Chassis truck fault monitoring method and device and chassis truck comprising device

Abstract

The invention discloses a chassis truck fault monitoring method and device and a chassis truck comprising the device. The method comprises the steps that state parameters of the chassis truck are obtained; processing the state parameters through a convolutional neural network; performing residual connection on the output of the convolutional neural network through a residual network; processing the output of the residual network through a long short-term memory recurrent neural network; and processing the output of the long-short-term memory recurrent neural network through a linear full-connection network to generate a fault monitoring result. According to the chassis truck fault monitoring method, the operation state of the chassis truck can be monitored in real time, and manual intervention is not needed.

Description

Technical Field

[0001] The present invention relates to a method and device for monitoring chassis vehicle faults, and a chassis vehicle including the device. Background Art

[0002] In order to evaluate chassis vehicle faults, maintenance personnel can inspect the chassis vehicle. However, depending on the location of the fault and the qualifications of the maintenance personnel themselves, there is a large uncertainty in manually evaluating chassis vehicle faults. In addition, it is not conducive to reducing labor costs by having maintenance personnel inspect the chassis vehicle. Therefore, a method for automatically monitoring chassis vehicle faults is needed. Summary of the Invention

[0003] The present disclosure provides a method for monitoring chassis vehicle faults, including: obtaining state parameters of the chassis vehicle, where the chassis vehicle is an electric chassis vehicle and the state parameters are the input current curve of the chassis vehicle motor; processing the state parameters through a convolutional neural network; performing a residual connection on the output of the convolutional neural network through a residual network; processing the output of the residual network through a long short-term memory recurrent neural network; and processing the output of the long short-term memory recurrent neural network through a linear fully-connected network to generate a fault monitoring result.

[0004] According to the method for monitoring chassis vehicle faults provided by the present disclosure, the convolutional neural network is a deep linear convolutional neural network configured to obtain spatial local features of the state parameters, and includes a first convolutional block and a second convolutional block, where the first convolutional block has a first convolutional kernel length and the second convolutional block has a second convolutional kernel length.

[0005] According to the method for monitoring chassis vehicle faults provided by the present disclosure, the convolutional neural network includes a pooling layer for reducing the dimension of the outputs of the first convolutional block and the second convolutional block.

[0006] According to the method for monitoring chassis vehicle faults provided by the present disclosure, the convolutional neural network includes a batch normalization layer for performing batch normalization processing on the output of the pooling layer.

[0007] According to the method for monitoring chassis vehicle faults provided by the present disclosure, the residual network is a deep residual network and includes a first deep residual layer for performing a residual connection on the output generated via the first convolutional block and the output generated via the second convolutional block.

[0008] According to the method for monitoring chassis vehicle faults provided by the present disclosure, the long short-term memory recurrent neural network is configured to obtain temporal features of the state parameters and includes a long short-term memory layer and a batch normalization layer, where the long short-term memory layer is used to obtain temporal features of the state parameters and the batch normalization layer is used to perform batch normalization processing on the output of the long short-term memory layer.

[0009] According to the chassis vehicle fault monitoring method provided by the present disclosure, wherein the linear fully connected network includes a global average pooling layer and a fully connected layer, and the fully connected layer includes a first fully connected layer and a second fully connected layer. Among them, the global average pooling layer is used to perform global average pooling processing on the output of the long short-term memory recurrent neural network, and the first fully connected layer and the second fully connected layer are used to reduce the dimension of the output of the global average pooling layer.

[0010] According to the chassis vehicle fault monitoring method provided by the present disclosure, wherein the activation function of the first fully connected layer is the relu function, which is used to prevent gradient explosion, and the activation function of the second fully connected layer is the softmax function, which is used to classify the fault types.

[0011] According to the chassis vehicle fault monitoring method provided by the present disclosure, wherein the chassis vehicle fault types include one or more of the following: the supply voltage is higher than the rated voltage of the chassis vehicle motor; the supply voltage is lower than the rated voltage of the chassis vehicle motor; the chassis vehicle lead screw is deformed; the chassis vehicle flap is deformed; and the chassis vehicle peg is broken.

[0012] According to the chassis vehicle fault monitoring method provided by the present disclosure, wherein the deep linear convolutional neural network includes a third convolutional block, and the length of the third convolutional kernel of the third convolutional block is equal to the smaller one of the length of the first convolutional kernel and the length of the second convolutional kernel.

