Fatigue test loading system state monitoring method and system based on digital twinning

Abstract

The invention relates to a fatigue test loading system state monitoring method and system based on digital twinning, belongs to the technical field of industrial equipment state monitoring, and solves the problems that a traditional method is single in monitoring means, low in informatization degree, poor in real-time performance and the like. The method comprises the following steps: firstly, building a physical space of a fatigue test loading system, deploying a loading system parameter sensor to collect parameters of the fatigue test loading system, then building a digital space of the fatigue test loading system, preprocessing data and dividing a data set, finding out front and back correlation characteristics of time series data by utilizing a log-LSTM model, and finally obtaining a fatigue test result. The feature extraction of the operation data of the fatigue test loading system is realized, the current state is effectively judged based on the time sequence feature, and the abnormal state classification result of the fatigue test loading system is finally output, so that the abnormal states of multiple components of the fatigue test loading system can be found in time, the stable operation of the test process is ensured, and the test efficiency is improved. And economic loss and research and development interruption caused by unnecessary shutdown are avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial equipment status monitoring, and specifically relates to a fatigue test loading system status monitoring method and system based on digital twins. Background Art

[0002] In modern industrial production and scientific experiments, fatigue testing equipment, as one of the key equipment, is widely used in the fields of material performance testing, structural strength assessment, and new product development. One of the core components of fatigue testing equipment is the fatigue testing loading system, and the stability and accuracy of its performance are directly related to the reliability and effectiveness of the test results.

[0003] However, traditional monitoring methods for fatigue test loading systems have many shortcomings. First, the monitoring methods are often relatively simple, mainly relying on direct measurements of force sensors and displacement sensors, and lack a comprehensive evaluation of the overall performance of the loading system. Secondly, the degree of informatization of monitoring data is not high, and the data collection, processing and analysis processes are relatively cumbersome, making it difficult to achieve real-time monitoring and intelligent early warning. Furthermore, existing monitoring methods lack sufficient adaptability and robustness to changes in system performance caused by environmental factors, equipment aging, etc.

[0004] With the rapid development of intelligent manufacturing and artificial intelligence technology, higher requirements are placed on the condition monitoring method of fatigue test loading system. Therefore, a method that can comprehensively, real-time and accurately monitor the performance of fatigue test loading system is urgently needed to ensure the stable operation of fatigue testing machine, improve test efficiency, reduce maintenance costs, and provide more reliable technical support for research in fields such as materials science and structural engineering. Summary of the invention

[0005] The present invention aims to provide a fatigue test loading system state monitoring method and system based on digital twins, so as to overcome the shortcomings of traditional fatigue test loading system state monitoring methods, such as single monitoring means, low degree of informatization, poor real-time performance, adaptability and robustness, so as to realize comprehensive evaluation, real-time monitoring and intelligent early warning of the fatigue test loading system state, and provide strong guarantee for the intelligent upgrading of testing machines and innovative development in the industrial field.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A fatigue test loading system condition monitoring method based on digital twins, the method comprising the following steps:

[0008] Step 1: Building a physical space for a fatigue test loading system, wherein the physical space includes a fatigue test loading system, a loading system parameter sensor, a data bus, and a status data server. The loading system parameter sensor is deployed in the fatigue test loading system to collect current, voltage, power, temperature, pressure in the pump, vibration, and filtering capacity parameters in real time. The system parameters collected by the loading system parameter sensor are connected to the data bus through a local area network, and are aggregated in the data bus and transmitted to the status data server for storage;

[0009] Step 2: Building a fatigue test loading system digital space, wherein the fatigue test loading system digital space sequentially cleans, resamples and standardizes the data transmitted from the state data server, and divides the processed data into a training set and a test set;

[0010] Step 3: Determine the hyperparameters of the log-LSTM model, and train the training set based on the log-LSTM model to generate a state recognition model;

[0011] Step 4: After encapsulating the model structure and weight parameters of the state recognition model, package them into a state recognition web service, upload the current system parameters collected in real time by the loading system parameter sensor to the state recognition web service for state recognition, output the abnormal state classification result of the fatigue test loading system, and feed back the abnormal state classification result to the fatigue test loading system.

