Speed sensor fault diagnosis method and device and storage medium

Abstract

The invention discloses a speed sensor fault diagnosis method and device and a storage medium, and belongs to the technical field of sensor fault diagnosis, and the method comprises the steps: obtaining the traction power data of a train, inputting the traction power data into a pre-trained residual BP network acceleration calculation model, and carrying out the calculation to obtain a predicted acceleration value; the predicted acceleration value is compared with an acceleration value actually measured by the speed sensor, and whether the speed sensor breaks down or not is judged; the real-time acceleration is mapped by collecting the train power data and compared with the actual running acceleration, whether the use condition of the sensor is intact or not is judged, a complex test platform does not need to be built, the construction cost of sensor health monitoring is reduced, compared with a test experience method, diagnosis errors caused by environmental influences are avoided, and the reliability of the sensor health monitoring is improved. And the diagnosis accuracy is higher.

Description

Technical Field

[0001] The present invention relates to a speed sensor fault diagnosis method, device and storage medium, and belongs to the technical field of sensor fault diagnosis. Background Art

[0002] Urban rail transit signaling systems are key technical equipment for ensuring reliable and safe train operation. Speed ​​sensors collect train speed parameters in real time and transmit them to the train control system through the signaling system for high-precision control. The accuracy of these measurements is crucial to the train's operational status and safety. Trains operate in a harsh environment. The train itself generates strong vibrations during movement, and electromagnetic interference from surrounding equipment can severely impact the train's sensor software and hardware. This can lead to data loss and distortion in the sensor's output, disrupting the train control system's analysis, processing, and control functions. This can cause the train control system to operate abnormally, severely threatening train safety and posing unforeseen safety risks and serious consequences.

[0003] Currently, there are two main methods for speed sensor fault diagnosis in engineering: test comparison and empirical testing. The test comparison method, which doesn't rely on experience and is fast, suffers from the complexity and high cost of setting up a test platform. The empirical testing method relies on waveform detection and empirical fault diagnosis, but it requires high operator skills and can significantly impact diagnostic results due to errors in judgment caused by complex operating environments. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a speed sensor fault diagnosis method, device and storage medium to meet the requirements of high-precision speed sensor fault diagnosis and high reliability and safety of train operation.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions: In a first aspect, the present invention provides a speed sensor fault diagnosis method, comprising: Obtain the train's traction power data and input it into the pre-trained residual BP network acceleration calculation model to calculate the predicted acceleration value; Compare the predicted acceleration value with the acceleration value actually measured by the speed sensor to determine whether the speed sensor is faulty; The training method of the residual BP network acceleration calculation model includes: Initialize the weights and thresholds of the residual BP network acceleration calculation model; Collect key parameters during train operation and calculate the train's traction power and acceleration data based on the key parameters; According to the time series characteristics of acceleration and power curves, the corresponding characteristic data are selected; According to the selected characteristic data, the residual BP network acceleration calculation model is established; The residual BP network acceleration calculation model is trained and verified using the train's traction power and acceleration data; It is determined whether the error obtained by verification exceeds the preset error threshold. If so, the weights and thresholds of the residual BP network acceleration calculation model are updated and retrained. Otherwise, the parameters of the residual BP network acceleration calculation model are saved and the training is completed.

[0006] Furthermore, when collecting key parameters during the operation of the train, a wireless data acquisition terminal is used. The wireless data acquisition terminal includes a Hall sensor, a sampling and conditioning circuit, an STM32 main control module and a 4G communication module. The sampling and conditioning circuit designs a differential operational amplifier circuit to filter out high-frequency noise and common-mode noise. The STM32 main control module samples the train current and voltage data at a fixed sampling frequency. The 4G communication module uses GPRS DTU to upload the data collected by the data acquisition terminal to the network server through the 4G network and TCP / IP protocol.

[0007] Furthermore, the collecting of key parameters during the train operation and calculating the train's traction power and acceleration data based on the key parameters include: Collect key parameters during train operation, including train traction current, auxiliary transformer current, and grid voltage data; The formula for calculating the train's traction power is as follows: (2.1); in, is the traction current, For auxiliary current, is the pantograph voltage, is the traction power, is the number of train carriages, is the number of train trailers, For the moment; The formula for calculating the train's acceleration is as follows: (2.2); in, is the train acceleration, is the train speed, The interval between the current moment and the previous moment.