[0013] According to the chassis vehicle fault monitoring method provided by the present disclosure, wherein the residual network is a deep residual network and includes a second deep residual layer, and a residual connection is performed on the output generated by the long short-term memory recurrent neural network and the output generated by the third convolutional block through the second deep residual network.

[0014] According to the chassis vehicle fault monitoring method provided by the present disclosure, wherein the fault monitoring result is associated with the occurrence probability corresponding to the fault type.

[0015] The present disclosure provides a chassis vehicle fault monitoring device, including: a state parameter acquisition module configured to acquire the state parameters of the chassis vehicle. Among them, the chassis vehicle is an electric chassis vehicle, and the state parameter is the input current curve of the chassis vehicle motor. The state parameter acquisition module includes a sensor and an analog-to-digital conversion chip. Among them, the sensor is configured to collect the analog signal of the state parameters of the chassis vehicle, and the analog-to-digital conversion chip is configured to convert the analog signal of the state parameters into a digital signal for transmission to the processor; a processor configured to receive and restore the digital signal of the state parameters, load the chassis vehicle fault monitoring model, and process the restored state parameters through the loaded chassis vehicle fault monitoring model to obtain a fault monitoring result.

[0016] According to the chassis vehicle fault monitoring device provided by the present disclosure, wherein the processor is a microprogram control unit MCU.

[0017] According to the chassis vehicle fault monitoring device provided by the present disclosure, wherein the sensor is a Hall current sensor.

[0018] According to the chassis vehicle fault monitoring device provided by the present disclosure, the loaded chassis vehicle fault monitoring model includes: a convolutional neural network module configured to process state parameters; a residual network module configured to perform residual connection on the output of the convolutional neural network module; a long short-term memory recurrent neural network module configured to process the output of the residual network module; and a linear fully connected network module configured to process the output of the long short-term memory recurrent neural network module to generate a fault monitoring result.

[0019] According to the chassis vehicle fault monitoring device provided by the present disclosure, the fault monitoring result is associated with the occurrence probability corresponding to the fault type.

[0020] The present disclosure provides a chassis vehicle including the chassis vehicle fault monitoring device described above. For specific details, reference may be made to the description above, and for the sake of brevity, it will not be repeated here.

[0021] The chassis vehicle fault monitoring method, device, and chassis vehicle including the device according to the present disclosure can monitor the operation state of the chassis vehicle in real time without manual intervention; accurately identify the risks of various fault types and improve the fault detection efficiency; the fault detection model is simple and does not require feature extraction; the operation is simple and reduces the professional requirements for operation and maintenance personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] From the following description in conjunction with the drawings, the above and other aspects, features, and advantages of specific embodiments of the present disclosure will become clearer, wherein:

[0023] Figure 1 is a schematic diagram of a state parameter acquisition and processing device of a chassis vehicle according to an embodiment of the present disclosure;

[0024] Figure 2 is a flowchart of a chassis vehicle fault monitoring method according to an embodiment of the present disclosure;

[0025] Figure 3a is a schematic structural diagram of a chassis vehicle fault monitoring model according to an embodiment of the present disclosure;

[0026] Figure 3b is another schematic structural diagram of a chassis vehicle fault monitoring model according to an embodiment of the present disclosure;

[0027] Figure 4is an internal view of a chassis vehicle to which a chassis vehicle fault monitoring model according to an embodiment of the present disclosure can be applied;

[0028] Figure 5 is a diagram of a chassis vehicle fault monitoring device according to the present disclosure. Detailed Embodiments

[0029] Before proceeding with the following detailed description, it may be advantageous to set forth definitions of certain words and phrases used throughout this disclosure. The terms "include" and "comprise" and their derivatives mean including but not limited to. The phrase "at least one," when used with a list of items, means that different combinations of one or more of the listed items may be used, and only one item from the list may be required. For example, "at least one of A, B, and C" includes any of the following combinations: A, B, C, A and B, A and C, B and C, A and B and C.

[0030] Definitions of other specific words and phrases are provided throughout this disclosure. Those of ordinary skill in the art should understand that, in many cases, if not most cases, such definitions apply to the prior and future use of the words and phrases so defined.

[0031] The following describes various embodiments of the principles of the present disclosure in this patent application document in conjunction with the accompanying drawings only by way of illustration and should not be construed in any way as limiting the scope of the present disclosure. Those skilled in the art will understand that the principles of the present disclosure can be implemented in any appropriately arranged system or device. In some cases, the actions described in the present disclosure may be performed in a different order and still achieve the desired result. Additionally, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing may be advantageous.

