A Modal Time-Frequency Diagram ResNet Health Management System for Mine Vacuum Contactor

Abstract

The present invention provides a modal time-frequency diagram ResNet health management system for a mine-used vacuum contactor, which overcomes the problems of few samples for contactor health management, aliasing in signal feature extraction, and low state recognition rate, and is used for maintenance warning. The present invention mainly collects the acceleration signal when the mine-used vacuum contactor is switched on to obtain the vibration signal of the mine-used vacuum contactor, and then uses the upper computer algorithm to extract the features of the vibration signal, obtains the time-frequency diagram and inputs it into the trained ResNet model to obtain the prediction result, and judges the state of the mine-used vacuum contactor through the prediction result, so as to realize the health management of the contact. In addition, the health management system of the present invention also simulates various working conditions of the mine-used vacuum contactor through simulation and collects the corresponding vibration signals, expanding the number of training samples, so that a ResNet model with higher accuracy can be trained.

Description

Technical Field

[0001] The present invention relates to the field of health management of mining vacuum contactors, and in particular to a contactor health management system which converts vibration signals into modal time-frequency diagrams and is based on ResNet. Background Art

[0002] According to the control instructions, the vacuum contactor for mining connects and disconnects the motor power supply line when the system is operating normally or fails. Its performance is closely related to the production efficiency and safety of the mine. However, the contact spacing of CKJ5-xxx / 1140 vacuum contactor is very short, only 3mm, and it is frequently opened and closed hundreds of times per hour, and even cuts off the short-circuit current of 4000A. In addition, the humid environment of the mine easily causes contact adhesion, spring fatigue, loose base, rusty iron core and other faults. If not identified in time, it will cause huge economic losses.

[0003] At present, the research on vacuum contactors focuses on offline fault types, fault causes and fault repair analysis. In order to promote the development of unmanned working face technology, online health management of vacuum contactors is on the agenda. Tang Bin et al. determined the fault type by sampling the closing coil current. [Tang Bin, Hu Guoqing, Liu Shuangqiang, Shen Huaqi, Huang Xumin, Yuan Yulin. Fault mode analysis of vacuum contactors based on the combination of simulation and experimental research [J]. High Voltage Electrical Appliances, 2021, 57(05): 175-181+188.] However, compared with vibration signals, coil current cannot fully characterize the health status of the contactor. Given that vibration signals exhibit certain non-stationarity and nonlinearity, it is necessary to develop a vacuum contactor health management system that can adaptively extract signal features and deeply mine the time-frequency characteristics of signals. Summary of the invention

[0004] The present invention is carried out to solve the above problems, and its purpose is to provide a modal time-frequency graph ResNet health management system for mining vacuum contactors. It mainly collects the vibration signal of the mining vacuum contactor under normal conditions and simulates the vibration signal of various faults of the mining vacuum contactor in an analog way and performs down-sampling processing on the collected vibration signal, expands the number of samples and then extracts the adaptive feature signal, obtains the time-frequency graph input model for iterative training to obtain the prediction result, and judges the state of the mining vacuum contactor through the prediction result, so as to realize the health management of the contact. The present invention adopts the following technical solutions:

[0005] The present invention provides a modal time-frequency diagram ResNet health management system for a mine-used vacuum contactor, which is characterized by including: a vibration signal acquisition module for acquiring the vibration signal of the mine-used vacuum contactor; an expert database module for providing training samples for the training of the ResNet model, wherein, through a simulation method, the vibration signal of the mine-used vacuum contactor under a fault condition is simulated, and the simulated vibration signal is downsampled to expand the number of the training samples; a characteristic signal extraction module for extracting signal characteristics of the acquired vibration signal by using a host computer algorithm to obtain a time-frequency diagram; and a ResNet fault diagnosis module for inputting the time-frequency diagram generated by the characteristic signal extraction module into the pre-trained ResNet model to obtain a state judgment of the mine-used vacuum contactor, thereby realizing the health management of the mine-used vacuum contactor.

