Intelligent diagnosis system and method for switch machine based on small sample multi-fault mode

Abstract

The application provides a switch machine intelligent diagnosis system and method based on a small sample multi-fault mode, and the system comprises: a data management module, which collects equipment operation condition data in real time, and stores the data in a source database in the data management module after data preprocessing; a fault diagnosis module, which takes the data in the source database as input, identifies and outputs normal data and fault data G0; a data enhancement module, which is used for expanding the fault data G0 identified by the fault diagnosis module into new fault data G1, and storing the fault data G0 and the new fault data G1 into an expansion database in the data management module; and a fault positioning module, which takes the data in the expansion database as input, identifies and outputs a fault type. The application realizes monitoring and management of the equipment operation process.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of industrial equipment fault diagnosis, and particularly relates to a turnout switch machine intelligent diagnosis system and method based on small sample multi-fault modes. BACKGROUND

[0002] With the rapid expansion of urban rail transit networks, the reliability of turnout switch machines, as the core equipment for controlling the direction of train operation, has become a key factor affecting the transportation efficiency and safety of rail transit systems.

[0003] Taking a subway operation system as an example, in the face of a large line network with more than 4 million passenger flows per day and more than 500 kilometers of operating mileage, the turnout equipment needs to complete more than 10 million high-precision switching operations per year. However, there are two problems in current engineering practice: one is that the traditional maintenance mode has structural defects such as cycle solidification and response lag, and the other is that there is a shortage of fault data and the equipment failure shows complex mechanism with hidden and sudden characteristics. With the help of an intelligent diagnosis system, the fault mechanism state of the equipment can be quickly located, the maintenance cost can be significantly reduced, resource waste can be avoided, the operation management process can be optimized, and a strong guarantee can be provided for the safe and stable operation of the rail transit system. SUMMARY

[0004] The purpose of the present application is to provide a turnout switch machine intelligent diagnosis system and method based on small sample multi-fault modes, which realizes the monitoring and management of the operation process of the equipment, facilitates the equipment managers and enterprise managers to have a more comprehensive understanding of the health status of all equipment, and facilitates more refined management. The technical scheme adopted is:

[0005] A turnout switch machine intelligent diagnosis system based on small sample multi-fault modes, comprising:

[0006] A data management module, which acquires equipment operation data in real time and stores the data preprocessed in a source database in the data management module;

[0007] A fault diagnosis module, which takes the data in the source database as input, identifies and outputs normal data and fault data G0;

[0008] A data enhancement module, which is used for expanding the fault data G0 identified by the fault diagnosis module into new fault data G1, and storing the fault data G0 and the new fault data G1 in an augmented database in the data management module;

[0009] A fault location module, which takes the data in the augmented database as input, identifies and locates the fault type.

[0010] Preferably, the fault diagnosis module is a trained hidden Markov model HMM.

[0011] Preferably, the fault diagnosis module trains the parameters of the hidden Markov model HMM based on a dynamic hyena optimization algorithm DHOA module and a Baum-Welch algorithm module;

[0012] The dynamic hyena optimization algorithm DHOA module replaces the exploration weight ( , which is a random number between 0 and 1, in the hyena optimization algorithm HOA with a dynamic exploration weight .

[0013] The M step of the Baum-Welch algorithm module updates the mean and variance based on the dynamic exploration weight , and trains the hidden Markov model HMM.

[0014] Preferably, the dynamic exploration weight is:

[0015] = .

[0016] wherein, - the current number of iterations of the dynamic hyena optimization algorithm DHOA module, - the total number of iterations of the dynamic hyena optimization algorithm DHOA module.

[0017] Preferably, the mobile flip bottleneck convolution MBConv of the fault positioning module introduces a sparse squeeze-and-excitation module, adaptive channel compression, and an H-Swish activation function.

[0018] Preferably, the data enhancement module is an exponential moving average type vector quantization variational autoencoder EMA_VQVAE.