[0032] The text and the drawings are provided only by way of example to assist in understanding the present disclosure. They should not be construed as limiting the scope of the claims appended hereto in any way. Throughout the drawings, the same reference numerals generally denote the same elements. Although certain embodiments and examples have been provided, it will be apparent to those skilled in the art based on the content of the present disclosure that the illustrated embodiments and examples can be changed without departing from the scope of the present disclosure.

[0033] Figure 1 is a schematic diagram of a state parameter acquisition and processing device for a chassis vehicle according to an embodiment of the present disclosure. Although Figure 1 the chassis vehicle 110, the state parameter acquisition module 120, and the processor 130 are shown separately therein, those skilled in the art can understand that the chassis vehicle 110, the state parameter acquisition module 120, and the processor 130 can be integrated together.

[0034] The chassis truck 110 may include any operable chassis truck, such as a chassis truck for transporting high-voltage electrical appliances such as high-voltage vacuum circuit breakers. The chassis truck is also known as a handcart.

[0035] The status parameter acquisition module 120 is configured to collect the status parameters of the chassis truck and send the status parameters of the chassis truck to the processor 130 for analysis. The status parameter acquisition module 120 includes a sensor 122 and an analog-to-digital conversion chip 124.

[0036] The sensor 122 is configured to collect analog signals of the status parameters from the chassis truck 110. The sensor 122 may include various sensors, including but not limited to current sensors (such as Hall current sensors), vibration sensors, speed sensors, temperature sensors, force sensors, and torque sensors. In the case where the sensor 122 is a current sensor, the current sensor may be connected in series in the circuit of the chassis truck motor and the power supply.

[0037] The analog-to-digital conversion chip 124 is configured to convert the analog signals of the status parameters collected by the sensor 122 into digital signals for subsequent transmission, processing, and analysis.

[0038] The processor 130 may be any conventional processor, such as a commercial CPU (Central Processing Unit), MCU (Microprogram Control Unit). Alternatively, the processor 130 may be a dedicated device, such as an ASIC or other hardware-based processor. The processor 130 may receive and restore the digital signals of the status parameters (for example, when the status parameter is current, restore the digital signal of the status parameter to a current curve), load the chassis truck fault monitoring model, and process the restored status parameters to obtain a fault monitoring result.

[0039] Figure 2 is a flowchart of a chassis truck fault monitoring method according to an embodiment of the present disclosure.

[0040] In block S202, the status parameters of the chassis truck can be obtained. For example, various status parameters such as those of the chassis truck 110 can be obtained through Figure 1 the status parameter acquisition module 120. The status parameters may include current, speed, temperature, force, torque, etc., but the present disclosure is not limited thereto.

[0041] In block S204, the status parameters can be processed by a convolutional neural network. The convolutional neural network may include various neural networks such as a deep linear convolutional neural network. The convolutional neural network can be configured to process the status parameters to obtain the spatial local features of the status parameters. The convolutional neural network can output the processing result in the form of a high-dimensional matrix.

[0042] In block S206, the output of the convolutional neural network can be subjected to residual connection through a residual network. The residual network can include a deep residual network. The residual network can be configured to mitigate the vanishing gradient problem of information transmission between different layers of the machine learning model; accelerate the convergence speed of model training; and achieve identity mapping, so as to reduce the negative impact of multiple layers on the training performance. The residual network can output the processing result in the form of a high-dimensional matrix.

[0043] In block S208, the output of the residual network can be processed through a long short-term memory recurrent neural network. The long short-term memory recurrent neural network can be configured to obtain the temporal features of the state parameters. The long short-term memory recurrent neural network can output the processing result in the form of a high-dimensional matrix.

[0044] In block S210, the output of the long short-term memory recurrent neural network can be processed through a linear fully connected network to generate a fault monitoring result. The linear fully connected network can be configured to map the input features to the output result to classify and output the fault monitoring result. For example, the probabilities of various types of fault monitoring results can be output as the result.

[0045] Figure 3a It is a schematic structural diagram of a chassis vehicle fault monitoring model according to an embodiment of the present disclosure.

[0046] As Figure 3a shown, the chassis vehicle fault monitoring model 300 can include a convolutional neural network 310, a residual network 320, a long short-term memory recurrent neural network 330, and a linear fully connected network 340. The chassis vehicle fault monitoring model 300 can be configured to process the state parameters 302 of the chassis vehicle to obtain the chassis vehicle fault monitoring result 304.