[0008] Furthermore, the corresponding characteristic data is selected based on the time series characteristics of the acceleration and power curves, wherein the characteristic data includes traction power data and acceleration data, and the selection method is as follows: Select the traction power data at the current moment and the previous k moments As a set of inputs to the neural network, if the traction power data before the current moment is less than k, it will be filled with 0; select the current moment Acceleration As a set of outputs of the neural network, the training samples are represented as: (2.3); in, is the traction power data at the current moment and the previous k moments , Acceleration data at the current moment .

[0009] Furthermore, the method further includes: performing normalization processing on the feature data using min-max normalization, the formula of which is as follows: (2.4); in, For an input variable Dimensional original data, For this input variable, the maximum value is For this input variable, the minimum value is for The normalized results.

[0010] Furthermore, the structure of the residual BP network acceleration calculation model includes a residual block structure and a residual neural network model structure, wherein: The residual block structure includes two hidden layers and one BN layer. The output value is nonlinearly activated using the ReLU activation function. The residual blocks use skip connections to jump and sum the input values ​​and activate them using the ReLU function. The output value serves as the input value of the next residual block. The residual neural network model structure includes 3 residual blocks and 2 fully connected layers. The hidden layer of the residual block has 32 nodes. The two fully connected layers serve as input and output layers respectively. The input layer has k+1 nodes and the output layer has 1 node, where k+1 is the sum of the traction power at the current moment and the previous k moments.

[0011] In a second aspect, the present invention provides a speed sensor fault diagnosis device, comprising: The data acquisition module is used to obtain the train's traction power data and input it into the pre-trained residual BP network acceleration calculation model to calculate the predicted acceleration value; A comparison module is used to compare the predicted acceleration value with the acceleration value actually measured by the speed sensor to determine whether the speed sensor is faulty; The training method of the residual BP network acceleration calculation model includes: Initialize the weights and thresholds of the residual BP network acceleration calculation model; Collect key parameters during train operation and calculate the train's traction power and acceleration data based on the key parameters; According to the time series characteristics of acceleration and power curves, the corresponding characteristic data are selected; According to the selected characteristic data, the residual BP network acceleration calculation model is established; The residual BP network acceleration calculation model is trained and verified using the train's traction power and acceleration data; It is determined whether the error obtained by verification exceeds the preset error threshold. If so, the weights and thresholds of the residual BP network acceleration calculation model are updated and retrained. Otherwise, the parameters of the residual BP network acceleration calculation model are saved and the training is completed.

[0012] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the aforementioned methods when executed by a processor.

[0013] In a fourth aspect, the present invention provides a computer device, comprising: Memory, used to store computer programs / instructions; A processor is configured to execute the computer program / instructions to implement the steps of any of the aforementioned methods.

[0014] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which implements the steps of any of the aforementioned methods when executed by a processor.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention primarily relies on mapping real-time acceleration through the collection of train power data, and compares it with the actual acceleration during operation to determine whether the sensor is in good condition. This eliminates the need to build a complex test platform, reducing the construction cost of sensor health monitoring. Furthermore, compared to empirical testing methods, this method is less susceptible to diagnostic errors caused by environmental influences, resulting in higher diagnostic accuracy.

[0016] 2. The present invention mainly relies on the design of residual BP network to map acceleration calculation, which solves the problem of gradient explosion when BP neural network learns complex models and improves the accuracy of acceleration calculation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a speed sensor fault diagnosis method based on a residual BP network of the present invention; Figure 2 This is a principle block diagram of the wireless data acquisition terminal of the present invention; Figure 3 This is a circuit diagram of a sampling and conditioning module of a wireless data acquisition terminal of the present invention; Figure 4 It is a residual block structure diagram in the residual BP network of the present invention; Figure 5 It is the residual BP network structure diagram of the present invention; Figure 6 This is a comparison diagram of the acceleration calculated by the residual BP network model of the present invention and the measured acceleration; Figure 7 This is a comparison diagram of the acceleration calculated by the residual BP network model of the present invention and the fault acceleration. DETAILED DESCRIPTION