[0012] Accordingly, the present invention also proposes a fatigue test loading system state monitoring system based on digital twins, which includes a physical space subsystem and a digital space subsystem:

[0013] The physical space subsystem includes:

[0014] Fatigue test loading system, used to provide fatigue load loading when performing fatigue test tasks;

[0015] A loading system parameter sensor is deployed in the fatigue test loading system and is used to collect the current, voltage, power, temperature, pump pressure, vibration and filtering capacity parameters of the fatigue test loading system in real time;

[0016] A data bus module, used to connect the loading system parameter sensor via a local area network, summarize the system parameters collected by the loading system parameter sensor and transmit the summarized data to a status data server;

[0017] The state data server is used to store the data transmitted by the data bus module;

[0018] The digital space subsystem includes:

[0019] A data preprocessing module, used to sequentially perform data cleaning, resampling and standardization on the data transmitted from the status data server;

[0020] A partitioning module is used to divide the processed data into a training set and a test set;

[0021] Model design module, used to determine the hyperparameters of the log-LSTM model;

[0022] A model training module, used for training the training set based on the log-LSTM model to generate a state recognition model;

[0023] The state recognition module is used to encapsulate the model structure and weight parameters of the state recognition model into a state recognition web service, receive the current system parameters collected in real time by the loading system parameter sensor and perform state recognition, output the abnormal state classification result of the fatigue test loading system, and feed back the abnormal state classification result to the fatigue test loading system.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention provides a fatigue test loading system state monitoring method and system based on digital twin. A multi-source sensor network is deployed to collect parameters such as current, voltage, vibration, and pressure, a mapping between the physical space of the fatigue test loading system and the digital space of the fatigue test loading system is constructed, and a log-LSTM model is used to find the before-and-after correlation features of time series data, so as to realize feature extraction of fatigue test loading system operation data, effectively judge the current state based on the time series features, and finally output the abnormal state classification result of the fatigue test loading system, so as to timely discover the abnormal states of multiple components of the fatigue test loading system, ensure the stable operation of the test process, and avoid economic losses and R&D interruptions caused by unnecessary downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a fatigue test loading system state monitoring method based on digital twins according to one of the embodiments of the present invention;

[0027] Figure 2 It is a principle block diagram of a fatigue test loading system state monitoring system based on digital twin according to another embodiment of the present invention;

[0028] Figure 3 Confusion matrix diagram for the model classification effect on the test set. DETAILED DESCRIPTION

[0029] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments. It is understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0030] like Figure 1 As shown, this embodiment provides a fatigue test loading system state monitoring method based on digital twins, which includes the following steps 1 to 3.

[0031] Step 1: Build the physical space of the fatigue test loading system, which includes the fatigue test loading system, loading system parameter sensors, data bus and status data server.

[0032] Among them, the fatigue test loading system is used to provide fatigue load loading when performing fatigue test tasks. It is a test equipment that provides fatigue loading for fatigue testing of engineering materials. It provides periodic reciprocating motion through hydraulic transmission to perform fatigue loading on the test sample. During operation, there are status data that can be obtained by sensors, including but not limited to current data, voltage data, power data, temperature data, pump pressure data, vibration data, filtration capacity data, etc.

[0033] Step 1 specifically includes the following processes:

[0034] Step 11: Confirm the parameters to be collected.

[0035] Perform fault analysis on the fatigue test loading system, and determine the subsystems in the fatigue test loading system that are prone to abnormalities based on historical statistics, including motors, hydraulic substations, hydraulic pumps, etc., and then confirm the parameter data that needs to be collected, including: current, voltage, power, temperature, pump pressure, vibration, filtration capacity, etc.