[0019] Preferably, the data enhancement module includes a residual block Res-Stack, and a vector quantization method applying an exponential moving average mechanism is applied to the output quantity of the first convolutional layer after the residual block Res-Stack, thereby enhancing the sparse fault sample representation capability through a stable codebook update mechanism, effectively solving the problems of gradient oscillation and encoding space collapse.

[0020] Specifically, a turnout switch machine intelligent diagnosis system based on small sample multi-fault mode includes a data management module, a fault diagnosis module, a data enhancement module, and a fault positioning module.

[0021] The data management module collects equipment operation data in real time through a circuit sensing system, and stores the data after data preprocessing in a source database;

[0022] The fault diagnosis module takes the data of the source database as input. The dynamic hamilton optimization algorithm (DHOA) and the Baum-Welch algorithm are used to cooperatively optimize the parameters of the hidden Markov model (HMM), and thus the DHO-BW_HMM can quickly and accurately perform accurate binary classification on the normal state and the fault state, thereby solving the timeliness problem.

[0023] The data enhancement module takes the fault data in the source database of the data management module as input, and uses the exponential moving average vector quantization variational autoencoder (EMA_VQVAE) to expand the data.

[0024] The fault location module takes the data of the expanded database as input, and uses the improved efficient network to locate the fault, accurately identifies the specific fault from the fourteen faults, and real-time displays and pushes to the user, and alarms the equipment fault, and then efficiently schedules the relevant personnel to repair.

[0025] Further, the data management module includes a data acquisition function, a data preprocessing function and a database management function.

[0026] The data acquisition function is used to store the real-time data of the circuit sensing system on the turnout switch machine in the source database.

[0027] The data preprocessing function is used for standardizing the original data.

[0028] The database management function is used for data transmission between the source database and the expanded database. The source database uses MySQL to store and manage real-time data, and uploads to the database for saving at a fixed time period every day, which is used for data analysis functions of fault diagnosis, data enhancement and the like, and can also provide real-time data query function for users. The expanded database is used to store new fault data and source fault data generated by the data enhancement module, and is used for data analysis functions of fault location and the like, and can also provide visual data query function for users.

[0029] The fault diagnosis module uses a classification recognition method, and uses the DHO-BW_HMM to quickly and accurately perform accurate binary classification on the normal state and the fault state. The corresponding method is as follows.

[0030] First, define the forward variable and the backward variable , and then calculate and .

[0031] represents the expected number of times of transition from state i to state j . representing the state i The expectation of a state transition occurs.

[0032] : Given an observation sequence O and a model , define the forward probability as the probability of the hidden state at time t given the observation sequence , denoted as .

[0033] The specific process is as follows:

[0034] First, initialize the forward probability of each hidden state at time 1: ; second, recursively calculate the forward probability T+1 from time 2 to time ;

[0035] Finally, the result is .

[0036] : Define the backward probability as the probability of the hidden state at time t given the observation sequence , denoted as .

[0037] The specific process is as follows:

[0038] First, initialize the backward probability of each hidden state at time T : ;

[0039] Second, recursively calculate the backward probability T-1 from time to time 1 ;

[0040] Finally, the result is .

[0041] According to the forward-backward variable, make max, we can get:

[0042] .

[0043] : The probability that the hidden state of the device at time t is i and the hidden state transition at time t+1 is j.

[0044] ;

[0045] : The probability that the hidden state of the device at time t is i .

[0046] ;

[0047] further cross E-M step iterations until the parameters converge.

[0048] use to represent the re-estimate of the initial state probability .

[0049] ;

[0050] use to represent the re-estimate of the initial state probability .

[0051] ;

[0052] where represents the expected number of times that the state i is transitioned to the state j , represents the expected number of times that the state i is in the state .

[0053] where is the current optimal value.

[0054] the observation probability distribution is fitted with a Gaussian model, where , respectively, represent the weight, mean and variance.

[0055] use to represent the re-estimate of the mean, with the following formula:

[0056] ;

[0057] = ;

[0058] ;

[0059] where is the current optimal value. is the set of hidden states at time .