[0018] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] Example 1: This example introduces a speed sensor fault diagnosis method, including: Obtain the train's traction power data and input it into the pre-trained residual BP network acceleration calculation model to calculate the predicted acceleration value; Compare the predicted acceleration value with the acceleration value actually measured by the speed sensor to determine whether the speed sensor is faulty; The training method of the residual BP network acceleration calculation model includes: Initialize the weights and thresholds of the residual BP network acceleration calculation model; Collect key parameters during train operation and calculate the train's traction power and acceleration data based on the key parameters; According to the time series characteristics of acceleration and power curves, the corresponding characteristic data are selected; According to the selected characteristic data, the residual BP network acceleration calculation model is established; The residual BP network acceleration calculation model is trained and verified using the train's traction power and acceleration data; It is determined whether the error obtained by verification exceeds the preset error threshold. If so, the weights and thresholds of the residual BP network acceleration calculation model are updated and retrained. Otherwise, the parameters of the residual BP network acceleration calculation model are saved and the training is completed.

[0020] like Figure 1 As shown, the speed sensor fault diagnosis method provided in this embodiment specifically involves the following steps: Step 1: Design a wireless data acquisition terminal with a built-in 4G communication module to realize online collection of train traction current, auxiliary transformer current, and grid voltage data; Step 2: Use the calculated train traction power and acceleration data as a training data set; Step 3: Select appropriate features based on the time series characteristics of the acceleration and power curves, and normalize the data as input and output data of the neural network model; Step 4: Design the residual BP network model structure and set the model parameters; Step 5: training the residual BP network based on the selected feature data and verifying the model validity, thereby obtaining the residual BP network train acceleration calculation model; Step 6: Set the input parameters, perform forward propagation calculations on the residual network, and compare with the actual measurements to determine whether a fault has occurred.

[0021] The wireless data acquisition terminal with built-in 4G communication module designed in step 1 includes: The wireless data acquisition terminal consists of a Hall effect sensor, a sampling and conditioning circuit, an STM32 main control module, and a 4G communication module. The sampling and conditioning circuit uses a differential op amp to filter out high-frequency and common-mode noise. The STM32 main control module samples train current and voltage data at a fixed sampling frequency. The 4G communication module uses a GPRS DTU to upload collected data to a network server via the 4G network and TCP / IP protocol.

[0022] The train traction power and acceleration calculated in step 2 include: Train traction power: The data acquisition terminal collects the voltage between the train and the catenary, the traction current, and the auxiliary transformer current during the train's operation, and indirectly calculates the traction power. Each EMU is equipped with a traction converter, and each trailer is equipped with an auxiliary converter. The train consists of multiple EMUs and trailers. Therefore, the traction power calculation formula is as follows: (2.1); in, is the traction current, For auxiliary current, is the pantograph voltage, is the traction power, is the number of train carriages, is the number of train trailers, For the moment.

[0023] Train acceleration: The train acceleration is calculated based on the train speed provided by the subway company. The formula is as follows: (2.2); in, is the train acceleration, is the train speed, The interval between the current moment and the previous moment.

[0024] In step 3, appropriate features are selected and normalized based on the time series characteristics of the acceleration and power curves, including: Neural network feature selection: select the traction power data at the current moment and the previous k moments As a set of inputs to the neural network, if the traction power data before the current moment is less than k, it will be filled with 0; select the current moment Acceleration As a set of outputs of the neural network, the training samples are represented as: (2.3); in, is the traction power data at the current moment and the previous k moments , Acceleration data at the current moment .

[0025] Normalization: Min-max normalization is used to normalize the feature data. The formula is as follows: (2.4); in, For an input variable Dimensional original data, For this input variable, the maximum value is For this input variable, the minimum value is for The normalized results.

[0026] In step 4, the design of the residual BP network model structure and module parameter settings include: Residual block structure design: 2 hidden layers, 1 BN layer, the output value uses the ReLU activation function for nonlinear activation; the residual blocks use skip connections to jump and sum the input values ​​and activate them with the ReLU function, and the output value serves as the input value of the next residual block; The residual neural network model structure design consists of three residual blocks and two fully connected layers. The hidden layer of the residual block has 32 nodes. The two fully connected layers serve as input and output layers, with the input layer having k+1 nodes and the output layer having 1 node, where k+1 is the sum of the traction power at the current moment and the previous k moments. Model parameter settings: select the LeakyReLU function as the activation function, Adam as the parameter update method, and MSE as the loss function.