[0036] Step 12: Deploy the sensor network and load system parameters.

[0037] The loading system parameter sensor is a general term for sensors used to collect various state parameters in real time during the operation of the fatigue test loading system, including intelligent circuit breakers, temperature sensors, pressure sensors, vibration sensors and intelligent filters, among which the intelligent circuit breaker is used to collect current, voltage and power parameters, the temperature sensor is used to collect temperature parameters, the pressure sensor is used to collect pressure parameters in the pump, the vibration sensor is used to collect vibration parameters, and the intelligent filter is used to collect filtering capacity parameters. The state data can be collected at different sampling frequencies. For example, the sampling frequency of the intelligent circuit breaker is 10Hz, the sampling frequency of the temperature sensor is 1Hz, the sampling frequency of the pressure sensor is 1Hz, the sampling frequency of the intelligent filter is 1Hz, and the sampling frequency of the vibration sensor is 100Hz. Among them, the intelligent circuit breaker can also be replaced by a combination of current sensors, voltage sensors and power sensors.

[0038] In this step, sensors related to the abnormal state of the fatigue test loading system are deployed in the fatigue test loading system to monitor the operating parameters of the fatigue test loading system in real time.

[0039] Step 13: Connect the loading system parameter sensor network to the data bus.

[0040] The data bus is a network structure that transfers data to a database. It is used to aggregate the data collected by all sensors and transmit the data through the TCP / IP protocol and store it in the status data server.

[0041] In this step, the sensor that collects data is connected to the data bus through the local area network. The system parameters collected by the loading system parameter sensor in real time are connected to the data bus through the local area network, summarized in the data bus, and different types of data are formatted for subsequent analysis.

[0042] Step 14: Data storage.

[0043] The status data server is an independent database server, which is generally built using MySQL database. It is used to store the operating status data of the fatigue test loading system collected by the sensor for subsequent analysis.

[0044] In this step, the data summarized in the data bus is transmitted to the status data server through the TCP / IP protocol. The status data server uses a relational database for data storage and also transmits the data to the constructed fatigue test loading system digital space.

[0045] Step 2: Build a digital space for the fatigue test loading system. The digital space for the fatigue test loading system is used to clean, resample and standardize the data transmitted from the status data server in sequence, and divide the processed data into training sets and test sets.

[0046] Step 21: Data cleaning.

[0047] The data stored in the state data server is cleaned, including removing outliers in the data, removing abnormal noise contained therein, and filling in missing values.

[0048] Step 22: Resample the data.

[0049] Since the sampling frequencies of sensor data are different, the cleaned data needs to be time-aligned and different resampling methods are used for different types of sensors.

[0050] For linear data, such as current data, voltage data, power data, etc., the linear interpolation method is used for resampling. The formula of the linear interpolation method is as follows:

[0051] ;

[0052] in, and are the target values ​​of two known data points, and They are and The corresponding index value, is the index value that needs to be interpolated. is an estimate obtained by linear interpolation.

[0053] For nonlinear data, such as pump pressure data, vibration data, filtration capacity data, etc., a nonlinear interpolation method is used for resampling. Specifically, a quadratic linear interpolation method is used. The formula of the quadratic linear interpolation method is as follows:

[0054] ;

[0055] in, is the horizontal coordinate of the point to be interpolated, is the interpolation result, that is, given When the value is, the ordinate is calculated by the quadratic interpolation formula; , , are the horizontal coordinates of the known points, which are usually arranged in ascending order, that is, < < ; , , They are respectively , , The corresponding vertical coordinate value.

[0056] Step 23: Data normalization.

[0057] Perform standardization operations on various types of data to obtain processed data. The formula for standardization operation is as follows:

[0058] ;

[0059] in, is time series data, is the mean of the data set, is the standard deviation of the data set, The data value after standardization.

[0060] Step 24: Training / testing dataset division.