[0047] The state parameters 302 of the chassis vehicle can be similar to the state parameters described with reference to Figure 2 block 202, and will not be elaborated here for the sake of brevity. In one embodiment, the data of the state parameters can be normalized. For example, the mean MEAN and standard deviation STD of the state parameters (such as the waveform of the state parameters) determined to be normal can be calculated. The other obtained state parameters X (such as the waveform of the state parameters) are normalized (X - MEAN) / STD to obtain the normalized state parameters for subsequent processing.

[0048] The convolutional neural network 310 can include various neural networks such as a deep linear convolutional neural network. The convolutional neural network 310 can be configured to process the state parameters to obtain different spatial local features of the state parameters. The convolutional neural network 310 can output the processing result in the form of a high-dimensional matrix.

[0049] The convolutional neural network 310 may include a first convolutional block 3101 and a second convolutional block 3105. The first convolutional block 3101 and the second convolutional block 3105 are configured to obtain different local spatial features of the state parameters 302 of the chassis vehicle.

[0050] The first convolutional block 3101 may include a first convolutional layer 3102, a first pooling layer 3103, and a first batch normalization layer 3104. The first convolutional layer 3102 may have a first convolutional kernel length and the number of convolutional kernels may be 64. For example, the first convolutional layer 3102 may have a first convolutional kernel length of 3. The activation function of the first convolutional layer 3102 may be "relu", and its padding method may be "same" (identical). The first convolutional layer 3102 is configured to receive the input as the state parameters of the chassis vehicle and capture the spatial local features of the state parameters of the chassis vehicle. The first pooling layer 3103 may include a max pooling layer, and its pooling length may be 2. The first pooling layer 3103 may be configured to perform dimensionality reduction, remove redundant information, compress features, simplify network complexity, reduce computational load, reduce memory consumption, and achieve invariance, etc. The first batch normalization layer 3104 may be configured to perform batch normalization on the processed data to reduce subsequent computational load for faster convergence.

[0051] The second convolutional block 3105 may include a second convolutional layer 3106, a second pooling layer 3107, and a second batch normalization layer 3108. The second convolutional layer 3106 may have a second convolutional kernel length and the number of convolutional kernels may be 64. For example, the second convolutional layer 3106 may have a second convolutional kernel length of 1. The activation function of the second convolutional layer 3106 may be "relu", and its padding method may be "same" (identical). The second convolutional layer 3106 is configured to receive the input as the state parameters of the chassis vehicle and capture the spatial local features of the state parameters of the chassis vehicle. The second pooling layer 3107 may include a max pooling layer, and its pooling length may be 2. The second pooling layer 3107 may be configured to perform dimensionality reduction, remove redundant information, compress features, simplify network complexity, reduce computational load, reduce memory consumption, and achieve invariance, etc. The second batch normalization layer 3108 may be configured to perform batch normalization on the processed data to reduce subsequent computational load for faster convergence.

[0052] The residual network 320 may include a deep residual network. The residual network 320 may include a first residual layer 321 such as a deep residual layer. A residual connection is made between the output generated via the first convolutional block 311 and the output generated via the second convolutional block 315 through the first residual layer 321. The first residual layer 321 may be configured to mitigate the vanishing gradient problem of information transmission between different layers of the machine learning model; accelerate the convergence speed of model training; and achieve an identity mapping, so as to reduce the negative impact of multiple layers on the training performance.

[0053] The long short-term memory recurrent neural network 330 may include a long short-term memory layer 331 and a batch normalization layer 332. The long short-term memory recurrent neural network 330 may be configured to obtain the temporal characteristics of the state parameters. The long short-term memory layer 331 may include, for example, 64 hidden layers. The hidden layers are configured to store and transmit information and for non-linear feature extraction, and may return the hidden state at each time step. The batch normalization layer 332 may be used to perform batch normalization on the data processed by the long short-term memory layer 331 to reduce the subsequent computational amount in order to achieve faster convergence.

[0054] The linear fully-connected network 340 may be configured to receive the output of the long short-term memory recurrent neural network 330 and generate a fault monitoring result. The linear fully-connected network 340 may map the input features to the output result to classify and output the fault monitoring result. For example, the probabilities of various types of fault monitoring results may be output as the result. The linear fully-connected network 340 may include a global average pooling layer 341, a first fully-connected layer 342, and a second fully-connected layer 343. The global average pooling layer 341 may be configured to perform global average pooling processing on the received data to enhance the connection between the features and the output fault types, reduce the dimension, and suppress overfitting. The first fully-connected layer 342 may be a fully-connected layer with 128 neurons. The activation function of the first fully-connected layer 342 may be "relu", and may be configured to reduce the dimension and prevent gradient explosion. The second fully-connected layer 343 may be a fully-connected layer with 6 neurons. The activation function of the second fully-connected layer 343 may be "softmax", and may be configured to reduce the dimension and classify, and output the chassis vehicle fault monitoring result 304.

