Fault diagnosis method of electric drive axle and training method of neural network model

Abstract

The invention provides a fault diagnosis method of an electric drive axle and a training method of a neural network model, and the method comprises the steps: collecting a vibration signal of a target part through a target vibration sensor, carrying out the processing, and outputting target data; inputting the time domain data into a first neural network model to output a first fault diagnosis result; inputting the frequency domain data into a second neural network model to output a second fault diagnosis result; inputting the normalized data into a third neural network model to output a third fault diagnosis result; and determining a target fault diagnosis result based on the three fault diagnosis results. According to the method, fault diagnosis can be carried out on three different dimensions of the time domain data, the frequency domain data and the normalized data through the multiple neural network models, the final target fault diagnosis result is determined by integrating the three fault diagnosis results, dependence on previous experience is not needed, a threshold range does not need to be selected, and the fault diagnosis efficiency is improved. Therefore, the fault identification efficiency and accuracy can be improved, and the time cost and the economic cost are saved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a fault diagnosis method for an electric drive bridge and a training method for a neural network model. Background Art

[0002] At present, the penetration rate of new energy vehicles in the market is increasing, and the electric drive axle is the power source of new energy vehicles. The electric drive axle is highly integrated with core components such as the drive motor, reducer and differential. The bearings and gears have high rotation times and meshing frequencies among the parts of the electric drive axle, and are very likely to be damaged first, thereby causing failures of the motor, reducer and differential. In the related technology, sensors are usually arranged on the housing of the electric drive axle to collect vibration signals, voltage signals, temperature signals, etc. generated in the working state. By performing feature processing on the collected signals and setting relevant thresholds, the fault diagnosis of the electric drive axle is completed. This method is very dependent on previous experience, and there is no accurate standard for the selection of the threshold range, resulting in low fault identification efficiency and accuracy, and high time and economic costs. Summary of the invention

[0003] The purpose of the present invention is to provide a fault diagnosis method for an electric drive axle and a training method for a neural network model to improve fault identification efficiency and reduce time and economic costs.

[0004] The present invention provides a fault diagnosis method for an electric drive axle, the method comprising: collecting a vibration signal of a target component through a target vibration sensor; wherein the target component comprises: a motor, a reducer and a differential in the electric drive axle; processing the vibration signal and outputting target data; wherein the target data comprises: time domain data, frequency domain data and normalized data corresponding to the vibration signal; inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; and determining a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result.

[0005] Furthermore, the target vibration sensor is a three-way vibration sensor; the vibration signal includes: a first signal in the axial direction, a second signal in the vertical direction, and a third signal in the horizontal direction.

[0006] Further, the target vibration sensor includes: a first vibration sensor, a second vibration sensor and a third vibration sensor; The first vibration sensor is arranged on the housing of the motor and is used to collect the vibration signal of the motor; The second vibration sensor is arranged at the bearing end of the reducer, and is used to collect the vibration signal of the reducer; The third vibration sensor is arranged on the housing of the differential and is used for collecting the vibration signal of the differential.

[0007] Furthermore, the step of processing the vibration signal and outputting target data includes: Performing noise reduction processing on the vibration signal to obtain a noise-reduced vibration signal; Normalizing the vibration signal after noise reduction to obtain normalized data; Slice the normalized data to obtain multiple data segments; For each data segment, according to a preset time domain processing method, the data in the data segment is subjected to time domain processing to obtain the time domain data corresponding to the data segment; wherein the time domain processing method includes: mean value calculation, variance calculation, absolute average value calculation, standard deviation calculation, root mean square value calculation, peak value calculation and minimum value calculation; Based on the time domain data corresponding to each data segment, the time domain data corresponding to the vibration signal is obtained; For each data segment, frequency domain processing is performed on the data in the data segment according to a preset frequency domain processing method to obtain frequency domain data corresponding to the data segment; wherein the frequency domain processing method includes: centroid frequency calculation, mean square frequency calculation, root mean square frequency calculation, frequency variance calculation and frequency standard deviation calculation; Based on the frequency domain data corresponding to each data segment, the frequency domain data corresponding to the vibration signal is obtained.

[0008] Furthermore, the first neural network model is configured with a first weight, the second neural network model is configured with a second weight, and the third neural network model is configured with a third weight; wherein the first weight is equal to the second weight; the third weight is higher than the first weight and higher than the second weight; and the sum of the first weight and the second weight is greater than the third weight.

[0009] Further, based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result, the step of determining the final target fault diagnosis result includes: if the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all the same, determining the first designated fault diagnosis result as the final target fault diagnosis result; wherein, the first designated fault diagnosis result is: the first fault diagnosis result, the second fault diagnosis result or the third fault diagnosis result; if the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all different, determining the third fault diagnosis result corresponding to the third neural network model configured with the third weight as the final target fault diagnosis result; if the first fault diagnosis result is the same as the second fault diagnosis result and different from the third fault diagnosis result, determining the second designated fault diagnosis result as the final target fault diagnosis result; wherein, the second designated fault diagnosis result is: the first fault diagnosis result and the second fault diagnosis result; if the first fault diagnosis result is different from the second fault diagnosis result and the third designated fault diagnosis result is the same as the third fault diagnosis result, determining the third fault diagnosis result as the final target fault diagnosis result; wherein, the third designated fault diagnosis result is the first fault diagnosis result or the second fault diagnosis result.

[0010] The present invention provides a training method for a neural network model, the method comprising: obtaining a preset initial data set; wherein the initial data set comprises: a first subset, a second subset, a third subset and a fourth subset; the first subset stores normal vibration data corresponding to an electric drive axle in a normal working state; the second subset stores motor fault vibration data corresponding to a motor fault; the third subset stores reducer fault vibration data corresponding to a reducer fault; the fourth subset stores differential fault vibration data generated under a differential fault; the initial data set is processed to obtain a time domain feature set, a frequency domain feature set and a normalized feature set corresponding to the initial data set; Based on the target feature set, a training set, a validation set and a test set are determined; wherein the target feature set is: a time domain feature set, a frequency domain feature set and a normalized feature set; based on the training set, an initial network model is trained to obtain a first model; based on the validation set, the first model is optimized to obtain a trained target neural network model, and the target neural network model is evaluated through the test set; wherein, if the target feature set is a time domain feature set, the target neural network model is a first neural network model; if the target feature set is a frequency domain feature set, the target neural network model is a second neural network model; if the target feature set is a normalized feature set, the target neural network model is a third neural network model.

[0011] The present invention provides a fault diagnosis device for an electric drive axle, the device comprising: an acquisition module, used for acquiring vibration signals of target components through a target vibration sensor; wherein the target components include: a motor, a reducer and a differential in the electric drive axle; a first processing module, used for processing the vibration signals and outputting target data; wherein the target data include: time domain data, frequency domain data and normalized data corresponding to the vibration signals; a first output module, used for inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; a second output module, used for inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; a third output module, used for inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; and a first determination module, used for determining a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result.