[0061] The processed data obtained in step 23 is divided into a test set and a training set according to different training tasks. For example, 80% of the data is divided as a training set, and 20% of the data is divided as a test set.

[0062] Step 3: In the fatigue test loading system digital space, determine the hyperparameters of the log-LSTM model, and train the training set based on the log-LSTM model to generate a state recognition model.

[0063] Step 31: Build the log-LSTM model.

[0064] The present invention uses log-LSTM as a state recognition model. The core of the log-LSTM model is the log-sigmoid activation function, and the network structure is mainly composed of an output gate, an input gate, a forget gate and a cell state.

[0065] The formula for the log-sigmoid activation function is as follows:

[0066] ;

[0067] in, Indicates the activated variable.

[0068] The formula of the forget gate is as follows:

[0069] f t =LogSigmoid( W f ⋅[ h t-1 , x t ]+ b f ) ;

[0070] in, represents the output of the forget gate, and Represent the weight and error of the forget gate respectively, is the hidden state at the previous moment, An input representing the current state.

[0071] The formula for the input gate is as follows:

[0072] i t =LogSigmoid( W i ⋅[ h t-1 , x t ]+ b i ) ;

[0073] in, represents the output of the input gate, and Represent the weight and error of the input gate respectively

[0074] The formula for updating the cell state is as follows:

[0075] ;

[0076] in, Indicates the cell state at the current moment, represents element-by-element multiplication, Indicates the state of the cell at the previous moment. Represents the candidate memory cell state at the current moment.

[0077] The formula for the output gate is as follows:

[0078] o t =LogSigmoid( W o ⋅[ h t-1 , x t ]+ b o ) ;

[0079] in, represents the output of the output gate, and Represent the weight and error of the output gate respectively.

[0080] The final output of the log-LSTM network is:

[0081] ;

[0082] in, is the final output, and tanh is the hyperbolic tangent activation function.

[0083] Step 32: Hyperparameter design and model training.

[0084] The log-LSTM model constructed in step 31 is designed with hyperparameters. The main design parameters include: the number of LSTM layers and the number of LSTMCell units included in each layer of LSTM structure, the number of fully connected layers, the number of batches, and the optimization rate. The hyperparameters of the log-LSTM model in this embodiment include: the number of LSTM layers is 5; the number of LSTMCell units included in each layer of LSTM structure is 2, the number of fully connected layers is 2, the number of batches is 2048, and the optimization rate is 0.001. In order to improve the accuracy, this embodiment uses the cross entropy loss function (CrossEntropyLoss) as the loss function for model training. The training loss drops to a stable level until the accuracy reaches more than 90%, and the classification effect of the model is verified on the test set. The results are as follows: Figure 3 As shown, Figure 3 This is a confusion matrix diagram of the model classification effect. In the confusion matrix, the ordinate is the true state, which indicates the true label of the sample, and the abscissa is the predicted state, which indicates the label judged by the model. The samples are divided into four categories: "normal", "primary degradation", "advanced degradation" and "failure". The value in the grid represents the number of matches between the results predicted by the model and the true label. After completing the training, save the model structure and weight parameters to obtain the state recognition model.

[0085] Step 4: Perform real-time state identification in the fatigue test loading system digital space.

[0086] After the model structure and weight parameters of the state recognition model obtained in step 32 are encapsulated, they are packaged into a state recognition web service, and the current system parameters collected in real time by the loading system parameter sensor are uploaded to the state recognition web service for state recognition. The state recognition web service outputs the abnormal state classification results of the fatigue test loading system, thereby realizing comprehensive evaluation, real-time monitoring and intelligent early warning of the fatigue test loading system state. After the abnormal state classification is completed, the abnormal state classification results, i.e., the state information, are fed back to the fatigue test loading system, and the fatigue test loading system can display the state information.