[0006] The modal time-frequency diagram ResNet health management system for the mine-used vacuum contactor provided by the present invention may further have the following technical characteristics: the vibration signal acquisition module includes a plurality of acceleration sensors, a constant current source, a high-speed acquisition card and a host computer. One end of the constant current source is respectively connected to the plurality of acceleration sensors to supply power to the plurality of acceleration sensors, and the other end of the constant current source is connected to one end of the high-speed acquisition card to serve as a signal channel. The other end of the high-speed acquisition card is connected to the host computer through Ethernet.

[0007] The modal time-frequency diagram ResNet health management system for the mine-used vacuum contactor provided by the present invention may further have the following technical characteristics: the acceleration sensor uses a CT1001L acceleration sensor; the constant current source uses a CT5201 constant current source adapter; the high-speed acquisition card uses an NI MCC USB-E1608 16-bit multi-functional Ethernet acquisition card to convert the vibration signal into a digital signal.

[0008] The modal time-frequency diagram ResNet health management system for the mine-used vacuum contactor provided by the present invention may further have the following technical characteristics: in the expert database module, the fault conditions include: a spring fatigue fault, which is simulated by reducing the opening spring of the mine-used vacuum contactor by 4 mm; a base loosening fault, which is simulated by loosening the fixing screws of the base of the mine-used vacuum contactor by 5 mm; and a core rusting fault, which is simulated by sprinkling dust on the contact surface between the core and the armature of the mine-used vacuum contactor.

[0009] The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor provided by the present invention may further have the following technical features: in the expert database module, the signal sampling rate is 500KHz, the sampling duration is 50ms, and the down-sampling frequency is 50KHz.

[0010] The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor provided by the present invention may further have the following technical features: in the characteristic signal extraction module, the upper computer algorithms include: the variational mode decomposition method, which is used to adaptively decompose the vibration signal into multiple intrinsic mode components with optimal central frequencies, so as to extract the signal characteristic information of the vibration signal; the grey wolf optimization algorithm, which is used to optimize the parameters of the variational mode decomposition method; and the wavelet time-frequency diagram algorithm, which is used to generate a wavelet time-frequency diagram, that is, the time-frequency diagram, by wavelet transform for the multiple intrinsic mode components decomposed by the variational mode decomposition method.

[0011] The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor provided by the present invention may further have the following technical features: the parameters of the variational mode decomposition method include the number of intrinsic mode components K and the penalty factor α. The grey wolf optimization algorithm uses the minimum envelope entropy as the fitness function, sets the range of the number of modal components K to 2-10 and the range of the penalty factor α to 200-4000, and searches for the number of modal components K and the penalty factor α corresponding to the minimum envelope entropy within the set range through iterative optimization, so as to obtain the optimized parameters.

[0012] The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor provided by the present invention may further have the following technical features: the parameters of the wavelet time-frequency diagram algorithm include the sampling frequency, the wavelet basis, and the scale sequence length. The sampling frequency is 50KHZ, the wavelet basis is cmor4-3, and the scale sequence length is 256.

[0013] The modal time-frequency diagram ResNet health management system for mining vacuum contactors provided by the present invention may also have such technical features, wherein training the ResNet model includes the following steps: step S1, data division of the time-frequency diagram generated by the feature signal extraction module to randomly obtain different training sets and test sets, step S2, data preprocessing of the time-frequency diagram in the training set, wherein data processing includes: cropping the training set time-frequency diagram using the resize function; selecting the intermediate image using the centercrop function; and randomly flipping, standardizing and normalizing the cropped training set time-frequency diagram, step S3, inputting the time-frequency diagram of the training set into the ResNet model, setting hyperparameters, and obtaining the trained ResNet model through iterative training.

