based on the mean and covariance to generate corresponding variables;
[0065] The fully connected layer FullyConnected+ReLU is used to fuse the variables corresponding to all dimensions to obtain the final health value.
[0066] Furthermore, the VLSTM is composed of a long short-term memory network LSTM and a VUE, and each VLSTM includes two parts in the time domain: encoding inference and decoding generation.
[0067] Furthermore, the coding inference estimates the mean and covariance for each dimension of the data feature and calculates the posterior probability distribution;
[0068] The decoding generation is to decode the posterior probability distribution, map it back to the original data space, and generate a variable for each dimensional data feature.
[0069] Furthermore, the health status judgment based on the health value is specifically as follows:
[0070] Damage status, health value RUL range is 0 <RUL<=0.2;
[0071] Sub-health state, the range of health value RUL is 0.2 <RUL<=0.75;
[0072] Health status, the range of health value RUL is 0.75 <RUL<=1.0。
[0073] The following describes in detail the implementation process of a method for evaluating the health status of an automated verification system in this embodiment from the perspective of model construction and training.
[0074] 1. Obtain the historical data of each sensor in the automated calibration system and perform preprocessing.
[0075] Assume that there are n sensors in the automated verification system. Based on the historical data x of each sensor i (i=1,2,...,n) calculate the mean μ and standard deviation σ, and perform z-sort on the data to obtain the normalized data x i :
[0076]
[0077] After normalization, the n sensor data at the same time are combined into a 1*n dimension vector XI = {x1, x2, ...x n};
[0078] The vector XI is resampled using the sliding window technique to obtain a series of input sequences {x1,x2,...x L}, {x2,x3,...x L+1}, ... {x n-L+1 ,x n-L+2 ,...x n}, L is the length of the sliding window;
[0079] The above sequences are combined into a two-dimensional data matrix X of (n-L+1)*L dimensions. The X matrix serves as the input for the subsequent training of the health assessment model:
[0080]
[0081] 2. Build a health assessment model and use the preprocessed historical data of each sensor for training.
[0082] In this embodiment, a convolutional neural network CNN and a multi-layer long short-term neural network MLSTM composed of LSTM and VUE are used to construct a health assessment model.
[0083] The CNN network, such as Figure 2 As shown in the figure, it includes a convolution layer Conv+ReLu, a downsampling layer MaxPooling, a fully connected layer FullyConnected+ReLU and a Softmax layer, which are used to extract data features from a two-dimensional data matrix X.
[0084] Specifically, the two-dimensional data matrix X is input into the CNN network. First, the dimension of the original two-dimensional data matrix X is increased to three dimensions through the convolution layer. Then, the data matrix is downsampled through the downsampling layer. Then, the downsampled data matrix is converted into a 1*1*T vector through the fully connected layer. Finally, the 1*1*T vector output by the fully connected layer is normalized to a vector T = [x1 x2 … x T ].
[0085] Vector T represents the sensor data characteristics at that moment. The vector T is then input into the MLSTM network to finally obtain the health value of the automated calibration system, and the health status is judged based on the health value.
[0086] The MLSTM network in this embodiment is as follows Figure 3As shown, it consists of m VLSTMs. Each VLSTM is composed of a long short-term memory network (LSTM) and a VUE. The structure of each VLSTM is as Figure 4 shown, mainly including two parts: encoding inference and decoding generation. The encoder estimates the mean μ i for each dimension x of the data features z and the covariance logσ z , and calculates the posterior probability distribution P(z|x i ). Then, in the decoder, z is decoded and mapped back to the original data space to obtain the variable r. After passing this variable r through the multi-layer VLSTM structure, and then through the fully connected layer, the predicted health value of the automated calibration system is obtained.
[0087] Based on U-Net, VUE incorporates the idea of variational autoencoders. By combining the advantages of U-Net and the variational inference mechanism of variational autoencoders, uncertainty and flexibility are introduced into the model, improving its performance. The functions of VUE are reflected in the following ways: In the encoder part, VUE encodes each dimension of the input data to generate the mean and covariance of the latent variable z, and estimates the posterior distribution P(z|x i ) through variational inference. In the decoder part, VUE uses the value sampled from the latent variable z and maps it back to the original data space through the decoding layer of the U-Net structure. Figure 4 In , VUE is mainly reflected through the structures of the encoder and decoder. Among them, the output of the VUE encoding part is directly connected to the decoding part, enabling features to be directly transferred from each layer of the encoder to the corresponding layer of the decoder, ensuring that more context information is used during the decoding process, thereby improving the prediction accuracy.
[0088] The historical data of each preprocessed sensor is used for training, including:
[0089] The historical data of each sensor is normalized using z-score to obtain the mean and standard deviation of the historical data of each sensor, and the mean A and standard deviation B are saved for the preprocessing of subsequent real-time data.
[0090] The sliding window method is used to process the sensor historical data sequence, and expert knowledge and experience are used to establish labels for the health status of the historical data. The established RUL value range in the health status label is 0 - 1, and the health status is divided into three categories, namely: damaged state (0 < RUL < 0.2), sub-healthy state (0.2 < RUL < 0.75), and healthy state (0.75 < RUL < 1.0).
[0091] The sensor historical data sequence is divided into a training set and a validation set in a ratio of 7:3, the training set is output to the health assessment model, and the model is trained using a stochastic gradient descent method.