[0060] Similarly, the variance is calculated: use to represent the re-estimate of the variance, with the following formula:

[0061] ;

[0062] Further, the trained DHO-BW_HMM is obtained: .

[0063] wherein the elements in are ; the elements in are observation probability distributions , i.e. , , .

[0064] is the current optimal value;

[0065] is the observation sequence at time ; O ;

[0066] The final model is updated with the forward and backward variables , , and then the forward algorithm or the backward algorithm is used for accurate binary classification.

[0067] Further, the data augmentation module uses a vector quantization variant autoencoder with an exponential moving average update mechanism to perform fault sample data augmentation, and then stores it in an augmented database.

[0068] The vector quantization variant autoencoder with an exponential moving average update mechanism uses an exponential moving average (EMA) to update the codebook, and the total loss function removes the vector quantization loss term , which respectively represent the reconstruction loss and the commitment loss. In the quantization process, it can be abstracted as:

[0069] ;

[0070] represents the number of feature vectors in the current small batch of samples mapped to the codebook entry , wherein is a smoothing attenuation coefficient. Third, the commitment loss prevents the latent space from diverging due to a lack of regularization by constraining the geometric distance between the encoder output and the selected embedding.

[0071] ​In addition, LPIPS (Learned Perceptual Image Patch Similarity) and SSIM (Structural Similarity) are selected as evaluation indexes, LPIPS measures the image perceptual quality through the distance of deep feature space, and captures the high-order semantic difference sensitive to human vision; SSIM quantifies the pixel-level structural similarity from three dimensions of brightness, contrast and structure. The combination of the two can comprehensively evaluate the perceptual quality and structural fidelity of the reconstructed data.

[0072] Further, the fault positioning module uses an improved efficient network to locate multiple faults and issue an alarm. The improved efficient network integrates a sparse squeezing and excitation (SE) module, adaptive channel compression and an H-Swish activation function. This method can accurately identify the type of fault, visualize and alarm the user.

[0073] A switch machine intelligent diagnosis method based on small sample multi-fault mode, characterized in that it comprises the following steps:

[0074] Step S1, the data management module pre-processes the real-time equipment data collected by the sensor and stores it in the source database;

[0075] Step S2, the fault diagnosis module requests equipment data from the source database and judges normal and fault through a statistical diagnosis model;

[0076] Step S3, the data enhancement module requests equipment data from the source database, two-dimensionizes the original data, and uses an exponential moving average type vector quantization variational autoencoder to expand the fault data, and stores the original fault data and the expanded fault data in the expansion database;

[0077] Step S4, the fault positioning module inputs the data in the expansion database and uses the improved efficient network to accurately locate 14 types of faults, real-time alarms the fault state, and dispatches corresponding tools and personnel for timely maintenance.

[0078] Compared with the prior art, the advantages of the present application are:

[0079] 1. The present application identifies and manages the health status of railway switch machine operation, which is convenient for equipment management personnel and enterprise managers to have a more comprehensive understanding of the health status of all equipment, and is also convenient for more detailed management, solving the two problems. In specific implementation, such as fault diagnosis and fault positioning module, the health status of industrial equipment is quickly identified, located and alarmed, which can more efficiently dispatch relevant personnel for maintenance.

[0080] 2、The application fuses the dynamic hippo algorithm and the Baum-Welch algorithm and builds DHO-BW_HMM, processed data is input into the model, two classification is carried out through the forward and backward algorithm, and the user is visually displayed through the system.

[0081] 3、The embodiment of the application improves the efficient network, fuses the sparse extrusion and excitation (SE) module, adaptive channel compression and H-Swish activation function, can more quickly and efficiently locate the fault, quickly identify the fault type and alarm, and then visually display the user through the system.