[0087] The present invention also proposes a fatigue test loading system state monitoring system based on digital twins, which includes a physical space subsystem for describing the physical entity part of the fatigue test loading system and a digital space subsystem for describing the digitized part of the fatigue test loading system, wherein the physical space subsystem includes a fatigue test loading system, a loading system parameter sensor, a data bus module and a state data server, and the digital space subsystem includes a data preprocessing module, a model training module and a state identification module.

[0088] Specifically, the fatigue test loading system is used to provide fatigue load loading when performing fatigue test tasks. It is a test equipment that provides fatigue loading for fatigue testing of engineering materials. It provides periodic reciprocating motion through hydraulic transmission to perform fatigue loading on the test sample. During operation, there is status data that can be obtained by sensors, including but not limited to current data, voltage data, power data, temperature data, pump pressure data, vibration data, filtration capacity data, etc.

[0089] The loading system parameter sensor is a general term for sensors used to collect various state parameters in real time during the operation of the fatigue test loading system, including intelligent circuit breakers, temperature sensors, pressure sensors, vibration sensors and intelligent filters, among which the intelligent circuit breaker is used to collect current, voltage and power parameters, the temperature sensor is used to collect temperature parameters, the pressure sensor is used to collect pressure parameters in the pump, the vibration sensor is used to collect vibration parameters, and the intelligent filter is used to collect filtering capacity parameters. The state data can be collected at different sampling frequencies. For example, the sampling frequency of the intelligent circuit breaker is 10Hz, the sampling frequency of the temperature sensor is 1Hz, the sampling frequency of the pressure sensor is 1Hz, the sampling frequency of the intelligent filter is 1Hz, and the sampling frequency of the vibration sensor is 100Hz. Among them, the intelligent circuit breaker can also be replaced by a combination of current sensors, voltage sensors and power sensors.

[0090] The data bus module is used to connect the loading system parameter sensor through the local area network, summarize the system parameters collected in real time by the loading system parameter sensor, and transmit and store the summarized data in the status data server through the TCP / IP protocol.

[0091] The status data server is an independent database server, which is generally built using MySQL database. It is used to store the operating status data of the fatigue test loading system collected by the sensor for subsequent analysis.

[0092] The state data server uses a relational database for data storage and also transmits the data to the data preprocessing module in the digital space subsystem for processing.

[0093] The data preprocessing module is a module used to process raw sensor data. It has data processing functions and can perform preliminary processing such as data cleaning, resampling and standardization on the data transmitted from the status data server in sequence to form usable data for analysis.

[0094] Furthermore, the data preprocessing module includes:

[0095] The data cleaning unit is used to clean the data stored in the state data server, including eliminating abnormal values ​​in the data, removing abnormal noise contained therein, and filling in missing values.

[0096] The resampling unit is used to resample linear data of different sampling frequencies by using a linear interpolation method and to resample nonlinear data of different sampling frequencies by using a quadratic linear interpolation method.

[0097] Since the sampling frequencies of sensor data are different, the cleaned data needs to be time-aligned, and different resampling methods are used for different types of sensor resampling units.

[0098] For linear data, such as current data, voltage data, power data, etc., the resampling unit uses a linear interpolation method for resampling. The formula of the linear interpolation method is as follows:

[0099] ;

[0100] in, and are the target values ​​of two known data points, and They are and The corresponding index value, is the index value that needs to be interpolated. is an estimate obtained by linear interpolation.

[0101] For nonlinear data, such as pump pressure data, vibration data, filtering capacity data, etc., the resampling unit uses a nonlinear interpolation method for resampling, specifically a quadratic linear interpolation method. The formula of the quadratic linear interpolation method is as follows:

[0102] ;

[0103] in, is the horizontal coordinate of the point to be interpolated, is the interpolation result, that is, given When the value is, the ordinate is calculated by the quadratic interpolation formula; , , are the horizontal coordinates of the known points, which are usually arranged in ascending order, that is, < < ; , , They are respectively , , The corresponding vertical coordinate value.

