Gearbox fault diagnosis method based on multi-sensor adaptive fusion

Abstract

The invention discloses a gearbox fault diagnosis method based on multi-sensor adaptive fusion, and the method comprises the following steps: obtaining the monitoring data of a plurality of sensors in a gearbox, and carrying out the adaptive weight fusion processing based on the multi-sensor data of the gearbox, so as to obtain the multi-sensor weight fusion data in different fault modes; and a multi-sensor multi-level data set is constructed based on the weighted fusion data and is used for training a multi-sensor fault diagnosis model so as to carry out fault diagnosis on the gearbox fault. According to the gearbox fault diagnosis method based on multi-sensor adaptive fusion, abnormal values are effectively eliminated, data redundancy is reduced, meanwhile, based on an adaptive weighted data fusion strategy of an optimal weight distribution principle, under the condition of not depending on expert experience, data noise interference is remarkably reduced, and the fault diagnosis efficiency is improved. And the contribution degree of each sensor is effectively balanced.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a gearbox fault diagnosis method based on multi-sensor adaptive fusion. Background Art

[0002] The gearbox is a basic transmission device, which is mainly used to transmit power to generators or other equipment to achieve functions such as acceleration and deceleration, change of direction, torque and clutch. At the same time, it is subjected to complex loads and harsh working environment during the transmission process, so it is very easy to have various faults. If not identified and handled in time, these faults may cause equipment shutdown, production interruption, and even cause major safety accidents. Therefore, the gearbox fault needs to be monitored and diagnosed accurately in real time. The fault diagnosis model is constructed by using the original signal data collected by the sensor to extract features and identify the fault type.

[0003] For example, a gearbox fault intelligent diagnosis method based on multi-sensor fusion is disclosed in the publication number CN202311865687.0. The original time series signals of multiple measuring points and multiple types collected by the multi-sensor data acquisition system are subjected to data-level weighted fusion based on the principle that the total correlation energy is proportional to the signal fusion weight, so as to obtain a multi-channel one-dimensional fusion signal; secondly, the time window overlapping slicing method is used to divide the fused signal into data segments of fixed length, and the continuous wavelet transform is used to convert each channel of the fused signal segment into two-dimensional time-frequency graph data; finally, a gearbox fault intelligent diagnosis model based on multi-head spatiotemporal attention and parallel GRU network is constructed, and the gearbox fault intelligent diagnosis model is trained using the two-dimensional time-frequency graph obtained by the conversion, and the gearbox fault classification result is determined using the trained gearbox fault intelligent diagnosis model.

[0004] However, the weight allocation mechanism in this data-level fusion strategy is overly dependent on statistical features such as signal strength, which not only reduces the significance of key information contained in low-energy signals during data fusion, but also often results in large errors in multi-sensor weighted fusion data, reducing data reliability. At the same time, the training data of the fault diagnosis model often only contains fused data, while ignoring the rich fault feature information contained in the original data, resulting in the model showing lower robustness and generalization ability when facing a variety of fault conditions. Summary of the invention

[0005] The purpose of the present invention is to provide a gearbox fault diagnosis method based on multi-sensor adaptive fusion to solve the problems in the current market raised by the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a gearbox fault diagnosis method based on multi-sensor adaptive fusion, comprising the following steps:

[0007] Acquire monitoring data of multiple sensors in the gearbox, and perform adaptive weighted fusion processing based on the multi-sensor data of the gearbox to obtain multi-sensor weighted fusion data under different fault modes;

[0008] A multi-sensor multi-level dataset is constructed based on weighted fusion data and used to train a multi-sensor fault diagnosis model for fault diagnosis of gearbox faults.