[0082] 4、The application develops an intelligent diagnosis system by systematizing the functions of the above-mentioned multiple modules, realizes the management and diagnosis of the current and power data of the railway turnout switch machine, and improves the efficiency of the management personnel and the staff scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0083] Figure 1 It is the system implementation architecture diagram of the application;

[0084] Figure 2 It is the system workflow diagram of the application;

[0085] Figure 3 It is the fault diagnosis method schematic diagram of the application;

[0086] Figure 4 It is the data enhancement method schematic diagram of the application using EMA_VQVAE;

[0087] Figure 5 It is the fault positioning method schematic diagram of the application using IEfficientNet;

[0088] Figure 6 It is the fault diagnosis method corresponding to the comparative experiment result diagram of the application;

[0089] Figure 7 It is the fault positioning method corresponding to the experimental result diagram of the application;

[0090] Figure 8 It is the experimental result diagram obtained based on the original EfficientNet model;

[0091] Figure 9 It is the experimental result diagram obtained by adding the sparse extrusion and excitation module in the original EfficientNet model. DETAILED DESCRIPTION

[0092] The small sample multi-fault mode based intelligent diagnosis system and method for railway switch machines will be described in more detail below in conjunction with the schematic drawings, in which the preferred embodiments of the present application are represented, it should be understood that the present application described herein can be modified by those skilled in the art, while still achieving the advantageous effects of the present application. Therefore, the following description should be understood as extensive knowledge for those skilled in the art, and not as a limitation of the present application.

[0093] As Figures 1-9 , the present application provides a small sample multi-fault mode based intelligent diagnosis system and method for railway switch machines, aiming at fault diagnosis and fault positioning for small sample multi-fault railway switch machines, including system scheme and method scheme.

[0094] In the intelligent diagnosis system scheme for railway switch machines, as Figure 1 shown, it includes data management module, fault diagnosis module, data enhancement module, fault positioning module, etc.

[0095] The data management module is mainly composed of data acquisition, data preprocessing and database management functions. The data acquisition function is used to collect real-time data of the circuit sensing system on the switch machine and store it in the database. Among them, the real-time data includes equipment running circuit data, including running current change, power change, etc. The data sampling frequency is 400ms.

[0096] The data preprocessing function is used to standardize the original data and generate original features. First, the collected original data is cleaned, and the abnormal values are deleted, the missing values are filled, etc., and the original features are generated by standardization.

[0097] The database management function is used for data transmission between the source database and the augmented database. The source database uses MySQL to store and manage real-time data, and uploads to the source database at fixed time period every day for data analysis functions of fault diagnosis, data enhancement, etc. The augmented database is used to store fault data and original fault data augmented by the data enhancement module, and is used for data analysis functions of fault positioning module, etc. It can also provide users with visual historical data query function. When the source database receives the data migration request sent by the augmented database, it will write SQL statements according to the request information, get the corresponding real-time data from the source database, send it to the augmented database and save the migration record for query.

[0098] The source database mainly stores the following types:

[0099] Table 1. Device field information table

[0100] Field name Meaning Type 1 # Normal A-phase current Normal A-phase current int 1 # Normal B-phase current Normal B-phase current int 1 # Normal C-phase current Normal C-phase current int 1 # Normal Power Normal Power int 2 # 1DQJ self-locking A-phase current 1DQJ self-locking A-phase current int 2 # 1DQJ self-locking B-phase current 1DQJ self-locking B-phase current int 2 # 1DQJ self-locking C-phase current 1DQJ self-locking C-phase current int 2 # 1DQJ self-locking Power 1DQJ self-locking Power int 3 # 2DQJ fails A-phase current 2DQJ fails A-phase current int 3 # 2DQJ fails B-phase current 2DQJ fails B-phase current int 3 # 2DQJ fails C-phase current 2DQJ fails C-phase current int 3 # 2DQJ fails Power 2DQJ fails Power int 4 # Indicates fault A-phase current Indicates fault A-phase current int 4 # Indicates fault B-phase current Indicates fault B-phase current int 4 # Indicates fault C-phase current Indicates fault C-phase current int 4 # Indicates fault Power Indicates fault Power int 5 # Disconnect X1 A-phase current Disconnect X1 A-phase current int 5 # Disconnect X1 B-phase current Disconnect X1 B-phase current int 5 # Disconnect X1 C-phase current Disconnect X1 C-phase current int 5 # Disconnect X1 Power Disconnect X1 Power int …… …… int 13 # Unlocking A-phase current Unlocking A-phase current int 13 # Unlocking B-phase current Unlocking B-phase current int 13 # Unlocking C-phase current Unlocking C-phase current int 13 # Unlocking Power Unlocking Power int 14 # Locking A-phase current Locking A-phase current int 14 #Locking B-phase current Locking B-phase current int 14 #Locking C-phase current Locking C-phase current int 14 #Locking Power Locking Power int 15 #Conversion A-phase current Conversion A-phase current int 15 #Conversion B-phase current Conversion B-phase current int 15 #Conversion C-phase current Conversion C-phase current int 15 #Conversion Power Conversion Power int