[0104] The standardization unit is used to perform standardization operations on various types of data to obtain processed data. The formula for standardization operation is as follows:

[0105] ;

[0106] in, is time series data, is the mean of the data set, is the standard deviation of the data set, The data value after standardization.

[0107] The partitioning module is used to divide the data processed by the data preprocessing module into a training set and a test set according to a certain ratio. For example, 80% of the data is divided as a training set and 20% of the data is divided as a test set.

[0108] The model design module is used to design the model structure and algorithm principle, and design the deep learning network structure. This embodiment adopts the log-LSTM structure for network design, builds the log-LSTM model and determines the hyperparameters of the log-LSTM model, including the number of LSTM network layers, the number of LSTMCell units, the number of fully connected layers, the number of batches, the optimization rate and the loss function, etc. The network structure and hyperparameters of the log-LSTM model can be found in the aforementioned method embodiment, which will not be repeated here.

[0109] The model training module is used to select the loss function after the hyperparameter design is completed in the model design module. The training set is trained based on the log-LSTM model to generate a state recognition model. The loss function uses the cross entropy loss function (CrossEntropyLoss) to improve the accuracy. During model training, the training loss drops to a stable level until the accuracy reaches more than 90%. The classification accuracy of the model is verified on the test set. The results are as follows: Figure 3 After the training is completed, the model structure and weight parameters are saved to obtain the state recognition model.

[0110] The state recognition module is used to service-encapsulate the model structure and weight parameters of the trained state recognition model, that is, after the model is encapsulated, it is packaged into a state recognition web service. It is also used to receive the current system parameters collected in real time by the loading system parameter sensor and perform state recognition based on the current system parameters. Finally, it outputs the abnormal state classification result of the fatigue test loading system, and feeds back the abnormal state classification result, that is, the state information, to the fatigue test loading system, so as to receive the state parameters and feed back the state information. The fatigue test loading system can display the fed-back state information.

[0111] The present invention provides a fatigue test loading system state monitoring method and system based on digital twin. A multi-source sensor network is deployed to collect parameters such as current, voltage, vibration, and pressure, a mapping between the physical space of the fatigue test loading system and the digital space of the fatigue test loading system is constructed, and a log-LSTM model is used to find the before-and-after correlation features of time series data, so as to realize feature extraction of fatigue test loading system operation data, effectively judge the current state based on the time series features, and finally output the abnormal state classification result of the fatigue test loading system, so as to timely discover the abnormal states of multiple components of the fatigue test loading system, ensure the stable operation of the test process, and avoid economic losses and R&D interruptions caused by unnecessary downtime.

[0112] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A fatigue test loading system condition monitoring method based on digital twins, characterized in that: The following steps are involved: Step 1: Building a physical space for a fatigue test loading system, wherein the physical space includes a fatigue test loading system, a loading system parameter sensor, a data bus, and a status data server. The loading system parameter sensor is deployed in the fatigue test loading system to collect current, voltage, power, temperature, pressure in the pump, vibration, and filtering capacity parameters in real time. The system parameters collected by the loading system parameter sensor are connected to the data bus through a local area network, and are aggregated in the data bus and transmitted to the status data server for storage; Step 2: Building a fatigue test loading system digital space, wherein the fatigue test loading system digital space sequentially cleans, resamples and standardizes the data transmitted from the state data server, and divides the processed data into a training set and a test set; Step 3: Determine the hyperparameters of the log-LSTM model, and train the training set based on the log-LSTM model to generate a state recognition model; Step 4: After encapsulating the model structure and weight parameters of the state recognition model, package them into a state recognition web service, upload the current system parameters collected in real time by the loading system parameter sensor to the state recognition web service for state recognition, output the abnormal state classification result of the fatigue test loading system, and feed back the abnormal state classification result to the fatigue test loading system.