[0009] Preferably, the gearbox multi-sensor data acquisition comprises the following steps:

[0010] Collect multi-sensor data of gearbox;

[0011] A1. Multiple sensors are pre-arranged at key positions of the gearbox to synchronously collect gearbox operation data;

[0012] A2. Execute step A1 for different gearbox failure mode operation states;

[0013] Weighted fusion processing of single sensor data;

[0014] B1. Use the consistency test theory based on eccentric distance to pre-process the measurement data of each sensor and eliminate the data with large measurement deviation in the multi-sensor system;

[0015] B2. Eliminate c observations from the calculated eccentricity distances in descending order, and perform single sensor weighted fusion on the remaining nc observations of the kth sensor. The smaller the observation value deviates from the center, the higher the consistency with other observation values, and the larger the corresponding weighting factor;

[0016] B3. After the kth sensor is processed by outlier removal and weighted fusion, the single sensor weighted fusion data of the remaining n–c observations can be obtained;

[0017] B4. Execute steps B1-B3 on the remaining sensor data under the fault mode to obtain multiple single sensor weighted fusion data;

[0018] B5. Execute steps B1-B4 on the operating data collected under different fault modes of the gearbox to obtain single sensor weighted fusion data of the operating data under different fault modes.

[0019] Preferably, after constructing the multi-sensor multi-level data set, the following steps are also included:

[0020] Construct a multi-sensor fault diagnosis model;

[0021] C1. Build a self-attention network, input and output data dimensions;

[0022] C2. Construct 1D-CNN. The input data is the output data of the self-attention network. The output data dimension is 1. The 1D-CNN network structure includes two convolution blocks. Each convolution block is composed of a convolution layer, a batch normalization layer, an activation layer, and a pooling layer in sequence. Finally, it is connected to a fully connected layer and the fault type is output through a Softmax classifier.

[0023] Training multi-sensor fault diagnosis models;

[0024] D1. Divide the gearbox fault diagnosis multi-sensor integrated data set into a training set and a test set in proportion to obtain a training set and a test set;

[0025] D2. Set the batch size and learning rate for model training, use cross entropy as the loss function, input the training set into the self-attention network and 1D-CNN in turn, and update the weights of neurons in different layers of 1D-CNN through error back propagation and gradient descent until the stopping condition is met and the model completes training;

[0026] Gearbox failure diagnosis;

[0027] E1. Input the test set into the final network model established and output the prediction results for each fault category;

[0028] E2. Select the maximum value from the prediction results, use the category corresponding to the maximum value as the category of gearbox fault diagnosis, and output the prediction label;

[0029] E3. Calculate the accuracy index based on the prediction results of the test set samples.

[0030] Preferably, the multi-sensor data adaptive weighted fusion processing specifically includes the following steps:

[0031] F1. Calculate the variance corresponding to the observation value of each sensor, and obtain the total mean square error corresponding to the multi-sensor weighted fusion data according to the multi-sensor weighted fusion data formula;

[0032] F2. Calculate the optimal weight factor corresponding to the kth sensor when the total mean square error is minimized by the Lagrange multiplier method, and obtain the multi-sensor weighted fusion data under the fault mode based on the optimal weight factor;

[0033] F3. Execute step F2 on the operating data collected under different fault modes of the gearbox to obtain multi-sensor weighted fusion data of the operating data under different fault modes.

[0034] Preferably, constructing a multi-sensor multi-level data set specifically includes the following steps:

[0035] G1. Integrate the weighted fusion data of each single sensor under different fault modes with the weighted fusion data of multiple sensors, and use the monitored gearbox fault type as a label;

[0036] G2. Divide the fused data into multiple discrete samples and construct a multi-sensor and multi-level dataset for gearbox fault diagnosis.

[0037] Preferably, the expression of multi-sensor weighted fusion data is as follows:

[0038]

[0039] Among them, W k is the weighting factor of the multi-sensor weighted fusion data of the kth sensor, is the fusion data of multiple sensors, m is the number of sensors, is the single sensor weighted fusion data of the kth sensor when the measurement data sample length is n;

[0040] The mean square error expression of multi-sensor weighted fusion data is as follows:

[0041]

[0042] Among them, W k is the multi-sensor fusion weight of the kth sensor, X is the monitoring value of the sensor, is the true value corresponding to the sensor monitoring value, is the variance of the kth sensor;

[0043] The calculation expression of the optimal weight factor is as follows:

[0044]

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. In the data preprocessing stage, the eccentricity-based method is used to perform consistency preprocessing on the single sensor measurement signal sequence, effectively eliminating outliers and reducing data redundancy. At the same time, the adaptive weighted data fusion strategy based on the optimal weight allocation principle not only significantly reduces data noise interference but also ensures that the contribution of each sensor is effectively balanced without relying on expert experience.