[0101] The expansion database mainly stores the following categories:

[0102] Table 2. Field information table

[0103] Field name Meaning Type 1 #1DQJ self-locking A-phase current 1DQJ self-locking A-phase current int 1 #1DQJ self-locking B-phase current 1DQJ self-locking B-phase current int 1 #1DQJ self-locking C-phase current 1DQJ self-locking C-phase current int 1 #1DQJ self-locking Power 1DQJ self-locking Power int 2 #2DQJ fails A-phase current 2DQJ fails A-phase current int 2 #2DQJ fails B-phase current 2DQJ fails B-phase current int 2 #2DQJ fails C-phase current 2DQJ fails C-phase current int 2 #2DQJ fails Power 2DQJ fails Power int 3 #Indicates fault A-phase current Indicates fault A-phase current int 3 #Indicates fault B-phase current Indicates fault B-phase current int 3 #Indicates fault C-phase current Indicates fault C-phase current int 3 #Indicates fault Power Indicates fault Power int 4 #Disconnect X1 A-phase current Disconnect X1 A-phase current int 4 #Disconnect X1 B-phase current Disconnect X1 B-phase current int 4 #Disconnect X1 C-phase current Disconnect X1 C-phase current int 4 #Disconnect X1 Power Disconnect X1 Power int 5 #Disconnect X2 A-phase current Disconnect X2 A-phase current int 5 #Disconnect X2 B-phase current Disconnect X2 B-phase current int 5 #Disconnect X2 C-phase current Disconnect X2 C-phase current int 5 #Disconnect X2 Power Disconnect X2 Power Disconnect X2 B-phase current int 5#Disconnect X2 C-phase current Disconnect X2 C-phase current int 5#Disconnect X2 Power Disconnect X2 Power int 6#Disconnect X3 A-phase current Disconnect X3 A-phase current int 6#Disconnect X3 B-phase current Disconnect X3 B-phase current int 6#Disconnect X3 C-phase current Disconnect X3 C-phase current int 6#Disconnect X3 Power Disconnect X3 Power int 7#Disconnect X4 A-phase current Disconnect X4 A-phase current int 7#Disconnect X4 B-phase current Disconnect X4 B-phase current int 7#Disconnect X4 C-phase current Disconnect X4 C-phase current int 7#Disconnect X4 Power Disconnect X4 Power int 8#Disconnect X5 A-phase current Disconnect X5 A-phase current int 8#Disconnect X5 B-phase current Disconnect X5 B-phase current int 8#Disconnect X5 C-phase current Disconnect X5 C-phase current int 8#Disconnect X5 Power Disconnect X5 Power int 9#The diode broken down A-phase current The diode broken down A-phase current int 9#The diode broken down B-phase current The diode broken down B-phase current int 9#The diode broken down C-phase current The diode broken down C-phase current int 9#The diode broken down Power The diode broken down Power int 10#The diode is open A-phase current The diode is open A-phase current int 10#The diode is open B-phase current The diode is open B-phase current int 10#The diode is open C-phase current The diode is open C-phase current int 10#The diode is open Power The diode is open Power int 11 Disconnect the circuit breaker A-phase current Disconnect the circuit breaker A-phase current int 11 Disconnect the circuit breaker B-phase current Disconnect the circuit breaker B-phase current int 11 Disconnect the circuit breaker C-phase current Disconnect the circuit breaker C-phase current int 11 Disconnect the circuit breaker Power Disconnect the circuit breaker Power int 12 Unlocking A-phase current Unlocking A-phase current int 12 Unlocking B-phase current Unlocking B-phase current int 12 Unlocking C-phase current Unlocking C-phase current int 12 Unlocking Power Unlocking Power int 13 Locking A-phase current Locking A-phase current int 13 Locking B-phase current Locking B-phase current int 13 Locking C-phase current Locking C-phase current int 13 Locking Power Locking Power int 14 Conversion A-phase current Conversion A-phase current int 14 Conversion B-phase current Conversion B-phase current int 14 Conversion C-phase current Conversion C-phase current int 14 Conversion Power Conversion Power int