[0012] The present invention provides a training device for a neural network model, the device comprising: an acquisition module, used to acquire a preset initial data set; wherein the initial data set comprises: a first subset, a second subset, a third subset and a fourth subset; the first subset stores normal vibration data corresponding to an electric drive axle in a normal working state; the second subset stores motor fault vibration data corresponding to a motor fault; the third subset stores reducer fault vibration data corresponding to a reducer fault; the fourth subset stores differential fault vibration data generated under a differential fault; a second processing module is used to process the initial data set to obtain a time domain feature set, a frequency domain feature set and a normalized feature set corresponding to the initial data set; a second determination ... A module is used to determine a training set, a validation set and a test set based on a target feature set; wherein the target feature set is: a time domain feature set, a frequency domain feature set and a normalized feature set; a training module is used to train an initial network model based on the training set to obtain a first model; an optimization module is used to optimize the first model based on the validation set to obtain a trained target neural network model, and evaluate the target neural network model through a test set; wherein, if the target feature set is a time domain feature set, the target neural network model is a first neural network model; if the target feature set is a frequency domain feature set, the target neural network model is a second neural network model; if the target feature set is a normalized feature set, the target neural network model is a third neural network model.

[0013] An electronic device provided by the present invention includes a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement any of the above-mentioned electric drive axle fault diagnosis methods or the above-mentioned neural network model training methods.

[0014] The electric drive axle fault diagnosis method and the neural network model training method provided by the present invention collect vibration signals of target components through target vibration sensors; wherein the target components include: motors, reducers and differentials in the electric drive axle; process the vibration signals and output target data; wherein the target data include: time domain data, frequency domain data and normalized data corresponding to the vibration signals; input the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; input the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; input the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; and determine a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result. This method can perform fault diagnosis in three different dimensions: time domain data, frequency domain data, and normalized data through a variety of neural network models, and determine the final target fault diagnosis result by combining the three fault diagnosis results. It does not need to rely on previous experience or select a threshold range, thereby improving fault identification efficiency and accuracy and saving time and economic costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a fault diagnosis method for an electric drive axle provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an electric drive axle provided by an embodiment of the present invention; Figure 3 A schematic diagram of the position of a target vibration sensor provided by an embodiment of the present invention; Figure 4 A flow chart of a fault diagnosis method for an electric drive axle provided by an embodiment of the present invention; Figure 5 A flowchart of a method for training a neural network model provided in an embodiment of the present invention.

[0017] Figure 6 A schematic diagram of a processing method of a neural network model provided by an embodiment of the present invention; Figure 7A schematic diagram of another processing method of a neural network model provided by an embodiment of the present invention; Figure 8 A schematic structural diagram of a fault diagnosis device for an electric drive axle provided by an embodiment of the present invention; Fig. 9 A schematic diagram of the structure of a training device for a neural network model provided by an embodiment of the present invention; Fig.10 A schematic structural diagram of an electronic device provided by an embodiment of the present invention.

[0018] Icons: 20-motor; 21-reducer; 22-differential; 23-main shaft 1; 24-main shaft 2; 25-half shaft 1; 26-half shaft 2; 27-gear 1; 28-gear 2; 29-gear 3; 30-gear 4; 31-bridge housing; 32-wheel hub; 40-first vibration sensor; 41-second vibration sensor; 43-third vibration sensor. DETAILED DESCRIPTION

[0019] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] At present, with the rapid consumption of traditional energy around the world, clean, reliable and renewable new energy technologies are the only way for the long-term development of mankind. In order to accelerate industrial transformation and achieve the goal of carbon neutrality, my country has begun to pay attention to the development of the new energy industry. As a result, the penetration rate of new energy vehicles in the market has increased. The electric drive axle in new energy vehicles is a highly integrated core component of the drive motor, reducer and differential. Compared with the traditional drive axle, the electric drive axle shortens the transmission chain, improves system efficiency, reduces overall mass, saves space and improves safety.

[0021] Because of the highly integrated characteristics of the electric drive axle, its structure is complex and expensive. Once a fault occurs, it is difficult for manpower to quickly find the location of the fault, resulting in a waste of time and economic costs. Bearings and gears have high rotation times and meshing frequencies among the parts of the electric drive axle, and are easily damaged first, causing failures of the motor, reducer and differential. Most of the existing technologies are based on vibration and noise; for example, when an abnormal sound occurs in the electric drive axle, the cause of the abnormal sound is preliminarily determined by parking or changing the driving direction, and then the specific fault location is determined by disassembly and inspection, but the fault generated at this time is often large. In order to find the fault location in advance, sensors are usually arranged on the housing of the electric drive axle to collect vibration signals, voltage signals, temperature signals, etc. generated in the working state. By processing the collected signals with features and setting relevant thresholds, the fault diagnosis of the electric drive axle is completed. This method relies heavily on previous experience, and there is no accurate standard for selecting the threshold range, resulting in low fault identification efficiency and accuracy, and high time and economic costs. Based on this, an embodiment of the present invention provides a fault diagnosis method for an electric drive axle and a training method for a neural network model. This technology can be applied to applications that require fault diagnosis of an electric drive axle.

[0022] To facilitate understanding of this embodiment, firstly, a fault diagnosis method of an electric drive axle disclosed in an embodiment of the present invention is introduced. Figure 1 As shown, the method comprises the following steps: Step S102, collecting vibration signals of target components through target vibration sensors; wherein the target components include: a motor, a reducer and a differential in an electric drive axle; The motor is the core power component of the electric drive axle, which can convert the electrical energy in the power supply into mechanical energy to drive the vehicle; the reducer can increase the torque by reducing the speed; the differential can make the left and right wheels rotate at different speeds to improve the stability of wheel operation during the turn. In actual implementation, the vibration signals corresponding to the motor, reducer and differential in the electric drive axle can be collected through the target vibration sensor.

[0023] Step S104, processing the vibration signal and outputting target data; wherein the target data includes: time domain data, frequency domain data and normalized data corresponding to the vibration signal; The above-mentioned time domain data is used to reflect the relationship between the signal and time; the above-mentioned frequency domain data is used to reflect the relationship between the signal and frequency; the above-mentioned normalized data is data mapped to a specific range. In actual implementation, by processing the above-mentioned collected vibration signal of the motor, the time domain data, frequency domain data and normalized data corresponding to the vibration signal can be output; by processing the above-mentioned collected vibration signal of the reducer, the time domain data, frequency domain data and normalized data corresponding to the vibration signal can be output; by processing the above-mentioned collected vibration signal of the differential, the time domain data, frequency domain data and normalized data corresponding to the vibration signal can be output.