[0055] The chassis vehicle fault monitoring result 304 may be similar to the fault monitoring result described in box 210 with reference to Figure 2 which will not be elaborated here for the sake of brevity.

[0056] Figure 3b is another structural schematic diagram of the chassis vehicle fault monitoring model according to an embodiment of the present disclosure.

[0057] As Figure 3bAs shown, the chassis vehicle fault monitoring model 310 may include a convolutional neural network 310, a residual network 320, a long short-term memory recurrent neural network 330, and a linear fully connected network 340. The chassis vehicle fault monitoring model 310 may be configured to process the state parameters 302 of the chassis vehicle to obtain the chassis vehicle fault monitoring result 304.

[0058] The convolutional neural network 310 may include various neural networks such as a deep linear convolutional neural network. The convolutional neural network 310 may be configured to process the state parameters to obtain different spatial local features of the state parameters. The convolutional neural network 310 may output the processing result in the form of a high-dimensional matrix.

[0059] The convolutional neural network 310 may include a first convolutional block 3101, a second convolutional block 3105, and a third convolutional block 3109. The first convolutional block 3101, the second convolutional block 3105, and the third convolutional block 3109 are configured to obtain different local spatial features of the state parameters 302 of the chassis vehicle.

[0060] The first convolutional block 3101 may include a first convolutional layer 3102, a first pooling layer 3103, and a first batch normalization layer 3104. The first convolutional layer 3102 may have a first convolutional kernel length and the number of convolutional kernels may be 64. For example, the first convolutional layer 3102 may have a first convolutional kernel length of 3. The activation function of the first convolutional layer 3102 may be "relu", and its padding method may be "same" (identical). The first convolutional layer 3102 is configured to receive the input as the state parameters of the chassis vehicle and capture the spatial local features of the state parameters of the chassis vehicle. The first pooling layer 3103 may include a max pooling layer, and its pooling length may be 2. The first pooling layer 3103 may be configured to perform dimensionality reduction, remove redundant information, compress the features, simplify the network complexity, reduce the computational amount, reduce the memory consumption, and achieve invariance, etc. The first batch normalization layer 3104 may be configured to perform batch normalization on the processed data to reduce the subsequent computational amount in order to achieve faster convergence.

[0061] The second convolutional block 3105 may include a second convolutional layer 3106, a second pooling layer 3107, and a second batch normalization layer 3108. The second convolutional layer 3106 may have a second convolutional kernel length and the number of convolutional kernels may be 64. For example, the second convolutional layer 3106 may have a second convolutional kernel length of 1. The activation function of the second convolutional layer 3106 may be "relu", and its padding method may be "same" (identical). The second convolutional layer 3106 is configured to receive an input as the state parameter of the chassis vehicle and capture the spatial local features of the state parameter of the chassis vehicle. The second pooling layer 3107 may include a max pooling layer, and its pooling length may be 2. The second pooling layer 3107 may be configured to reduce the dimension, remove redundant information, compress features, simplify network complexity, reduce the amount of computation, reduce memory consumption, and achieve invariance, etc. The second batch normalization layer 3108 may be configured to perform batch normalization on the processed data to reduce the subsequent amount of computation so as to achieve faster convergence.

[0062] The third convolutional block 3109 may include a third convolutional layer 3110, a third pooling layer 3111, and a third batch normalization layer 3112. The third convolutional layer 3110 may have a third convolutional kernel length and the number of convolutional kernels may be 64. For example, the third convolutional kernel length of the third convolutional layer 3110 may be equal to the smaller one of the first convolutional kernel length and the second convolutional kernel length to reduce the amount of computation. The activation function of the third convolutional layer 3110 may be "relu", and its padding method may be "same" (identical). The third convolutional layer 3110 is configured to receive an input as the state parameter of the chassis vehicle and capture the spatial local features of the state parameter of the chassis vehicle. The third pooling layer 3111 may include a max pooling layer, and its pooling length may be 2. The third pooling layer 3111 may be configured to reduce the dimension, remove redundant information, compress features, simplify network complexity, reduce the amount of computation, reduce memory consumption, and achieve invariance, etc. The third batch normalization layer 3112 may be configured to perform batch normalization on the processed data to reduce the subsequent amount of computation so as to achieve faster convergence.