[0014] Function and Effect of the Invention

[0015] According to the modal time-frequency diagram ResNet health management system of the mining vacuum contactor of the present invention, since the expert database module is used to expand the down-sampled data of the mining vacuum contactor and the ResNet fault diagnosis module is used to train the ResNet model to obtain the prediction result, the current operating status of the contactor can be identified, and the health management of the contactor is realized. Therefore, it is easy to solve the problems of low accuracy in identifying the operating status of the contactor and high randomness due to insufficient data. At the same time, since the collected vibration signal is decomposed by the optimal parameter variational mode to obtain K modal signals, and the K modal signals are generated into a wavelet time-frequency diagram for learning and state recognition, the present invention can more deeply mine the characteristic information of the original vibration signal, better learn the characteristics of each time-frequency diagram, and realize a higher accuracy judgment of the operating status of the contactor. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a logic block diagram of the modal time-frequency diagram ResNet health management system of the mining vacuum contactor in an embodiment of the present invention;

[0017] Figure 2 It is a structural schematic diagram of a vibration signal acquisition module of a ResNet health management system of a modal time-frequency diagram of a vacuum contactor for mining in an embodiment of the present invention;

[0018] Figure 3 It is an example diagram of a wavelet time-frequency diagram based on modal variables of a ResNet health management system for a mine vacuum contactor in an embodiment of the present invention;

[0019] Figure 4 It is a modal time-frequency diagram of the mining vacuum contactor in an embodiment of the present invention and a real-time health management flow diagram of the ResNet health management system. DETAILED DESCRIPTION

[0020] In order to make the technical means, creative features, achieved purposes and effects realized by the present invention easy to understand, the modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor of the present invention will be specifically described below in conjunction with the embodiments and the drawings.

[0021] <Embodiment>

[0022] Figure 1 It is the logic block diagram of the modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor in the embodiment of the present invention.

[0023] As Figure 1 shown, the modal time-frequency diagram ResNet health management system 100 of the mine-used vacuum contactor includes a vibration signal acquisition module 11, an expert database module 12, a characteristic signal extraction module 13, a model training module 14, a data storage module 15, a ResNet fault diagnosis module 16, and a control module 17 for controlling each of the above modules.

[0024] The vibration signal acquisition module 11 obtains the vibration signal of the mine-used vacuum contactor by collecting the acceleration signal when the mine-used vacuum contactor is switched on.

[0025] Figure 2 It is the structural schematic diagram of the vibration signal acquisition module of the modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor in the embodiment of the present invention;

[0026] In this embodiment, as Figure 3 shown, the vibration signal acquisition module 11 includes an acceleration sensor 111, a constant current source 112, a high-speed acquisition card 113, and a host computer 114. Among them, the acceleration sensor uses a CT1001L acceleration sensor. The sensitivity of the sensor is 10 mV / g, the operating frequency range is 0.5 KHz - 10 KHz, the weight is 10 g, the measurement range is 500 g, and the voltage output range is ±5 V. Since the contactor vibration signal is often different when collected at different positions, but the fluctuation trend is roughly the same. After multiple measurements, three positions with the richest vibration signals are selected to install the acceleration sensor 111. One of the acceleration sensors 111 is installed on the base through an M5 screw, and the other two acceleration sensors 111 are respectively adsorbed on the upper and lower parts of the contactor through magnetic suction seats; the constant current source 112 uses a CT5201 constant current source adapter. The constant current source 112 is connected to the acceleration sensor 111 to supply power to the acceleration sensor 111, and at the same time acts as a signal channel, connects to the high-speed acquisition card 113, and is connected to the host computer 114 through Ethernet; the high-speed acquisition card 113 selected is a NI MCC USB-E1608 16-bit multi-functional Ethernet acquisition card.

[0027] Specifically, when collecting the vibration signal of the mine-used vacuum contactor, first switch the switch of the constant current source 112 from the off end to the on end. When the green light is on, it indicates that the adapter has successfully powered the sensor. At the same time, the constant current source 112 also has a signal output gain function, which can amplify the input signal by ten times. Connect the input port of the high-speed acquisition card 113 to the output signal interface of the adapter and connect it to the upper computer 114 through Ethernet to realize the acquisition and storage of data.