[0024] Step S106, inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; The first neural network model can be an ANN (Artificial Neural Network) model, which is an algorithmic mathematical model of a biological neural network, used for information processing and pattern recognition. The ANN model may include an input layer, a hidden layer, and an output layer. The input layer is used to receive input information, the hidden layer is used to self-adjust internal features, and the output layer outputs the first fault diagnosis result. The number of nodes of the neurons in the output layer can be set to 4; in actual implementation, the time domain data can be sent to the first neural network model, and the first neural network model can output the first fault diagnosis result according to the time domain data.

[0025] Step S108, inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; The above-mentioned second neural network model can be a DBN (Deep Belief Network) model, which is a deep neural network based on unsupervised learning and has powerful feature extraction and expression capabilities. The DBN model is stacked by RBM (Restricted Boltzmann Machine, a generative random artificial neural network). RBM includes a visible layer and a hidden layer. The visible layer is used to receive the input data of the upper layer, and the hidden layer learns the internal features of the feature matrix and inputs it to the next layer. Finally, the second fault diagnosis result is output through the fully connected layer. The fully connected layer can be set to have 4 output nodes. In actual implementation, the above-mentioned frequency domain data can be sent to the second neural network model, and the second neural network model can output the second fault diagnosis result according to the frequency domain data.

[0026] Step S110, inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; The third neural network model can be a CNN (Convolutional Neural Network) model, which is a deep learning model specifically used to process data with a grid structure. The CNN model consists of a convolution layer, a pooling layer, a fully connected layer, and a classification layer. The convolution layer performs a convolution operation with the data to generate a feature map, wherein the first convolution layer is used to receive the initial data, the fully connected layer is connected to the back of the convolution layer for downsampling, and the fully connected layer is in front of the output layer to expand the final feature map. The classification layer outputs the third fault diagnosis result, and the number of output nodes can be set to 4. In actual implementation, the normalized data can be sent to the third neural network model, and the third neural network model can output the third fault diagnosis result based on the normalized data.

[0027] Step S112, determining a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result.

[0028] The above-mentioned first fault diagnosis result, second fault diagnosis result and third fault diagnosis result may be the same or different, and it is necessary to combine these three fault diagnosis results to determine the final target fault diagnosis result; for example, if the actual fault is a motor fault, after processing in the above manner, the first fault diagnosis result and the second fault diagnosis result are both motor faults, and the third fault diagnosis result is a reducer fault. In this case, by combining the three fault diagnosis results, it can usually be determined that the final target fault diagnosis result is a motor fault, etc.

[0029] The above-mentioned fault diagnosis method of the electric drive axle collects the vibration signal of the target component through the target vibration sensor; wherein the target component includes: the motor, the reducer and the differential in the electric drive axle; the vibration signal is processed to output the target data; wherein the target data includes: the time domain data, the frequency domain data and the normalized data corresponding to the vibration signal; the time domain data is input into the pre-trained first neural network model to output the first fault diagnosis result through the first neural network model; the frequency domain data is input into the pre-trained second neural network model to output the second fault diagnosis result through the second neural network model; the normalized data is input into the pre-trained third neural network model to output the third fault diagnosis result through the third neural network model; based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result, the final target fault diagnosis result is determined. This method can perform fault diagnosis in three different dimensions of time domain data, frequency domain data and normalized data through multiple neural network models, and determine the final target fault diagnosis result by combining the three fault diagnosis results, without relying on previous experience or selecting a threshold range, thereby improving the efficiency and accuracy of fault identification and saving time and economic costs.

[0030] The embodiment of the present invention further provides another electric drive axle fault diagnosis method, which is implemented on the basis of the method in the above embodiment, and comprises the following steps: Step 1: collecting vibration signals of target components through target vibration sensors; wherein the target components include: a motor, a reducer and a differential in an electric drive axle; The target vibration sensor is a three-way vibration sensor; the three-way vibration sensor can simultaneously obtain signals in the axial direction, the vertical direction and the horizontal direction: specifically, the vibration signal includes: a first signal in the axial direction, a second signal in the vertical direction and a third signal in the horizontal direction.

[0031] The target vibration sensor includes: a first vibration sensor, a second vibration sensor and a third vibration sensor; the first vibration sensor is arranged on the housing of the motor to collect the vibration signal of the motor; the second vibration sensor is arranged at the bearing end of the reducer to collect the vibration signal of the reducer; the third vibration sensor is arranged on the housing of the differential to collect the vibration signal of the differential.

[0032] like Figure 2 The schematic diagram of the structure of an electric drive axle shown in the figure mainly comprises a motor, a main shaft 1, a main shaft 2, a reducer, a differential, a half shaft 1, a half shaft 2, a wheel hub, a gear and a bridge housing.

[0033] The motor is the core power component of the electric drive axle, which can convert the electrical energy in the power supply into mechanical energy to drive the vehicle; the motor cooperates with the main shaft 1. The reducer increases the torque by reducing the speed, and the reducer cooperates with the gears between the main shaft 2 and the main shaft 1 to receive the force and torque from the motor. The differential can make the left and right wheels rotate at different speeds, improving the stability of wheel operation during turning. One end of the differential is connected to the half shaft 1, and the other end is connected to the half shaft 2. The half shaft 1 is connected to the reducer through gears. The half shaft 1 and the half shaft 2 are connected through the bevel gear of the reducer. The two half shafts drive the two wheel hubs respectively. The bridge housing surrounds the reducer, differential, half shafts and other related components, mainly used to avoid interference from the external environment and provide structural support.

[0034] The following is an explanation of how to select the target vibration sensor. Assuming that the maximum speed of the given motor is 12000rpm, the gear ratio of gear 1 to gear 2 is 1:2; the gear ratio of gear 3 to gear 4 is 1:2; since spindle 1 is connected to the motor, the maximum speed of spindle 1 is 12000rpm. According to the transmission ratio formula , the maximum speed of spindle 2 is 6000rpm, where the number of teeth 1 / the number of teeth 2 corresponds to the gear ratio of gear 1 to gear 2, the output shaft 2 corresponds to the maximum speed of spindle 2, and the output shaft 1 corresponds to the maximum speed of spindle 1; similarly, according to the maximum speed of spindle 2 and the gear ratio of gear 3 to gear 4, the maximum speed of half shaft 1 is 3000rpm. According to the frequency formula , where n is the maximum speed; the rotational frequencies of spindle 1, spindle 2 and half-shaft 1 are 200 Hz, 100 Hz and 50 Hz respectively. Therefore, it can be determined that the maximum speed of spindle 1, spindle 2 and half-shaft 1 is 12000 rpm and the maximum rotational frequency is 200 Hz.