[0092] After the initial model training is completed, the model is verified using the validation set. When the model accuracy on the validation set reaches 93%, the model training is considered complete.
[0093] 3. The real-time data collected by the current sensors of the automated verification system are input into the trained health assessment model to obtain health value data.
[0094] After the health assessment model is trained, the real-time data of the sensor is collected, and the real-time data is z-score normalized using the mean A and standard deviation B during the training process; the data of each sensor at L moments are selected to form the model input matrix X, and the matrix X is input into the trained model to obtain the RUL value of the automated calibration equipment. The RUL value is between 0 and 1, and the health status is judged based on the health value.
[0095] Embodiment 2
[0096] In one embodiment of the present disclosure, a system for evaluating the health status of an automated verification system is provided, including a data acquisition module and a health evaluation module:
[0097] The data acquisition module is configured to: acquire the real-time data currently collected by each sensor of the automated verification system to be evaluated, and pre-process it into a data matrix;
[0098] The health evaluation module is configured to: input the data matrix into the trained health evaluation model to obtain the health status of the automated verification system;
[0099] Among them, the health assessment model uses a convolutional neural network to extract data features, inputs the extracted data features into a multi-layer long-short time neural network, obtains a health value, and makes a health status judgment based on the health value.
[0100] Embodiment 3
[0101] The purpose of this embodiment is to provide a computer-readable storage medium.
[0102] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in a method for evaluating the health status of an automated verification system as described in the first embodiment of the present disclosure.
[0103] Embodiment 4
[0104] The purpose of this embodiment is to provide an electronic device.
[0105] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a method for evaluating the health status of an automated verification system as described in the first embodiment of the present disclosure are implemented.
[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for evaluating the health status of an automated verification system, characterized in that: include: Obtain the real-time data collected by each sensor of the automated verification system to be evaluated, and pre-process it into a data matrix; Inputting the data matrix into the trained health assessment model to obtain the health status of the automated verification system; Among them, the health assessment model uses a convolutional neural network to extract data features, inputs the extracted data features into a multi-layer long-short time neural network, obtains a health value, and makes a health status judgment based on the health value.
2. A method for evaluating the health status of an automated verification system as claimed in claim 1, characterized in that: The preprocessing is specifically as follows: Normalize the real-time data collected by each sensor to form a vector of 1*n dimension, where n is the number of sensors; Use a sliding window of length L to resample the vector to obtain several sequences; Combine the sequences into a (n-L+1)*L dimensional data matrix.
3. A method for evaluating the health status of an automated verification system as claimed in claim 1, characterized in that: The convolutional neural network includes a convolutional layer Conv+ReLu, a downsampling layer MaxPool ing, a fully connected layer FullyConnected+ReLU and a Softmax layer; The convolution layer Conv+ReLu is used to increase the dimension; The downsampling layer MaxPool ing is used for downsampling; The fully connected layer FullyConnected+ReLU is used to reduce the dimension; The Softmax layer is used for normalization, and finally obtains the vector [x1 x2…x T ], where T is the vector dimension.
4. A method for evaluating the health status of an automated verification system as claimed in claim 1, characterized in that: The multi-layer long short-time neural network is composed of several VLSTMs and a fully connected layer FullyConnected+ReLU; The VLSTM processes each dimension of the data features based on the mean and covariance to generate corresponding variables; The fully connected layer FullyConnected+ReLU is used to fuse the variables corresponding to all dimensions to obtain the final health value.
5. A method for evaluating the health status of an automated verification system as claimed in claim 4, characterized in that: The VLSTM is composed of a long short-term memory network LSTM and a VUE, and each VLSTM includes two parts in the time domain: encoding inference and decoding generation.
6. A method for evaluating the health status of an automated verification system as claimed in claim 5, characterized in that: The coding inference is to estimate the mean and covariance of each dimension of the data feature and calculate the posterior probability distribution; The decoding generation is to decode the posterior probability distribution, map it back to the original data space, and generate a variable for each dimensional data feature.
7. A method for evaluating the health status of an automated verification system as claimed in claim 1, characterized in that: The health status judgment based on the health value is specifically as follows: Damage status, health value RUL range is 0 <RUL<=0.2; Sub-health state, the range of health value RUL is 0.2 <RUL<=0.75; Health status, the range of health value RUL is 0.75 <RUL<=1.0。 8. A system for evaluating the health status of an automated verification system, characterized in that: Including data acquisition module and health assessment module: The data acquisition module is configured to: acquire the real-time data currently collected by each sensor of the automated verification system to be evaluated, and pre-process it into a data matrix; The health evaluation module is configured to: input the data matrix into the trained health evaluation model to obtain the health status of the automated verification system; Among them, the health assessment model uses a convolutional neural network to extract data features, inputs the extracted data features into a multi-layer long-short time neural network, obtains a health value, and makes a health status judgment based on the health value.
9. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; a processor for executing the computer-readable instructions; Wherein, when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 7 is executed.
10. A storage medium, characterized in that: The computer-readable instructions are non-transitorily stored, wherein when the computer-readable instructions are executed by a computer, the method of any one of claims 1 to 7 is performed.
Citation Information
Cited By
Test bed pneumatic valve state intelligent detection method and system
CN121521484A