[0104] A fault diagnosis module, which uses the preprocessed data collected by the data management module as input to perform fault diagnosis. The state monitoring uses the DHO-BW_HMM method:

[0105] For this project, the data type is mainly current. The data from the source database is used as input. First, the samples are balanced and stratified sampling is performed, and then the model is input for classification. The process is as shown in Figure 3 , and the specific process is as follows:

[0106] Step 1: Initialize the initial probability .

[0107] Step 2: E Step

[0108] According to and , the probability expectation value of each sample at time t is calculated i , and the hidden state transition is j . The probability of the hidden state of the device at time t is i . .

[0109] Step 3, M Step, using the expectation value obtained in E step to update the model parameters .

[0110] Step 4: Loop E-M Step until the parameter converges, obtaining the trained DHO-BW_HMM: .

[0111] Where, E the expectation value obtained in step is .

[0112] Update the final model with the forward and backward variable , , and then use the forward algorithm or backward algorithm for accurate binary classification.

[0113] The data augmentation module is used for the enhancement of fault data classified by the fault diagnosis module. The data is first two-dimensionalized and then input into the model. The encoder extracts the features of the input data through a convolutional neural network and maps these features to a discrete latent space. To achieve discrete latent representation, an Exponential Moving Average Vector Quantization (EMA_VQVAE) mechanism is introduced to map the continuous latent space to a limited vector dictionary, thereby ensuring that the generated latent representation is discrete. The decoder is responsible for restoring the discrete latent representation output by the encoder to an approximation of the original data. The decoder also uses a convolutional neural network to input the discrete latent representation and generate reconstructed data as close as possible to the original data through learning. The enhanced fault data and the source fault data are then stored in the augmented database. The specific process is shown in Figure 4

[0114] The EMA_VQVAE (data augmentation module) is used in Fig. 4 to expand the fault data in this paper. The model inputs the fault data pictures in this paper, and after training and reconstruction, it can reconstruct pictures highly similar to the original pictures.

[0115] The improved efficient network (IEfficientNet) is used to locate and identify multiple faults. The specific process is shown in Figure 5

[0116] In Fig. 5, the IEfficientNet (fault location module) is used to diagnose and locate the 14 fault states in this paper.

[0117] After 250 rounds of training of the model with the picture data of the 14 fault states as input, the trained final model is obtained. Then, by inputting a picture of any fault state, the system can determine the probability of the input being a certain fault state according to the final model.

[0118] ​​For example, input the picture of fault state 2, the system can output: the probability of this picture being fault state 0 is 0.1%, being fault state 1 is 0.1%, being fault state 2 is 99%, being fault state 3 is 0.1%, being fault state 4 is 0.1%, being fault state 5 is 0.1%, being fault state 6 is 0.1%, being fault state 7 is 0.1%, being fault state 8 is 0.1%, being fault state 9 is 0%, being fault state 10 is 0.1%, being fault state 11 is 0%, being fault state 12 is 0%, and being fault state 13 is 0%.