[0035] According to the Nyquist sampling theorem, the sampling frequency of the target vibration sensor should be greater than twice the highest frequency of the sampled signal. At the same time, in order to prevent the target vibration sensor from being slightly offset due to vibration during the sampling process and to capture more details, after comprehensive consideration, the sampling frequency of the target vibration sensor can be selected as 4000Hz, and the layout method is glue bonding. In practical applications, the appropriate connection method can be selected according to the sampling frequency, for example, adsorption connection, glue bonding or bolt connection can be used. Generally, the higher the sampling frequency of the target vibration sensor, the more reliable the connection method needs to be.

[0036] like Figure 3The schematic diagram of the position of a target vibration sensor shown in the figure can determine the layout position of the target vibration sensor on the bridge housing according to the maximum force direction of the key components of the electric drive bridge and the interference of the signal collection, that is, the position with the largest vibration amplitude can be found. This position usually corresponds to the position with the maximum force, and the collected vibration signal is relatively more accurate. Specifically, Figure 3 As shown, the first vibration sensor 1 is arranged on the housing of the motor and close to the output shaft to prevent the motor from interfering with the gearbox during the transmission process; the second vibration sensor 2 is arranged on the bearing end of the reducer; and the third vibration sensor 3 is arranged on the housing of the differential close to the half shaft 1. The first vibration sensor, the second vibration sensor and the third vibration sensor are all three-way vibration sensors, which can simultaneously obtain the first axial signal of the motor, the reducer and the differential, the second signal in the vertical direction and the third signal in the horizontal direction. The first signal, the second signal and the third signal obtained are usually analog quantities, and the analog quantities can be converted into digital signals through A / D conversion, and the converted digital signals are saved in the database in the form of vibration signal files, wherein each vibration signal file contains three channels, and the three channels are arranged in columns, the first channel stores the second signal in the vertical direction, the second channel stores the third signal in the horizontal direction, and the third channel stores the first signal in the axial direction.

[0037] Step 2: performing noise reduction processing on the vibration signal to obtain a noise-reduced vibration signal; Specifically, the signal of each channel of the vibration signal file can be decomposed by discrete wavelet change to obtain wavelet coefficients of different scales. Among them, the wavelet basis function used for decomposition selects Daubechies wavelet, and the number of decomposition layers is generally 3 to 5 layers. Next, the wavelet coefficients need to be threshold processed to remove the noise. The threshold selection adopts the VisuShrink method. The threshold processing selects the hard threshold processing method. If the absolute value of the parameter in the wavelet coefficient is less than the threshold, it is set to 0, otherwise it is retained. Finally, the wavelet coefficients after threshold processing are inversely transformed, the denoised signal is reconstructed and replaced to the original position to obtain the denoised vibration signal. The denoising processing method can refer to the relevant technology and will not be repeated here.

[0038] Step 3, normalizing the vibration signal after noise reduction to obtain normalized data; The signal of each channel after noise reduction is normalized, and the normalization formula is: in, represents the channel; Represents the timing The corresponding amplitude value below; Represents the minimum amplitude of the channel; Represents the maximum amplitude of the channel. The signal of each channel is normalized and replaced to the original position to obtain normalized data.

[0039] Step 4: Segment the normalized data to obtain multiple data segments; The normalized data is segmented, and the length of each data segment can be Determine; among them, is the sampling frequency, is the rotation speed. The sampling frequency is 4000Hz (i.e., the sampling frequency of the target vibration sensor determined above), the minimum rotation speed is 3000rpm (i.e., the minimum of the maximum rotation speeds of the main shaft 1, main shaft 2, and semi-shaft 1 above), and N is 80 timing points. In practical applications, in order to meet the number of sampling points in a single cycle, the number of timing points actually set usually needs to be greater than N. For example, the length of each data segment can be set to 100 timing points, and then the vibration signal file can be divided into equal lengths of 100 timing points. The dimension of each data segment after segmentation is 3 columns and 100 rows.

[0040] Step 5: for each data segment, perform time domain processing on the data in the data segment according to a preset time domain processing method to obtain time domain data corresponding to the data segment; wherein the time domain processing method includes: mean calculation, variance calculation, absolute average calculation, standard deviation calculation, root mean square value calculation, peak value calculation and minimum value calculation; For each data segment, the above mean calculation is to add and average the data in the data segment; the above variance calculation is to calculate the average of the squares of the differences between each data and the mean in the data segment, which is used to describe the degree of dispersion of the data distribution; the above absolute mean calculation is to calculate the average of the absolute values ​​of the differences between each data and the mean in the data segment; the above standard deviation calculation can be used to measure the degree of dispersion of the data in the data segment, which indicates the degree of deviation between the data in the data segment and the mean; the above root mean square value calculation is to sum the squares of each data in the data segment and then take the square root of the average value; the above peak value calculation is to obtain the maximum value in the data segment; the above minimum value calculation is to obtain the minimum value in the data segment. In actual implementation, time domain processing can be performed for each data segment according to the above time domain processing method to obtain the time domain data corresponding to the data segment. For example, the dimension of the data segment before processing is 3 columns and 100 rows, and the dimension of the data segment after time domain processing is 3 columns and 7 rows.

[0041] Step 6, based on the time domain data corresponding to each data segment, obtaining the time domain data corresponding to the vibration signal; The time domain data corresponding to each data segment is summarized to obtain the time domain data corresponding to the vibration signal.

[0042] Step 7: for each data segment, perform frequency domain processing on the data in the data segment according to a preset frequency domain processing method to obtain frequency domain data corresponding to the data segment; wherein the frequency domain processing method includes: centroid frequency calculation, mean square frequency calculation, root mean square frequency calculation, frequency variance calculation and frequency standard deviation calculation; For each data segment, the centroid frequency is the frequency of the data component with a larger component in the frequency spectrum, reflecting the distribution of the data power spectrum; the mean square frequency is the weighted average of the square of the data frequency; the root mean square frequency is the arithmetic square root of the mean square frequency, which represents the square root of the weighted average of the square of the data frequency; the frequency variance is to calculate the square of the difference between each data and the square number, and then average these square values; the frequency standard deviation can represent the degree of frequency dispersion, that is, the square root of the average sum of the squares of the deviations between the actual frequency and the average frequency. In actual implementation, frequency domain processing can be performed on each data segment according to the above frequency domain processing method to obtain the frequency domain data corresponding to the data segment. For example, the dimension of the data segment before processing is 3 columns and 100 rows, and the dimension of the data segment after time domain processing is 3 columns and 5 rows.

[0043] Step eight, based on the frequency domain data corresponding to each data segment, obtain the frequency domain data corresponding to the vibration signal.