2. The fatigue test loading system state monitoring method based on digital twin according to claim 1 is characterized in that: The data resampling in step 2 includes: Linear interpolation is used for linear data, and the formula is: ； in, and are the target values ​​of two known data points, and They are and The corresponding index value, is the index value that needs to be interpolated. is an estimate obtained by linear interpolation; The quadratic linear interpolation method is used for nonlinear data, and the formula is: ； in, is the horizontal coordinate of the point to be interpolated, is the interpolation result, that is, given When the value is, the ordinate is calculated by the quadratic interpolation formula; , , are the horizontal coordinates of the known points, which are arranged in ascending order, that is, < < ; , , They are respectively , , The corresponding vertical coordinate value.

3. The fatigue test loading system state monitoring method based on digital twin according to claim 1 or 2 is characterized in that: The log-LSTM model includes: The input gate, forget gate, and output gate all use log-sigmoid activation function; A cell state updating unit, which dynamically updates the cell state according to the outputs of the input gate and the forget gate; The output gate generates the final state classification result through a hyperbolic tangent activation function.

4. The fatigue test loading system state monitoring method based on digital twin according to claim 1 or 2 is characterized in that: In step 3, the cross entropy loss function is used to train the model until the training accuracy reaches more than 90%.

5. The fatigue test loading system state monitoring method based on digital twin according to claim 1 or 2 is characterized in that: The loading system parameter sensors include an intelligent circuit breaker, a temperature sensor, a pressure sensor, a vibration sensor and an intelligent filter, wherein the sampling frequency of the intelligent circuit breaker is 10 Hz, the sampling frequencies of the temperature sensor, the pressure sensor and the intelligent filter are 1 Hz, and the sampling frequency of the vibration sensor is 100 Hz.

6. The fatigue test loading system state monitoring method based on digital twin according to claim 1 or 2 is characterized in that: The hyperparameters of the log-LSTM model include: the number of LSTM layers is 5, the number of LSTMCell units in each layer is 2, the number of fully connected layers is 2, the number of batches is 2048, and the optimization rate is 0.

001.

7. The fatigue test loading system state monitoring method based on digital twin according to claim 1 or 2 is characterized in that: The data bus transmits data to the status data server via TCP / IP protocol.

8. The fatigue test loading system state monitoring method based on digital twin according to claim 1 or 2 is characterized in that: The state data server uses a MySQL database for data storage.

9. The fatigue test loading system condition monitoring system based on digital twin is characterized by: Including physical space subsystem and digital space subsystem: The physical space subsystem includes: Fatigue test loading system, used to provide fatigue load loading when performing fatigue test tasks; A loading system parameter sensor is deployed in the fatigue test loading system and is used to collect the current, voltage, power, temperature, pump pressure, vibration and filtering capacity parameters of the fatigue test loading system in real time; A data bus module, used to connect the loading system parameter sensor via a local area network, summarize the system parameters collected by the loading system parameter sensor and transmit the summarized data to a status data server; The state data server is used to store the data transmitted by the data bus module; The digital space subsystem includes: A data preprocessing module, used to sequentially perform data cleaning, resampling and standardization on the data transmitted from the status data server; A partitioning module is used to divide the processed data into a training set and a test set; Model design module, used to determine the hyperparameters of the log-LSTM model; A model training module, used for training the training set based on the log-LSTM model to generate a state recognition model; The state recognition module is used to encapsulate the model structure and weight parameters of the state recognition model into a state recognition web service, receive the current system parameters collected in real time by the loading system parameter sensor and perform state recognition, output the abnormal state classification result of the fatigue test loading system, and feed back the abnormal state classification result to the fatigue test loading system.

10. The fatigue test loading system state monitoring system based on digital twin according to claim 9 is characterized in that: The data preprocessing module comprises: Data cleaning unit, used to remove outliers in the data, remove abnormal noise contained in it, and fill in missing values; A resampling unit, for resampling linear data of different sampling frequencies by using a linear interpolation method and resampling nonlinear data of different sampling frequencies by using a quadratic linear interpolation method; The standardization unit is used to perform standardization operations on data to obtain processed data.

Images