[0119] The model considers the problem from three dimensions of width (W) , depth (D) , and resolution (R) , and abstracts the entire network as:

[0120] ;

[0121] The convolution network layer can be defined as a function: i , wherein, f represents an operator, represents that the operator is executed times in Stage i , represents the feature matrix input into Stage X , i represents the height, width, and channel number of . In order to explore the influence of the three factors on the final accuracy, we add X to the formula, and we can obtain the abstracted optimization problem, that is: ;

[0122] ;

[0123] ;

[0124] ;

[0125] ;

[0126] wherein, d is used to scale the depth , r is used to scale the resolution and , w is used to scale the channel number of the feature matrix​ , To limit memory , is the limit of FLOPs.Then a mixing factor φ is used to unify the scaling , which is calculated as follows:

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] ;

[0132] ;

[0133] So the total FLOPs ratio can be approximately expressed as , when the limit ≈ 2, for any FLOPs equivalent increase times.Further fix , and search based on the formula given above, get the best parameters . Further input into the network structure for training, testing, updating parameters, and further fault positioning.

[0134] Among them, including fault alarm, for receiving different health state information sent by the fault positioning module, and according to the fault information real-time push device health state, also can through the mobile phone app and so on carries out the alarm of multi-platform.

[0135] The system workflow of the embodiment of the application, as shown in Figure 2 , includes the following steps:

[0136] 1、The data management module pre-processes the real-time device data collected by the sensor and stores it in the source database. At the same time, the data of the source database is used as input for fault diagnosis, and the fault data in the source database is used as input for data enhancement.

[0137] 2、The fault diagnosis module requests device data from the source database and judges the abnormality through DHO-BW_HMM.

[0138] Subsequently, the fault data is used as input for the data enhancement module to expand the data.

[0139] 3. The fault positioning module requests device data from the augmented database, and will use the improved efficient network to identify and locate multiple faults, and then alarm the system and push it to the device manager.

[0140] The average accuracy of each 10 rounds is solved as shown in FIG. 6 using the model DHO-BW_HMM in this paper, compared with MCA-LSTM, COA_HSMM, GA_HSMM, DHO_RandomForest and RQA_SVM models under the same experimental conditions. Figure 6 As can be seen from the figure, the model proposed in this paper has the highest average accuracy on the data set in this paper.

[0141] Where Epochs represents the number of iterations, and Average Accuracy represents the average accuracy.

[0142] FIG. 7 is an experimental result diagram corresponding to the fault positioning method. The final model obtained by training the IEfficientNet (ImporvedEfficientNet) model for 250 rounds is tested on the test set. Among them, the model IEfficientNet is trained to divide the training set and the test set. The training set is used for model training, and the test set data is not used for training the model of the training set data, and is used to verify the performance of the model to prevent the model from overfitting.

[0143] Where 0~13 represents different types of faults. The ordinate is the true fault state, and the abscissa is the predicted fault state.

[0144] As shown in the 0-0 position, the fault state 0 (i.e. the state of “1DQJ self-closing”), 34 data are tested, and all 34 data are correctly identified, i.e. 100% recognition.

[0145] The 11-11 position represents the fault state 11 (i.e. the state of “unlocking the card group”), 35 data are tested, 34 data are correctly identified, and the recognition rate is 97.1%; One data is identified as state 13 (i.e. one data is identified as the state of “conversion card block”), so the recognition rate of this model for state 11 does not reach 100%.

[0146] FIGS. 8 and 9 are ablation experiments, which are used to prove the effectiveness of the two improved parts (sparse squeezing and excitation module and H-Swish activation function) of the present application.

[0147] FIG. 8 is the result of training using the original EfficientNet model without improvement, and the training conditions are consistent with Figure 7 .

[0148] Figure 9 is a reference in the original EfficientNet model, the sparse squeeze and excitation module is added to carry on the experiment, the training condition is consistent with Figure 7.