[0044] The frequency domain data corresponding to each data segment is summarized to obtain the frequency domain data corresponding to the vibration signal.

[0045] Step nine, inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; Step ten, inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; Step 11, inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; The first neural network model is configured with a first weight, the second neural network model is configured with a second weight, and the third neural network model is configured with a third weight; wherein the first weight is equal to the second weight; the third weight is higher than the first weight and higher than the second weight; and the sum of the first weight and the second weight is greater than the third weight.

[0046] The above-mentioned first weight, second weight and third weight can be set according to actual needs. Usually, the data of the third weight will be higher than the first weight and the second weight, and the sum of the first weight and the second weight will usually be higher than the third weight; for example, the first weight can be set to 0.3, the second weight can be set to 0.3, the third weight can be set to 0.4, etc.

[0047] Step 12: if the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all the same, determine the first designated fault diagnosis result as the final target fault diagnosis result; wherein the first designated fault diagnosis result is: the first fault diagnosis result, the second fault diagnosis result or the third fault diagnosis result; In actual implementation, after obtaining the above-mentioned first fault diagnosis result, second fault diagnosis result and third fault diagnosis result, a comprehensive judgment can be made based on the first weight corresponding to the first neural network model, the second weight corresponding to the second neural network model, and the third weight corresponding to the third neural network model to determine the final target fault diagnosis result; specifically, if the first fault diagnosis result output by the first neural network model, the second fault diagnosis result output by the second neural network model, and the third fault diagnosis result output by the third neural network model are of the same type, for example, all are motor faults, then the comprehensive result is the common result of the three, that is, the final target fault diagnosis result is a motor fault.

[0048] Step thirteen, if the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all different, determining the third fault diagnosis result corresponding to the third neural network model configured with the third weight as the final target fault diagnosis result; If the first fault diagnosis result output by the first neural network model, the second fault diagnosis result output by the second neural network model, and the third fault diagnosis result output by the third neural network model are all different, since the third weight corresponding to the third neural network model is the largest, the third fault diagnosis result output by the third neural network model can be determined as the final target fault diagnosis result. In actual implementation, the fault type corresponding to the output result of the neural network can be assigned 1, and the other results can be assigned 0. For example, if the output result of the first neural network model is a motor fault, the corresponding diagnosis result of the motor is 1, and the corresponding diagnosis results of the reducer and differential are 0. After the first weight assignment calculation, the motor fault assignment is 0.3; the output result of the second neural network model is a reducer fault, then the corresponding diagnosis result of the reducer is 1, and the rest are 0. After the second weight assignment calculation, the reducer fault assignment is 0.3; the output result of the third neural network model is a differential fault, and similarly, the differential fault assignment is 0.4; the assignment results of the three neural network models are comprehensively compared, and the differential fault assignment is the largest, then the final target diagnosis result is determined to be a differential fault.

[0049] Step 14: if the first fault diagnosis result is the same as the second fault diagnosis result and different from the third fault diagnosis result, determine the second designated fault diagnosis result as the final target fault diagnosis result; wherein the second designated fault diagnosis result is: the first fault diagnosis result and the second fault diagnosis result; If the first fault diagnosis result output by the first neural network model is the same as the second fault diagnosis result output by the second neural network model, but different from the third fault diagnosis result output by the third neural network model, since the sum of the first weight corresponding to the first neural network model and the second weight corresponding to the second neural network model is greater than the third weight corresponding to the third neural network model, the comprehensive result can be determined based on the first fault diagnosis result and the second fault diagnosis result, that is, the first fault diagnosis result or the second fault diagnosis result can be used to determine the final target fault diagnosis result.

[0050] Step fifteen, if the first fault diagnosis result is different from the second fault diagnosis result, and the third designated fault diagnosis result is the same as the third fault diagnosis result, determine the third fault diagnosis result as the final target fault diagnosis result; wherein the third designated fault diagnosis result is the first fault diagnosis result or the second fault diagnosis result.

[0051] If the first fault diagnosis result is different from the second fault diagnosis result, but one of them is the same as the third fault diagnosis result, since the third weight corresponding to the third neural network model is the highest, the final target fault diagnosis result can be the third fault diagnosis result.

[0052] The above-mentioned fault diagnosis method of the electric drive axle arranges a three-way vibration sensor on the surface of the bridge housing of the electric drive axle to collect vibration signals of key components; processes the acquired vibration signals to reduce the learning difficulty of subsequent multiple types of neural network models; establishes multiple types of neural network models through artificial neural networks, machine learning and deep learning, namely, a first neural network model, a second neural network model and a third neural network model, and judges the fault of the electric drive axle through multiple types of fault diagnosis models. If there is no fault, it works normally, otherwise the specific fault type is output.

[0053] Whether the electric drive axle is faulty can be comprehensively judged based on the fault diagnosis results output by the three neural network models. If it is determined that the electric drive axle is not faulty, it can be considered that the electric drive axle is operating normally. If it is determined that the electric drive axle is faulty, the final target fault diagnosis result can be determined in the above manner. For example, the corresponding fault type of the output is a motor fault, a reducer fault or a differential fault.

[0054] For easier understanding, see Figure 4 The flowchart of a fault diagnosis method for an electric drive axle shown in the figure first arranges the target vibration sensor of the electric drive axle, collects the vibration signal, processes the vibration signal, establishes multiple types of neural network models, and outputs the target fault diagnosis result of the electric drive axle. If there is no fault, it means that the electric drive axle is working normally. If there is a fault, the motor fault, reducer fault or differential fault can be output.

[0055] With the rapid development of science and technology since the last century, a large number of computer technologies have emerged. Representative ones include artificial intelligence networks, machine learning, deep learning, etc. These technologies have the characteristics of strong adaptive learning ability, wide application fields, and good generalization performance, so they are widely used in various fields. However, they require huge amounts of data for feature mining to raise the upper limit of technology. In summary, this scheme attempts to realize the fault diagnosis of the electric drive axle based on data-driven and artificial intelligence, so as to quickly and accurately determine the location of the fault, reduce dependence on manual experience, improve diagnostic recognition efficiency and accuracy, and save time and economic costs.

[0056] The electric drive axle integrates key components such as the motor, reducer and differential. In related technologies, whenever a fault occurs, it is difficult for maintenance personnel to accurately and quickly troubleshoot the fault due to the high degree of integration of this structure. This solution collects vibration signals and diagnoses based on the vibration signals, reducing the frequency of disassembly and troubleshooting, thereby improving maintenance efficiency.

[0057] This solution establishes multiple types of neural network models, which can perform fault judgment in three different dimensions: normalized data, time domain data, and frequency domain data, avoiding the absoluteness of a single neural network model. It also avoids the limitations of the existing technology of using threshold range division to judge the fault of the electric drive axle, reducing interference from human subjectivity.