[0027] In step 5, the residual BP network model is trained based on the selected feature data. The training process is as follows: Step 51, initializing the weights and thresholds of the residual neural network model; Step 52: construct a training sample set by selecting the traction power data at each moment and the previous k moments and the acceleration corresponding to each moment; Step 53: Use the training set samples to train the residual network; Step 54, use the validation set samples to test the model accuracy; Step 55, determine whether the error exceeds the preset error threshold. If so, update the residual neural network weights and thresholds, and jump to step 53 to retrain the neural network. Otherwise, save the residual neural network model parameters and complete the training. Among them, RMSE is selected as the error evaluation indicator: (2.5); Where, is the sample size, is the true value, is the corresponding predicted value; In step 6, the residual network forward propagation calculation is performed and compared with the actual value for fault analysis, including: Forward propagation calculation: load the trained residual neural network model parameters and input , output acceleration Calculation results.

[0028] Fault diagnosis: The neural network training data set is non-fault data. The error between the acceleration calculated by the trained model and the acceleration value under non-fault conditions is very small, so it can be compared and analyzed with the fault data. d If the error between the acceleration calculated by the neural network and the measured value at a certain proportion of the moments is greater than the set threshold, the sensor is judged to be faulty.

[0029] The following describes the contents involved in the above embodiment in conjunction with a preferred embodiment.

[0030] The present invention uses a domestic subway line as an example. In this example, the train traction current, auxiliary transformer current and network voltage data of a B-type train are collected during the actual operation from Monday to Thursday. The subway company provides the train speed data at the corresponding time. The present invention provides a speed sensor fault diagnosis method based on residual BP network. The specific process is as follows: Figure 1 As shown, the steps include: Step 1: Design a wireless data acquisition terminal such as Figure 2 As shown in the figure, it consists of a Hall sensor, a sampling and conditioning circuit, an STM32 main control module and a 4G communication module. The sampling and conditioning circuit structure is as follows Figure 3 As shown in the figure, the designed differential op amp circuit has an operating bandwidth of 1kHz, an amplitude attenuation of -4.4dB in the low-frequency phase, and an amplitude attenuation of -20dB / decade in the high-frequency phase, effectively filtering out high-frequency noise and common-mode noise above 1kHz. The STM32 main control module is configured with a timer TIM for PWM control. Sampling is triggered on the rising edge of the PWM clock, and the analog-to-digital converter ADC samples the train current and voltage data at a sampling frequency of 100kHz. The 4G communication module uses a GPRS DTU to upload data collected by the data acquisition terminal to a network server via the 4G network and TCP / IP protocol. The data acquisition terminal performs smoothing filtering on every 1000 points of the collected data before transmitting it to the DTU wireless communication module and transmitting it to the server via the serial port. Therefore, the data in the server is transmitted at 0.1s intervals.

[0031] Step 2. In this example, the actual running speed of the train is related to the running power of the train. When the train is in traction and braking conditions, the train power and speed are positively correlated. When the train is in the inertia stage, its speed is usually greater than 40 km / h. At this time, the train traction motor is not working and the traction power is close to 0. Directly selecting speed as the neural network output will make the model difficult to fit. Therefore, acceleration is selected as the neural network output data set.

[0032] Traction power: The data acquisition terminal collects the voltage between the train and the catenary, the traction current, and the auxiliary transformer current during the train's operation, and indirectly calculates the traction power. Each EMU is equipped with a traction converter, and each trailer is equipped with an auxiliary converter. Since this example collects data for a Type B train, which consists of four EMUs and two trailers, the traction power calculation formula is as follows: (1.1); in, is the traction current, For auxiliary current, is the pantograph voltage, is the traction power.

[0033] Train acceleration: The train acceleration is calculated based on the train speed provided by the subway company. The formula is as follows: (1.2); in, is the train acceleration, is the train speed, The interval between the current moment and the previous moment.

[0034] Step 3: When selecting features, select the current moment And the first 20 moments And the traction power data of the first 2 seconds As a set of inputs to the neural network, if at the current moment If the previous traction power data is less than 20, it will be filled with 0 ; Select the current time Acceleration As a set of outputs of the neural network, the training samples are expressed as: (1.3); in, Traction power data at the current moment and 2 seconds ago , Acceleration data at the current moment .

[0035] This example uses min-max normalization to normalize the feature data. The formula is as follows: (1.4); in, For an input variable Dimensional original data, For this input variable, the maximum value is For this input variable, the minimum value is for The normalized results.