[0063] The residual network 320 may include a deep residual network. The residual network 320 may include a first residual layer 321 and a second residual layer 322 such as deep residual layers. Residual connection is performed on the output generated via the first convolutional block 311 and the output generated via the second convolutional block 315 through the first residual layer 321. The first residual layer 321 may be configured to alleviate the problem of gradient disappearance in the transfer of information between different layers of the machine learning model; accelerate the convergence speed of model training; and achieve identity mapping, so as to reduce the negative impact of multiple layers on the training performance.

[0064] The long short-term memory recurrent neural network 330 may include a long short-term memory layer 331 and a batch normalization layer 332. The long short-term memory recurrent neural network 330 may be configured to obtain the temporal features of the state parameters. The long short-term memory layer 331 may include, for example, 64 hidden layers. The hidden layers are configured to store and transmit information and for non-linear feature extraction, and may return the hidden state at each time step. The batch normalization layer 332 may be used to perform batch normalization on the data processed by the long short-term memory layer 331 to reduce the subsequent computational amount, so as to achieve faster convergence.

[0065] Through the second residual layer 322 included in the residual network 320, a residual connection can be performed on the output generated by the long short-term memory recurrent neural network 330 and the output generated by the third convolutional block 3109. The second residual layer 322 may be configured to alleviate the problem of vanishing gradients in the transfer of information between different layers of the machine learning model; accelerate the convergence speed of model training; and implement an identity mapping, so that the negative impact of multiple layers on the training performance is reduced.

[0066] The linear fully-connected network 340 may be configured to receive the output of the second residual layer 322 and generate a fault monitoring result. The linear fully-connected network 340 may map the input features to the output result to classify and output the fault monitoring result. For example, the probabilities of various types of fault monitoring results may be output as the result. The linear fully-connected network 340 may include a global average pooling layer 341, a first fully-connected layer 342, and a second fully-connected layer 343. The global average pooling layer 341 may be configured to perform global average pooling processing on the received data, so as to enhance the connection between the features and the output fault types, reduce the dimension, and suppress overfitting. The first fully-connected layer 342 may be a fully-connected layer with 128 neurons. The activation function of the first fully-connected layer 342 may be "relu", and may be configured to reduce the dimension and prevent gradient explosion. The second fully-connected layer 343 may be a fully-connected layer with 6 neurons. The activation function of the second fully-connected layer 343 may be "softmax", and may be configured to reduce the dimension and classify, and output the chassis vehicle fault monitoring result 304.

[0067] Through Figure 3b The shown chassis vehicle fault monitoring model can monitor the operation state of the chassis vehicle in real time without manual intervention; accurately identify the risks of various fault types and improve the fault detection efficiency; the fault detection model is simple and does not require feature extraction; the operation is simple, reducing the professional requirements for operation and maintenance personnel.

[0068] Figure 4 is an internal view of the chassis vehicle to which the chassis vehicle fault monitoring model according to an embodiment of the present disclosure can be applied.

[0069] The chassis truck fault monitoring model can be used to monitor the faults of the chassis truck. The chassis truck can be a chassis truck (also known as a trolley) configured to transport and operate high-voltage electrical appliances such as high-voltage vacuum circuit breakers.

[0070] The chassis truck can be configured to move between the test position and the working position of the switchgear to transport and operate high-voltage electrical appliances. Currently, electric chassis trucks are gradually replacing traditional chassis trucks, driving by motors instead of manual cranking, reducing the workload of operators, and at the same time enabling the function of "one-key sequence control", greatly reducing the operation and maintenance costs. Although replacing manual labor with electrodes reduces the possibility of misoperation caused by inexperienced operation and maintenance personnel, the potential faults of electric chassis trucks are becoming more difficult to detect, and are often not discovered until serious problems are caused. The existing online monitoring functions of electric chassis trucks are relatively single, and most only provide overcurrent and overvoltage monitoring. Therefore, there is a need for a real-time monitoring method for electric chassis trucks that can online real-time monitor the driving motor status of the electric chassis truck and can quickly diagnose or predict common fault types.