[0058] In addition, this solution is based on data-driven and combines artificial intelligence technology to propose a fault diagnosis method for the electric drive axle, which can complete the judgment of faulty components without disassembling the electric drive axle, thereby saving manpower, time and economic costs.

[0059] The embodiment of the present invention provides a training method for a neural network model, such as Figure 5 As shown, the method includes: Step S502, obtaining a preset initial data set; wherein the initial data set includes: a first subset, a second subset, a third subset and a fourth subset; the first subset stores normal vibration data corresponding to the electric drive axle in a normal working state; the second subset stores motor fault vibration data corresponding to a motor fault; the third subset stores reducer fault vibration data corresponding to a reducer fault; the fourth subset stores differential fault vibration data generated under a differential fault; In this embodiment, the first vibration sensor, the second vibration sensor and the third vibration sensor arranged on the electric drive axle can simultaneously obtain the first axial signal, the second vertical signal and the third horizontal signal of the motor, the reducer and the differential. The first signal, the second signal and the third signal obtained are usually analog quantities, and the analog quantities are converted into digital signals through A / D conversion. The digital signals are summarized into a database, which includes four major categories: Class A (corresponding to the above-mentioned first subset), Class B (corresponding to the above-mentioned second subset), Class C (corresponding to the above-mentioned third subset) and Class D (corresponding to the above-mentioned fourth subset). Class A stores vibration signal files under normal working conditions, Class B stores vibration signal files generated under motor failure, Class C stores vibration signal files generated under reducer failure, and Class D stores vibration signal files generated under differential failure. Among them, the vibration signal files stored in Class A should contain vibration signals generated under normal working conditions of the motor, reducer and differential. Each vibration signal file in the database contains three channels, which are arranged in columns. The first channel stores the second signal in the vertical direction, the second channel stores the third signal in the horizontal direction, and the third channel stores the first signal in the axial direction.

[0060] Step S504, processing the initial data set to obtain a time domain feature set, a frequency domain feature set and a normalized feature set corresponding to the initial data set; In actual implementation, the vibration signal files in the above database can be subjected to wavelet threshold denoising to reduce the interference of noise; the data of each channel after denoising is normalized to prevent extreme points from interfering with the subsequent neural network model; the normalized signal is subjected to data slicing operation, and all the data fragments obtained after slicing can be summarized into a normalized feature set, which is the same as the categories contained in the original database, that is, it contains four major categories: normal, motor failure, reducer failure and differential failure. After time domain processing of each data fragment, it is summarized into a time domain feature set, which also contains four major categories: normal, motor failure, reducer failure and differential failure. After frequency domain processing of each data fragment, it is summarized into a frequency domain feature set, which also contains four major categories: normal, motor failure, reducer failure and differential failure.

[0061] Step S506, determining a training set, a validation set, and a test set based on the target feature set; wherein the target feature set is: a time domain feature set, a frequency domain feature set, and a normalized feature set; Step S508, training the initial network model based on the training set to obtain a first model; Step S510, optimize the first model based on the validation set to obtain a trained target neural network model, and evaluate the target neural network model through the test set; wherein, if the target feature set is a time domain feature set, the target neural network model is a first neural network model; if the target feature set is a frequency domain feature set, the target neural network model is a second neural network model; if the target feature set is a normalized feature set, the target neural network model is a third neural network model.

[0062] like Figure 6 A schematic diagram of a processing method of a neural network model is shown. In actual implementation, the time domain feature set can be divided into a training set, a validation set, and a test set at a ratio of 6:2:2; the frequency domain feature set can be divided into a training set, a validation set, and a test set at a ratio of 6:2:2; the normalized feature set can be divided into a training set, a validation set, and a test set at a ratio of 6:2:2.

[0063] The training set is used to train the corresponding neural network model and determine the parameters such as weights and biases within the model; the validation set is used to verify the diagnostic performance of the corresponding neural network model, adjust the parameters and hyperparameters within the model, and optimize the model performance; the test set is used to evaluate the final performance of the corresponding neural network model. The training set, validation set, and test set all contain four types of samples: normal electric drive axle, motor failure, reducer failure, and differential failure.

[0064] like Figure 7 The schematic diagram of another processing method of a neural network model shown is to input a training set of a time domain feature set into a first neural network model (such as an ANN model) to determine the parameters within the model, and then input a validation set into the trained first neural network model to optimize and adjust the internal parameters, and finally input a test set into the adjusted first neural network model for evaluation and output of a diagnostic result.

[0065] The training set of the frequency domain feature set is input into the second neural network model (such as the DBN model) to determine the parameters inside the model. Then the validation set is input into the trained second neural network model to optimize and adjust the internal parameters. Finally, the test set is input into the adjusted second neural network model for evaluation and output of the diagnostic results.

[0066] The training set of the normalized feature set is input into a third neural network model (such as a CNN model) to determine the parameters within the model. The validation set is then input into the trained third neural network model to optimize and adjust the internal parameters. Finally, the test set is input into the adjusted third neural network model for evaluation and output of the diagnostic results.

[0067] In the related art, the electric drive axle has a complex structure, the signal collection of each component is difficult, and the collected signals are prone to interfere with each other. This solution processes the collected signals in the time domain and frequency domain to expand the fault information unique to each component, thereby improving the prediction accuracy of various neural network models.

[0068] In addition, related technologies collect and process the signals of the electric drive axle, and then set a certain range of thresholds to determine whether a fault has occurred. The division of the threshold range of this type of technology is uncertain, and it is impossible to make an objective evaluation based on past experience. This solution uses the collected signals as the data basis, and processes them to reduce mutual interference. At the same time, it uses artificial intelligence technology to establish a variety of neural network models for adaptive fault feature learning to complete the fault diagnosis of the electric drive axle, thereby achieving a fast and accurate diagnosis effect, and reducing the dependence on previous experience and the influence of human subjectivity.

[0069] In the above-mentioned fault diagnosis method of the electric drive axle, the target vibration sensor adopts a three-way vibration sensor, which reduces the complexity of the target vibration sensor in the layout process. The first signal in the axial direction, the second signal in the vertical direction and the third signal in the horizontal direction of each key component of the electric drive axle can be obtained simultaneously through a single target vibration sensor, and the collected signal is converted into a digital signal by A / D, which is convenient for storage into a historical period database, providing data support for the subsequent various types of neural network models, and effectively improving the accuracy of fault diagnosis.

[0070] This solution processes the converted digital signal, and uses wavelet threshold denoising and normalization processing to reduce the interference of noise on the vibration signal during the acquisition process; then, the time domain analysis and frequency domain analysis are used to extract the characteristics of the vibration signal in the time domain and frequency domain respectively, which helps to improve the efficiency of various types of neural network models in the adaptive feature learning process.