[0036] Step 4: The residual block structure of the residual BP network selected in this example is as follows: Figure 4 As shown in the figure, the design consists of 2 hidden layers and 1 BN layer. The output value uses the ReLU activation function for nonlinear activation. The skip connection between the residual blocks is used to jump and sum the input values ​​and activate them with the ReLU function. The output value is used as the input value of the next residual block. The residual BP network model structure is as follows Figure 5 As shown, it consists of 3 residual blocks and 2 fully connected layers. The hidden layer of the residual block has 32 nodes. The 2 fully connected layers serve as input and output layers respectively. The input layer has 21 nodes and the output layer has 1 node. 21 is the sum of the traction power at the current moment and the previous 2 seconds. Model parameter settings: select the LeakyReLU function as the activation function, Adam as the parameter update method, and MSE as the loss function.

[0037] Step 5: training the residual BP network based on the selected feature data and verifying the model validity, thereby obtaining the residual BP network train acceleration calculation model; Step 51, initializing the weights and thresholds of the residual neural network model; Step 52: construct a training sample set by selecting the traction power data at each moment and the previous 2 seconds and the corresponding acceleration at each moment; Step 53: Use the training set samples to train the residual network; Step 54, use the validation set samples to test the model accuracy; Step 55, determine whether the error exceeds the preset error threshold. If so, update the residual neural network weights and thresholds, and jump to step 53 to retrain the neural network. Otherwise, save the residual neural network model parameters and complete the training. Step 6: Load the trained acceleration calculation model and convert the train traction power data of a certain section into As model input, output acceleration Compare with the actual interval running acceleration, such as Figure 6 As shown in the figure, the error between the trained residual BP network acceleration calculation model and the actual acceleration is only 2.84%.

[0038] The neural network training data set is non-fault data. The error between the acceleration calculated by the trained model and the acceleration value under non-fault conditions is very small, so it can be compared and analyzed with the fault data. When the error between the acceleration calculated by the neural network and the measured value for 2 consecutive seconds is greater than 6%, the sensor is judged to be faulty. Figure 7 As shown in Figure 1, when the velocity sensor waveform is lost, the model-calculated acceleration is compared with the measured acceleration. The fault strategy can effectively identify whether the sensor is faulty.

[0039] Embodiment 2: This embodiment provides a speed sensor fault diagnosis device, comprising: The data acquisition module is used to obtain the train's traction power data and input it into the pre-trained residual BP network acceleration calculation model to calculate the predicted acceleration value; A comparison module is used to compare the predicted acceleration value with the acceleration value actually measured by the speed sensor to determine whether the speed sensor is faulty; The training method of the residual BP network acceleration calculation model includes: Initialize the weights and thresholds of the residual BP network acceleration calculation model; Collect key parameters during train operation and calculate the train's traction power and acceleration data based on the key parameters; According to the time series characteristics of acceleration and power curves, the corresponding characteristic data are selected; According to the selected characteristic data, the residual BP network acceleration calculation model is established; The residual BP network acceleration calculation model is trained and verified using the train's traction power and acceleration data; It is determined whether the error obtained by verification exceeds the preset error threshold. If so, the weights and thresholds of the residual BP network acceleration calculation model are updated and retrained. Otherwise, the parameters of the residual BP network acceleration calculation model are saved and the training is completed.

[0040] The specific functional implementation of each of the above modules can be found in the relevant content of the method in Example 1 and will not be elaborated on here.

[0041] Example 3: This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods described in Example 1 are implemented.

[0042] Embodiment 4: This embodiment provides a computer device, including: Memory, used to store computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of any one of the methods described in Example 1.

[0043] Example 5: This embodiment provides a computer program product, including a computer program / instruction, which implements the steps of any method described in Example 1 when executed by a processor.

[0044] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0045] Those skilled in the art will appreciate that embodiments of the present disclosure may be provided as methods, systems, or computer program products. Thus, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0046] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0047] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0048] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure and are not intended to limit its scope of protection. Although the present disclosure has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the disclosed claims to be approved.

Claims

1. A speed sensor fault diagnosis method, characterized in that: include: Obtain the traction power data of the train, input it into the pre-trained residual BP network acceleration calculation model, and calculate the predicted acceleration value; Compare the predicted acceleration value with the acceleration value actually measured by the speed sensor to determine whether the speed sensor is faulty; The training method of the residual BP network acceleration calculation model includes: Initialize the weights and thresholds of the residual BP network acceleration calculation model; Collect key parameters during train operation and calculate the train's traction power and acceleration data based on the key parameters; According to the time series characteristics of acceleration and power curves, the corresponding characteristic data are selected; According to the selected characteristic data, a residual BP network acceleration calculation model is established; The residual BP network acceleration calculation model is trained and verified through the train's traction power and acceleration data; It is determined whether the error obtained by verification exceeds the preset error threshold. If so, the weight and threshold of the residual BP network acceleration calculation model are updated and retrained. Otherwise, the parameters of the residual BP network acceleration calculation model are saved and the training is completed.