[0028] The expert database module 12 is used to provide training samples for the training of the ResNet model. Among them, through simulation, the vibration signals of the mine-used vacuum contactor under normal conditions and fault conditions are simulated, and the simulated vibration signals are downsampled to expand the number of the training samples.

[0029] In this embodiment, the simulation method is adopted to simulate the normal condition and three fault conditions of the contactor, and the corresponding vibration signals are collected. These three fault conditions are: spring fatigue fault, base loosening fault, and iron core rusting fault.

[0030] Specifically, set the signal sampling rate to 500KHz and the sampling duration to 50ms. 60 groups of data are collected for the normal state and the three fault states respectively, and 20 groups are collected by each acceleration sensor. The spring fatigue fault is simulated by reducing the opening spring of the mine-used vacuum contactor by 4mm. The base loosening fault is simulated by loosening the base fixing screws of the mine-used vacuum contactor by 5mm. The iron core rusting and contamination fault is simulated by sprinkling dust on the contact surface between the iron core and the armature of the mine-used vacuum contactor. In view of the problem of the incomplete health management database of the vacuum contactor, the downsampling method is adopted, and the downsampling frequency of the collected vibration signal is set to 50KHz, so that the number of samples is expanded to ten times the original, and each group is equivalent to 600 groups of data, increasing the data samples, reducing the randomness of the contactor state judgment, facilitating more accurate judgment of the contactor state, and better realizing health management.

[0031] The feature signal extraction module 13 uses the upper computer algorithm to extract the signal features of the collected vibration signal to obtain a time-frequency diagram.

[0032] In this embodiment, the upper computer algorithm includes the grey wolf optimization algorithm, the variational mode decomposition method, and the wavelet time-frequency diagram algorithm.

[0033] The variational mode decomposition method is used to adaptively decompose the collected vibration signal into multiple intrinsic mode components with the best center frequencies, so as to extract the signal feature information of the vibration signal.

[0034] Grey Wolf Optimization Algorithm is used to optimize the parameters of the variational mode decomposition method. Since the number of intrinsic mode components K and the penalty factor α are highly correlated with the accuracy of variational mode decomposition, the most suitable number of intrinsic mode components K and penalty factor α need to be selected when using the Grey Wolf algorithm to optimize the variational mode decomposition parameters after variational mode decomposition. In this embodiment, first, the range of the number of decomposed intrinsic mode components K is set to 2 - 10, and the range of the penalty factor α is set to 200 - 4000. By continuously iterating and optimizing, the number of intrinsic mode components K and the penalty factor α selected for the minimum envelope entropy of the intrinsic mode function are found within the set range, and the optimal variational mode decomposition algorithm parameters are obtained.

[0035] Wavelet time-frequency diagram algorithm is used to generate a wavelet time-frequency diagram, that is, the time-frequency diagram, from the multiple intrinsic mode components obtained by decomposing the variational mode decomposition method through wavelet transform. In this embodiment, the sampling frequency of the wavelet time-frequency diagram algorithm is 50KHZ, the wavelet basis is cmor4 - 3, and the length of the used scale sequence is 256.

[0036] Figure 3 It is an example diagram of the wavelet time-frequency diagram based on modal variables of the modal time-frequency diagram ResNet health management system for mine vacuum contactors in the embodiments of the present invention.

[0037] As Figure 3 shown, in this embodiment, the variational mode decomposition method and the optimized parameters are used to decompose the collected vibration signal to obtain K intrinsic mode components, and then through wavelet transform, a wavelet time-frequency diagram is generated.

[0038] Model training module 14 uses the training samples provided in the expert database module 12 to train the model and obtains a trained ResNet model. Among them, the ResNet model uses ResNet50 as the backbone network.