[0149] In summary, the present application obtains real-time data of equipment operation by microcomputer circuit monitoring system, establishes bottom model algorithm by using multi-fault diagnosis technology, and evaluates and monitors the health status of the equipment according to the sensing data; the system comprises a data management module for collecting and storing original current, power and other data in the database after preprocessing; a fault diagnosis module for adopting heuristic algorithm to optimize statistical model for normal and fault state binary classification; a data enhancement module for adopting exponential moving average type vector quantization variational autoencoder to expand small sample fault data; a fault positioning module for adopting improved efficient network to position multiple faults on the enhanced fault data; and an intelligent diagnosis system for data expansion, identification and alarm of the equipment status; the present application can realize monitoring, fault diagnosis, data enhancement and fault positioning of railway turnout switch machine status, which helps managers to know the health status of the equipment in time and realize more refined management.

[0150] The above is only the preferred embodiment of the present application, and does not limit the present application in any way. Any person skilled in the art can make any equivalent replacement, modification or change of the technical scheme and technical content disclosed in the present application without departing from the scope of the technical scheme of the present application, which still falls within the protection scope of the present application.

Claims

1. A small sample multi-fault mode based intelligent diagnosis system for a switch machine, characterized in that, Comprise: a data management module that collects equipment operation data in real time, and stores the pre-processed data in a source database in the data management module; a fault diagnosis module that takes the data in the source database as input, identifies and outputs normal data and fault data G0; a data enhancement module that expands the fault data G0 identified by the fault diagnosis module into new fault data G1, and stores the fault data G0 and the new fault data G1 in an augmented database in the data management module; a fault location module that takes the data in the augmented database as input, identifies and locates the fault type; the fault diagnosis module is a trained hidden Markov model HMM; the fault diagnosis module trains the parameters of the hidden Markov model HMM based on a dynamic river horse optimization algorithm DHOA module and a Baum-Welch algorithm module; The dynamic exploration weight in the dynamic hippo optimization algorithm DHOA module is replaced by the dynamic exploration weight ;​ The M-step of the Baum-Welch algorithm module is based on dynamic exploration weights updating the mean and variance, and training a hidden Markov model (HMM); The dynamic exploration weight Is: = ； wherein, - the current number of iterations of the dynamic herd optimization algorithm DHOA module, - the total number of iterations of the dynamic herd optimization algorithm DHOA module.

2. The small sample based multi-failure mode intelligent diagnosis system for switch machines according to claim 1, wherein, the mobile flip bottleneck convolution MBConv of the fault location module introduces a sparse squeeze and excitation module, adaptive channel compression and H-Swish activation function.

3. The small sample based multi-failure mode intelligent diagnosis system for switch machines according to claim 1, wherein, The data enhancement module is an exponential moving average type vector quantization variational autoencoder EMA_VQVAE.

4. The small sample multi-failure mode based switch machine intelligent diagnosis system according to claim 3, characterized in that, The data enhancement module includes a residual block Res-Stack, and applies a vector quantization method with an exponential moving average mechanism to the output of the first convolution layer after the residual block Res-Stack.

5. A turnout switch machine intelligent diagnosis method based on small sample multi-fault mode, based on the turnout switch machine intelligent diagnosis system based on small sample multi-fault mode of any one of claims 1-4, characterized in that, Comprise the following steps: Step S1, the data management module pre-processes the real-time equipment data collected by the sensor and stores it in the source database; Step S2, the fault diagnosis module requests equipment data from the source database and judges normal and fault through a statistical diagnosis model; Step S3, the data enhancement module requests equipment data from the source database, two-dimensions the original data, and uses an exponential moving average type vector quantization variational autoencoder to expand the fault data, and stores the original fault data and the expanded fault data in the augmented database; Step S4, the fault location module inputs the data in the augmented database and uses the improved efficient network to accurately locate 14 kinds of faults, and real-time alarms the fault state, and then dispatches the corresponding tools and personnel for timely maintenance.

Images