[0071] As Figure 4 shown, the chassis truck can include a grounding knife interlock 411, a lead screw 412, an auxiliary switch 413, a flap 414, and a motor 415. Those skilled in the art should understand that the chassis truck can include more or fewer components. The grounding knife interlock 411 can be used to control that the chassis truck cannot move when the grounding knife is in the closed position. The lead screw 412 can be configured to rotate forward and backward to respectively control the chassis truck to be cranked in and out. The auxiliary switch 413 can be configured to connect and disconnect the motor circuit through contacts. The flap 414 can be configured to be interlocked with the chassis truck on the circuit breaker to prevent the circuit breaker from closing during the process of being cranked in and out. The peg 415 (the number of which can be two) can be configured to limit the test position and the operating position of the chassis truck. The current waveform of the motor 415 can be obtained as a state parameter for fault monitoring. However, the present disclosure is not limited thereto, and different state parameters can be obtained through installing different sensors for fault monitoring.

[0072] In one embodiment, the common fault types of the chassis truck can include:

[0073] Fault 1: The supply voltage is higher than the rated voltage of the motor 415 of the chassis truck;

[0074] Fault 2: The supply voltage is lower than the rated voltage of the motor 415 of the chassis truck;

[0075] Fault 3: The lead screw 412 of the chassis truck is deformed;

[0076] Fault 4: The flap 414 of the chassis truck is deformed;

[0077] Fault 5: The peg 415 of the chassis truck is broken.

[0078] In one embodiment, the fault monitoring result is associated with the occurrence probability corresponding to the fault type. For example, a chassis vehicle fault monitoring model can be trained to output a fault monitoring result including a total of six conditions, namely normal conditions and the above five faults. The fault monitoring result can correspond to the occurrence probability of the above six conditions. For example, the fault monitoring result can be data in the form of a one-dimensional matrix of R = [a, b, c, d, e, f]. Among them, a is the probability of the normal condition, b is the probability of fault 1, c is the probability of fault 2, d is the probability of fault 3, e is the probability of fault 4, and f is the probability of fault 5. Faults can be artificially generated to simulate the above common chassis vehicle faults, and the state parameters corresponding to the faults can be obtained to train the fault training model.

[0079] In one embodiment, the maximum value among a, b, c, d, e, and f (i.e., max(a, b, c, d, e, f)) can be obtained, and the condition corresponding to the maximum value is used as the result of this fault monitoring.

[0080] In one embodiment, in response to determining that the maximum value among a, b, c, d, e, and f is greater than or equal to the first threshold, the condition corresponding to the maximum value among a, b, c, d, e, and f is used as the diagnostic result of this fault monitoring.

[0081] In one embodiment, in response to determining that the maximum value among a, b, c, d, e, and f is less than the first threshold, the condition corresponding to the maximum value among a, b, c, d, e, and f is used as the prediction result of this fault monitoring. Alternatively, in response to determining that the maximum value among a, b, c, d, e, and f is less than the first threshold, the condition corresponding to the probability greater than the second threshold (where the second threshold is less than the first threshold) among a, b, c, d, e, and f is used as the prediction result of this fault monitoring.

[0082] Figure 5 It is a diagram of a chassis vehicle fault monitoring device according to the present disclosure.

[0083] As Figure 5 shown, the chassis vehicle fault monitoring device 500 may include a state parameter acquisition module 512 and a processor 502.

[0084] The state parameter acquisition module 512 may be configured to acquire the state parameters of the chassis vehicle. Figure 5 The sensor 514 in Figure 1 may be similar to Figure 5 the sensor 122 in Figure 1 and the analog-to-digital conversion chip 516 in

[0085] The processor 502 may include a convolutional neural network module 504, a residual network module 506, a long short-term memory recurrent neural network module 508, and a linear fully-connected network module 510. Figure 5 The processor 502 in Figure 1 may be similar to the processor 130 in

[0086] According to an embodiment of the present disclosure, a chassis vehicle may include a chassis vehicle fault monitoring device 500 as shown in Figure 5 to perform chassis vehicle fault monitoring. For the sake of brevity, the description is not repeated here.

[0087] Although the present disclosure has been described with exemplary embodiments, various changes and modifications may be suggested to those skilled in the art. The present disclosure is intended to cover such changes and modifications that fall within the scope of the appended claims.

[0088] Any description in the present invention should not be construed as implying that any particular element, step, or function is an essential element that must be included within the scope of the claims. The scope of the subject matter of a patent application is defined only by the claims.