[0039] In this embodiment, the model training module 14 first preprocesses the data, and then inputs the time-frequency diagram of the preprocessed data into the ResNet50 model for iterative training to obtain a trained ResNet model and stores it in the data storage module 15.

[0040] Specifically, first, the time-frequency diagrams of the intrinsic mode components are partitioned. The time-frequency diagrams of the normal state and the three fault states of the contactor are placed in a folder respectively. Each folder contains 600 time-frequency diagrams corresponding to a type of vibration signal. The training set and the test set are randomly partitioned according to the ratio of 8:2. The time-frequency diagrams of the training set are cropped by the resize function, and then the middle image with a size of 224*224*3 is selected by the centercrop function. The cropped time-frequency diagrams are randomly flipped horizontally, normalized and standardized. Then, 1920 randomly selected and preprocessed training set modal time-frequency diagrams are input into the ResNet50 model. The hyperparameters are set as follows: the number of pictures processed in each batch batch_size is set to 16, the number of training iterations epochs is set to 200, the learning rate is set to 0.001, the Adam algorithm is used as the optimizer, and the cross-entropy loss function is selected as the loss function. Iterative training is carried out to obtain a trained ResNet model and save it to the data storage module 15. Finally, the data in the test set is input into the saved ResNet model, and the classification accuracy of 480 test set time-frequency diagrams is used to test the model.

[0041] The data storage module 15 is used to store the trained ResNet model.

[0042] The ResNet fault diagnosis module 16 inputs the time-frequency diagram (corresponding to the vibration signal collected in real time) generated by the feature signal extraction module 13 into the ResNet model stored in the data storage module 15 to obtain the state judgment of the mine-used vacuum contactor, so as to realize the health management of the mine-used vacuum contactor.

[0043] Figure 4 It is a schematic diagram of the real-time health management flow of the modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor in the embodiment of the present invention.

[0044] As Figure 4 shown, based on the above-mentioned modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor, the health management process of the mine-used vacuum contactor is as follows: First, the vibration signal acquisition module 11 collects the acceleration signal when the mine-used vacuum contactor is closed in real time to obtain the real-time vibration signal of the mine-used vacuum contactor. Then, the feature signal extraction module 13 extracts the feature signal from the vibration signal collected in real time to obtain a time-frequency diagram. Finally, the ResNet fault diagnosis module 16 inputs the time-frequency diagram into the trained ResNet model to obtain a prediction result, so as to be able to judge the operating state of the mine-used vacuum contactor through the prediction result and realize contact health management.

[0045] Function and effect of the embodiment

[0046] According to the modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor provided by this embodiment, since the expert database module is used to downsample and expand the data of the mine-used vacuum contactor and the ResNet fault diagnosis module is used to train the ResNet model to obtain the prediction result, it is convenient to efficiently identify the current operating state of the contactor and realize the health management of the contactor. And because the collected vibration signal is decomposed into K modal signals through variational mode decomposition with optimal parameters, and a wavelet time-frequency diagram is generated from the K modal signals for learning and state recognition, the characteristic information of the original vibration signal can be mined more deeply, the characteristics of each time-frequency diagram can be learned better, and the operating state of the contactor can be judged with higher accuracy. The health state can be judged more accurately.

[0047] In addition, since three of the most common fault conditions are simulated by simulation and the corresponding vibration signals are obtained, the number of training samples is effectively increased, so that a better model with higher prediction accuracy can be trained, and thus the health state of the mine-used vacuum contactor can be judged more accurately.

[0048] The above embodiments are only used to illustrate the specific implementation manners of the present invention, and the present invention is not limited to the description scope of the above embodiments.