[0071] The embodiment of the present invention provides a fault diagnosis device for an electric drive axle, such as Figure 8 As shown, the device comprises: The acquisition module 80 is used to acquire the vibration signal of the target component through the target vibration sensor; wherein the target component includes: the motor, the reducer and the differential in the electric drive axle; The first processing module 81 is used to process the vibration signal and output target data; wherein the target data includes: time domain data, frequency domain data and normalized data corresponding to the vibration signal; A first output module 82, used for inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; A second output module 83, used for inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; A third output module 84, used for inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; The first determination module 85 is used to determine a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result.

[0072] The fault diagnosis device of the above-mentioned electric drive axle can perform fault diagnosis in three different dimensions of time domain data, frequency domain data and normalized data through a variety of neural network models, and determine the final target fault diagnosis result by combining the three fault diagnosis results. It does not need to rely on previous experience or select a threshold range, thereby improving fault identification efficiency and accuracy and saving time and economic costs.

[0073] Furthermore, the target vibration sensor is a three-way vibration sensor; the vibration signal includes: a first signal in the axial direction, a second signal in the vertical direction, and a third signal in the horizontal direction.

[0074] Furthermore, the target vibration sensor includes: a first vibration sensor, a second vibration sensor and a third vibration sensor; the first vibration sensor is arranged on the housing of the motor to collect the vibration signal of the motor; the second vibration sensor is arranged at the bearing end of the reducer to collect the vibration signal of the reducer; the third vibration sensor is arranged on the housing of the differential to collect the vibration signal of the differential.

[0075] Furthermore, the first processing module is also used to: perform noise reduction processing on the vibration signal to obtain a noise-reduced vibration signal; perform normalization processing on the noise-reduced vibration signal to obtain normalized data; perform segmentation processing on the normalized data to obtain multiple data segments; for each data segment, perform time domain processing on the data in the data segment according to a preset time domain processing method to obtain time domain data corresponding to the data segment; wherein the time domain processing method includes: mean calculation, variance calculation, absolute average calculation, standard deviation calculation, root mean square value calculation, peak value calculation and minimum value calculation; based on the time domain data corresponding to each data segment, obtain the time domain data corresponding to the vibration signal; for each data segment, perform frequency domain processing on the data in the data segment according to a preset frequency domain processing method to obtain the frequency domain data corresponding to the data segment; wherein the frequency domain processing method includes: center of gravity frequency calculation, mean square frequency calculation, root mean square frequency calculation, frequency variance calculation and frequency standard deviation calculation; based on the frequency domain data corresponding to each data segment, obtain the frequency domain data corresponding to the vibration signal.

[0076] Furthermore, the first neural network model is configured with a first weight, the second neural network model is configured with a second weight, and the third neural network model is configured with a third weight; wherein the first weight is equal to the second weight; the third weight is higher than the first weight and higher than the second weight; and the sum of the first weight and the second weight is greater than the third weight.

[0077] Further, the first determination module is also used for: if the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all the same, determining the first designated fault diagnosis result as the final target fault diagnosis result; wherein, the first designated fault diagnosis result is: the first fault diagnosis result, the second fault diagnosis result or the third fault diagnosis result; if the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all different, determining the third fault diagnosis result corresponding to the third neural network model configured with a third weight as the final target fault diagnosis result; if the first fault diagnosis result is the same as the second fault diagnosis result and different from the third fault diagnosis result, determining the second designated fault diagnosis result as the final target fault diagnosis result; wherein, the second designated fault diagnosis result is: the first fault diagnosis result and the second fault diagnosis result; if the first fault diagnosis result is different from the second fault diagnosis result and the third designated fault diagnosis result is the same as the third fault diagnosis result, determining the third fault diagnosis result as the final target fault diagnosis result; wherein, the third designated fault diagnosis result is the first fault diagnosis result or the second fault diagnosis result.

[0078] The fault diagnosis device for an electric drive axle provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned electric drive axle fault diagnosis method embodiment. For the sake of brief description, for matters not mentioned in the embodiment of the fault diagnosis device for an electric drive axle, reference may be made to the corresponding contents in the aforementioned electric drive axle fault diagnosis method embodiment.

[0079] The embodiment of the present invention provides a training device for a neural network model, such as Fig. 9 As shown, the device comprises: The acquisition module 90 is used to acquire a preset initial data set; wherein the initial data set includes: a first subset, a second subset, a third subset and a fourth subset; the first subset stores normal vibration data corresponding to the electric drive axle in a normal working state; the second subset stores motor fault vibration data corresponding to a motor fault; the third subset stores reducer fault vibration data corresponding to a reducer fault; the fourth subset stores differential fault vibration data generated under a differential fault; The second processing module 91 is used to process the initial data set to obtain a time domain feature set, a frequency domain feature set and a normalized feature set corresponding to the initial data set; A second determination module 92 is used to determine a training set, a validation set and a test set based on a target feature set; wherein the target feature set is: a time domain feature set, a frequency domain feature set and a normalized feature set; A training module 93 is used to train an initial network model based on a training set to obtain a first model; The optimization module 94 is used to optimize the first model based on the validation set to obtain a trained target neural network model, and evaluate the target neural network model through the test set; wherein, if the target feature set is a time domain feature set, the target neural network model is the first neural network model; if the target feature set is a frequency domain feature set, the target neural network model is the second neural network model; if the target feature set is a normalized feature set, the target neural network model is the third neural network model.

[0080] The training device of the above neural network model processes the collected signals in the time domain and frequency domain to expand the fault information unique to each component, thereby improving the prediction accuracy of various types of neural network models.

[0081] The embodiment of the present invention further provides an electronic device, see Fig.10 As shown, the electronic device includes a processor 130 and a memory 131, wherein the memory 131 stores machine executable instructions that can be executed by the processor 130, and the processor 130 executes the machine executable instructions to implement the above-mentioned electric drive axle fault diagnosis method and neural network model training method.

[0082] Further, Fig.10 The electronic device shown further includes a bus 132 and a communication interface 133 , and the processor 130 , the communication interface 133 and the memory 131 are connected via the bus 132 .

[0083] The memory 131 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 133 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 132 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0084] The processor 130 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 130. The above processor 130 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 131, and the processor 130 reads the information in the memory 131 and completes the steps of the method of the above embodiment in combination with its hardware.

[0085] An embodiment of the present invention also provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the above-mentioned electric drive axle fault diagnosis method and neural network model training method. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0086] The computer program product of the electric drive axle fault diagnosis method and the neural network model training method provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be found in the method embodiments, which will not be repeated here.