Claims

1. A method for monitoring faults of a chassis vehicle, including: obtaining the state parameters of the chassis vehicle, wherein the chassis vehicle is an electric chassis vehicle, and the state parameters are the input current curves of the chassis vehicle motor; processing the state parameters through a convolutional neural network; performing residual connection on the output of the convolutional neural network through a residual network; processing the output of the residual network through a long short-term memory recurrent neural network; processing the output of the long short-term memory recurrent neural network through a linear fully connected network to generate a fault monitoring result.

2. The method according to claim 1, wherein, the convolutional neural network is a deep linear convolutional neural network configured to obtain the spatial local features of the state parameters, and includes a first convolutional block and a second convolutional block, wherein the first convolutional block has a first convolutional kernel length, and the second convolutional block has a second convolutional kernel length.

3. The method according to claim 2, wherein, the convolutional neural network includes a pooling layer for dimensionality reduction of the outputs of the first convolutional block and the second convolutional block.

4. The method according to claim 3, wherein, the convolutional neural network includes a batch normalization layer for batch normalization processing of the output of the pooling layer.

5. The method according to claim 2, wherein, the residual network is a deep residual network, and includes a first deep residual layer, and performs residual connection on the output generated via the first convolutional block and the output generated via the second convolutional block through the first deep residual layer.

6. The method according to claim 1, wherein, the long short-term memory recurrent neural network is configured to obtain the temporal features of the state parameters, and includes a long short-term memory layer and a batch normalization layer, the long short-term memory layer is used to obtain the temporal features of the state parameters, and the batch normalization layer is used to perform batch normalization processing on the output of the long short-term memory layer.

7. The method according to claim 1, wherein, the linear fully connected network includes a global average pooling layer and a fully connected layer, the fully connected layer includes a first fully connected layer and a second fully connected layer, wherein the global average pooling layer is used to perform global average pooling processing on the output of the long short-term memory recurrent neural network, and the first fully connected layer and the second fully connected layer are used to perform dimensionality reduction on the output of the global average pooling layer.

8. The method according to claim 1, wherein, the activation function of the first fully connected layer is the relu function for preventing gradient explosion, and the activation function of the second fully connected layer is the softmax function for classifying the fault types.

9. The method according to claim 8, wherein, the fault types of the chassis vehicle include one or more of the following: the supply voltage is higher than the rated voltage of the chassis vehicle motor; the supply voltage is lower than the rated voltage of the chassis vehicle motor; the lead screw of the chassis vehicle is deformed; the tipping plate of the chassis vehicle is deformed; and the peg of the chassis vehicle is broken.

10. The method according to claim 2, wherein, the deep linear convolutional neural network includes a third convolutional block, wherein the third convolutional kernel length of the third convolutional block is equal to the smaller one of the first convolutional kernel length and the second convolutional kernel length.

11. The method according to claim 10, wherein, The residual network is a deep residual network and includes a second deep residual layer. A residual connection is performed on the output generated by the long short-term memory recurrent neural network and the output generated by the third convolutional block through the second deep residual network.

12. The method according to claim 1, wherein, the fault monitoring result is associated with the occurrence probability corresponding to the fault type.

13. A chassis vehicle fault monitoring device, comprising: a status parameter acquisition module configured to acquire the status parameters of the chassis vehicle, wherein the chassis vehicle is an electric chassis vehicle, the status parameter is the input current curve of the chassis vehicle motor, and the status parameter acquisition module includes a sensor and an analog-to-digital conversion chip, wherein, the sensor is configured to collect the analog signal of the status parameters of the chassis vehicle, the analog-to-digital conversion chip is configured to convert the analog signal of the status parameter into a digital signal for transmission to the processor; a processor configured to receive and restore the digital signal of the status parameter, load the chassis vehicle fault monitoring model, and process the restored status parameter through the loaded chassis vehicle fault monitoring model to obtain a fault monitoring result.

14. The device according to claim 13, wherein, the processor is a microprogram control unit MCU.

15. The device according to claim 13, wherein, the sensor is a Hall current sensor.

16. The device according to claim 13, wherein, the loaded chassis vehicle fault monitoring model includes: a convolutional neural network module configured to process the status parameter; a residual network module configured to perform a residual connection on the output of the convolutional neural network module; a long short-term memory recurrent neural network module configured to process the output of the residual network module; a linear fully-connected network module configured to process the output of the long short-term memory recurrent neural network module to generate a fault monitoring result.

17. The device according to claim 16, wherein, the fault monitoring result is associated with the occurrence probability corresponding to the fault type.

18. A chassis vehicle comprising the chassis vehicle fault monitoring device according to any one of claims 13-17.

Images