Claims

1. A modal time-frequency diagram ResNet health management system for a mine-used vacuum contactor, characterized in that Including: A vibration signal acquisition module, which is used to acquire the vibration signal of the mine-used vacuum contactor; An expert database module, which is used to provide training samples for the training of the ResNet model. Among them, through a simulation method, the vibration signal of the mine-used vacuum contactor under fault conditions is simulated, and the simulated vibration signal is downsampled to expand the number of the training samples; A feature signal extraction module, which uses a host computer algorithm to extract signal features from the acquired vibration signal to obtain a time-frequency diagram; and A ResNet fault diagnosis module, which inputs the time-frequency diagram generated by the feature signal extraction module into the pre-trained ResNet model to obtain the state judgment of the mine-used vacuum contactor, so as to realize the health management of the mine-used vacuum contactor. Among them, in the feature signal extraction module, the host computer algorithm includes: A variational mode decomposition method, which is used to adaptively decompose the vibration signal into multiple intrinsic mode components with the best center frequencies, so as to extract the signal feature information of the vibration signal; A grey wolf optimization algorithm, which is used to optimize the parameters of the variational mode decomposition method; and A wavelet time-frequency diagram algorithm, which is used to generate a wavelet time-frequency diagram, that is, the time-frequency diagram, from the multiple intrinsic mode components decomposed by the variational mode decomposition method through wavelet transform. The parameters of the wavelet time-frequency diagram algorithm include a sampling frequency, a wavelet basis, and a scale sequence length. The sampling frequency is 50KHz, the wavelet basis is cmor4-3, and the scale sequence length is 256. Training the ResNet model includes the following steps: Step S1, perform data partitioning on the time-frequency diagram generated by the feature signal extraction module to randomly obtain different training sets and test sets. Step S2, perform data preprocessing on the time-frequency diagram in the training set. Among them, the data processing includes: Using the resize function to crop the time-frequency diagram of the training set; Using the centercrop function to select the middle image; and Performing random flipping, standardization, and normalization processing on the cropped time-frequency diagram of the training set. Step S3, input the time-frequency diagram of the training set into the ResNet model, and set hyperparameters, and obtain the trained ResNet model through iterative training.

2. The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor according to claim 1, wherein: Among them, The vibration signal acquisition module includes a plurality of acceleration sensors, a constant current source, a high-speed acquisition card, and a host computer. One end of the constant current source is respectively connected to the plurality of acceleration sensors to supply power to the plurality of acceleration sensors. The other end of the constant current source is connected to one end of the high-speed acquisition card, acting as a signal channel. The other end of the high-speed acquisition card is connected to the host computer through Ethernet.

3. The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor according to claim 2, wherein: Among them, The acceleration sensor adopts a CT1001L acceleration sensor. The constant current source adopts a CT5201 constant current source adapter. The high-speed acquisition card used is a NI MCC USB-E1608 16-bit multifunctional Ethernet acquisition card, which converts the vibration signal into a digital signal.

4. Modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor according to claim 1, It is characterized in that Among them, in the expert database module, the fault conditions include Spring fatigue fault, simulated by reducing the opening spring of the mine-used vacuum contactor by 4 mm Base loosening fault, simulated by loosening the fixing screws of the base of the mine-used vacuum contactor by 5 mm; and Core rust fault, simulated by sprinkling dust on the contact surface between the core and the armature of the mine-used vacuum contactor.

5. The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor according to claim 1, characterized in that Among them, In the expert database module, the signal sampling rate is 500 KHz, the sampling duration is 50 ms, and the downsampling frequency is 50 KHz.

6. The modal time-frequency diagram ResNet health management system of the mine-used vacuum contactor according to claim 1, characterized in that Among them, The parameters of the variational mode decomposition method include the number of intrinsic mode components K and the penalty factor α The grey wolf optimization algorithm uses the minimum envelope entropy as the fitness function, sets the range of the number of modal components K to 2-10 and the range of the penalty factor α to 200-4000, and through iterative optimization, searches for the number of modal components K and the penalty factor α corresponding to the minimum envelope entropy within the set range, so as to obtain the optimized parameters.

Images