2. The speed sensor fault diagnosis method according to claim 1, characterized in that: When collecting key parameters during the operation of the train, a wireless data acquisition terminal is used. The wireless data acquisition terminal includes a Hall sensor, a sampling and conditioning circuit, an STM32 main control module and a 4G communication module. The sampling and conditioning circuit designs a differential operational amplifier circuit to filter out high-frequency noise and common-mode noise. The STM32 main control module samples the train current and voltage data at a fixed sampling frequency. The 4G communication module uses GPRS DTU to upload the data collected by the data acquisition terminal to a network server through a 4G network and TCP / IP protocol.

3. The speed sensor fault diagnosis method according to claim 1, characterized in that: The collecting of key parameters during the train operation and calculating the train's traction power and acceleration data according to the key parameters include: Collect key parameters during train operation, including train traction current, auxiliary transformer current, and grid voltage data; Calculate the traction power of the train, the formula is as follows: （2.1）; in, is the traction current, As auxiliary variable current, is the pantograph voltage, is the traction power, is the number of train carriages, is the number of train trailers, for the moment; The formula for calculating the acceleration of the train is as follows: （2.2）; in, is the train acceleration, is the train speed, is the interval between the current moment and the previous moment.

4. The speed sensor fault diagnosis method according to claim 1, characterized in that: According to the time series characteristics of the acceleration and power curves, the corresponding characteristic data are selected, wherein the characteristic data includes traction power data and acceleration data, and the selection method is as follows: Select the traction power data at the current moment and the previous k moments As a set of inputs to the neural network, if the traction power data before the current moment is less than k, it will be filled with 0; select the current moment Acceleration As a set of outputs of the neural network, the training samples are represented as: （2.3）; in, is the traction power data at the current moment and the previous k moments , The acceleration data at the current moment .

5. The speed sensor fault diagnosis method according to claim 1, characterized in that: The method further includes: normalizing the feature data using min-max normalization, the formula of which is as follows: （2.4）; in, For an input variable Dimensional original data, To do this, enter the maximum value of the variable. For this input variable the minimum value is for The normalized results.

6. The speed sensor fault diagnosis method according to claim 1, characterized in that: The structure of the residual BP network acceleration calculation model includes a residual block structure and a residual neural network model structure, wherein: The residual block structure includes 2 hidden layers and 1 BN layer, and the output value is nonlinearly activated using the ReLU activation function; the input value jumps and sums are jumped between the residual blocks using skip connections and activated with the ReLU function, and the output value is used as the input value of the next residual block; The residual neural network model structure includes 3 residual blocks and 2 fully connected layers. The hidden layer of the residual block has 32 nodes. The two fully connected layers serve as input and output layers respectively. The input layer has k+1 nodes and the output layer has 1 node, where k+1 is the sum of the traction power at the current moment and the previous k moments.

7. A speed sensor fault diagnosis device, characterized in that: include: The data acquisition module is used to obtain the traction power data of the train, input it into the pre-trained residual BP network acceleration calculation model, and calculate the predicted acceleration value; A comparison module is used to compare the predicted acceleration value with the acceleration value actually measured by the speed sensor to determine whether the speed sensor is faulty; The training method of the residual BP network acceleration calculation model includes: Initialize the weights and thresholds of the residual BP network acceleration calculation model; Collect key parameters during train operation and calculate the train's traction power and acceleration data based on the key parameters; According to the time series characteristics of acceleration and power curves, the corresponding characteristic data are selected; According to the selected characteristic data, a residual BP network acceleration calculation model is established; The residual BP network acceleration calculation model is trained and verified through the train's traction power and acceleration data; It is determined whether the error obtained by verification exceeds the preset error threshold. If so, the weight and threshold of the residual BP network acceleration calculation model are updated and retrained. Otherwise, the parameters of the residual BP network acceleration calculation model are saved and the training is completed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer device, characterized in that: include: Memory, for storing computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images