[0087] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault diagnosis method for an electric drive axle, characterized in that: The method comprises: The target vibration sensor is used to collect vibration signals of the target component; wherein the target component includes: the motor, the reducer and the differential in the electric drive axle; Processing the vibration signal and outputting target data; wherein the target data includes: time domain data, frequency domain data and normalized data corresponding to the vibration signal; Inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; Inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; Inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; A final target fault diagnosis result is determined based on the first fault diagnosis result, the second fault diagnosis result, and the third fault diagnosis result.

2. The method according to claim 1, characterized in that The target vibration sensor is a three-way vibration sensor; the vibration signal includes: a first signal in the axial direction, a second signal in the vertical direction, and a third signal in the horizontal direction.

3. The method according to claim 1, characterized in that The target vibration sensor includes: a first vibration sensor, a second vibration sensor and a third vibration sensor; The first vibration sensor is arranged on the housing of the motor and is used to collect a vibration signal of the motor; The second vibration sensor is arranged at the bearing end of the reducer and is used to collect the vibration signal of the reducer; The third vibration sensor is arranged on the housing of the differential and is used to collect the vibration signal of the differential.

4. The method according to claim 1, characterized in that: The step of processing the vibration signal and outputting target data comprises: Performing noise reduction processing on the vibration signal to obtain a noise-reduced vibration signal; Normalizing the vibration signal after noise reduction to obtain the normalized data; Slice the normalized data to obtain multiple data segments; For each of the data segments, time domain processing is performed on the data in the data segment according to a preset time domain processing method to obtain time domain data corresponding to the data segment; wherein the time domain processing method includes: mean calculation, variance calculation, absolute average value calculation, standard deviation calculation, root mean square value calculation, peak value calculation and minimum value calculation; Based on the time domain data corresponding to each of the data segments, obtaining the time domain data corresponding to the vibration signal; For each of the data segments, frequency domain processing is performed on the data in the data segment according to a preset frequency domain processing method to obtain frequency domain data corresponding to the data segment; wherein the frequency domain processing method includes: centroid frequency calculation, mean square frequency calculation, root mean square frequency calculation, frequency variance calculation and frequency standard deviation calculation; Based on the frequency domain data corresponding to each of the data segments, the frequency domain data corresponding to the vibration signal is obtained.

5. The method according to claim 1, characterized in that The first neural network model is configured with a first weight, the second neural network model is configured with a second weight, and the third neural network model is configured with a third weight; wherein the first weight is equal to the second weight; the third weight is higher than the first weight and higher than the second weight; and the sum of the first weight and the second weight is greater than the third weight.

6. The method according to claim 5, characterized in that The step of determining a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result comprises: If the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all the same, determining the first designated fault diagnosis result as the final target fault diagnosis result; wherein the first designated fault diagnosis result is: the first fault diagnosis result, the second fault diagnosis result or the third fault diagnosis result; If the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result are all different, determining the third fault diagnosis result corresponding to the third neural network model configured with the third weight as the final target fault diagnosis result; If the first fault diagnosis result is the same as the second fault diagnosis result and different from the third fault diagnosis result, determining the second designated fault diagnosis result as the final target fault diagnosis result; wherein the second designated fault diagnosis result is: the first fault diagnosis result and the second fault diagnosis result; If the first fault diagnosis result is different from the second fault diagnosis result, and the third designated fault diagnosis result is the same as the third fault diagnosis result, the third fault diagnosis result is determined as the final target fault diagnosis result; wherein the third designated fault diagnosis result is the first fault diagnosis result or the second fault diagnosis result.

7. A training method for a neural network model, characterized in that: The method comprises: Acquire a preset initial data set; wherein the initial data set includes: a first subset, a second subset, a third subset and a fourth subset; the first subset stores normal vibration data corresponding to the electric drive axle in a normal working state; the second subset stores motor fault vibration data corresponding to a motor fault; the third subset stores reducer fault vibration data corresponding to a reducer fault; the fourth subset stores differential fault vibration data generated under a differential fault; Processing the initial data set to obtain a time domain feature set, a frequency domain feature set, and a normalized feature set corresponding to the initial data set; Based on the target feature set, a training set, a validation set and a test set are determined; wherein the target feature set is: the time domain feature set, the frequency domain feature set and the normalized feature set; Training an initial network model based on the training set to obtain a first model; The first model is optimized based on the validation set to obtain a trained target neural network model, and the target neural network model is evaluated by the test set; wherein, if the target feature set is a time domain feature set, the target neural network model is a first neural network model; if the target feature set is a frequency domain feature set, the target neural network model is a second neural network model; if the target feature set is a normalized feature set, the target neural network model is a third neural network model.

8. A fault diagnosis device for an electric drive axle, characterized in that: The device comprises: A collection module, used for collecting vibration signals of a target component through a target vibration sensor; wherein the target component includes: a motor, a reducer and a differential in the electric drive axle; A first processing module, used to process the vibration signal and output target data; wherein the target data includes: time domain data, frequency domain data and normalized data corresponding to the vibration signal; A first output module, used for inputting the time domain data into a pre-trained first neural network model to output a first fault diagnosis result through the first neural network model; A second output module, used for inputting the frequency domain data into a pre-trained second neural network model to output a second fault diagnosis result through the second neural network model; A third output module, used for inputting the normalized data into a pre-trained third neural network model to output a third fault diagnosis result through the third neural network model; The first determination module is used to determine a final target fault diagnosis result based on the first fault diagnosis result, the second fault diagnosis result and the third fault diagnosis result.

9. A training device for a neural network model, characterized in that: The device comprises: An acquisition module is used to acquire a preset initial data set; wherein the initial data set includes: a first subset, a second subset, a third subset and a fourth subset; the first subset stores normal vibration data corresponding to the electric drive axle in a normal working state; the second subset stores motor fault vibration data corresponding to a motor fault; the third subset stores reducer fault vibration data corresponding to a reducer fault; the fourth subset stores differential fault vibration data generated under a differential fault; A second processing module is used to process the initial data set to obtain a time domain feature set, a frequency domain feature set and a normalized feature set corresponding to the initial data set; A second determination module is used to determine a training set, a validation set and a test set based on a target feature set; wherein the target feature set is: the time domain feature set, the frequency domain feature set and the normalized feature set; A training module, used for training an initial network model based on the training set to obtain a first model; An optimization module is used to optimize the first model based on the validation set to obtain a trained target neural network model, and evaluate the target neural network model through the test set; wherein, if the target feature set is a time domain feature set, the target neural network model is a first neural network model; if the target feature set is a frequency domain feature set, the target neural network model is a second neural network model; if the target feature set is a normalized feature set, the target neural network model is a third neural network model.

10. An electronic device, characterized in that: It includes a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the fault diagnosis method of the electric drive axle described in any one of claims 1 to 6, or the training method of the neural network model described in claim 7.

Images