[0047] 2. By integrating the multi-sensor weighted fusion data with the fusion signals of each single sensor and integrating the label information, a multi-sensor multi-level data set for gearbox fault diagnosis is constructed. This can not only complete the integrity of the data information and make up for the missing information of the multi-sensor weighted fusion data, but also effectively eliminate the redundant information in the original signal and improve the data quality and analysis efficiency.

[0048] 3. Obtain corresponding multi-sensor weighted fusion data according to different fault modes, integrate the data of different fault modes through multi-level data sets, and then realize multi-sensor adaptive fusion of gearbox fault diagnosis by building diagnostic models, training diagnostic models and fault diagnosis, which can realize real-time monitoring and accurate diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of the present invention;

[0050] Figure 2 It is a time domain waveform diagram before and after the multi-sensor weighted fusion data under the gear pitting state of the present invention;

[0051] Figure 3 It is a curve diagram of the accuracy of the multi-sensor fault diagnosis model of the present invention;

[0052] Figure 4 It is a confusion matrix diagram of the classification results of the multi-sensor fault diagnosis model of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0054] A gearbox fault diagnosis method based on multi-sensor adaptive fusion, comprising:

[0055] Acquire monitoring data of multiple sensors in the gearbox, and perform adaptive weighted fusion processing based on the multi-sensor data of the gearbox to obtain multi-sensor weighted fusion data under different fault modes;

[0056] A multi-sensor multi-level dataset is constructed based on weighted fusion data and used to train a multi-sensor fault diagnosis model for fault diagnosis of gearbox faults.

[0057] In a specific embodiment, the above diagnostic method comprises the following specific steps:

[0058] S1: Collect gearbox multi-sensor data;

[0059] S2: weighted fusion processing of single sensor data;

[0060] S3: Adaptive weighted fusion processing of multi-sensor data;

[0061] S4: Construct multi-sensor multi-level dataset;

[0062] S5: Construct a multi-sensor fault diagnosis model;

[0063] S6: training multi-sensor fault diagnosis model;

[0064] S7: Perform fault diagnosis on gearbox fault.

[0065] The specific steps are as follows:

[0066] Step 1, collecting gearbox multi-sensor data;

[0067] Step 1: Use m sensors pre-arranged at key positions of the gearbox to synchronously collect gearbox operation data X1, X2, ..., X m ;

[0068] Step 2: Execute step 1 for different fault mode operating states of the gearbox;

[0069] Step 2: weighted fusion processing of single sensor data;

[0070] Step 1: Use the consistency test theory based on eccentric distance to pre-process the measurement data of each sensor and eliminate the data with large measurement deviation in the multi-sensor system. The expression of eccentric distance is as follows:

[0071]

[0072] Among them, x i is the ith observation value among the n observation values ​​of the kth sensor, is the average value of all observations of the kth sensor;

[0073] Step 2: Remove c observations from the calculated eccentric distances in descending order, and perform single sensor weighted fusion processing on the remaining nc observations of the kth sensor. The smaller the observation value deviates from the center, the higher the consistency with other observation values, and the larger the corresponding weighting factor. The weighting factor expression is as follows:

[0074]

[0075] where 0≤w(i)≤1, and To avoid the complexity of calculating the weighting factor when the eccentric distance is 0, the d(i)+1 method is used;

[0076] Step 3: After the kth sensor is processed by outlier removal and weighted fusion, the single sensor weighted fusion data of the remaining n–c observations can be obtained. The single sensor weighted fusion data expression is as follows:

[0077]

[0078] in, is the single sensor weighted fusion data of the kth sensor when the measurement data sample length is n;

[0079] Step 4: Execute steps 1 to 3 on the remaining sensor data to obtain k single sensor weighted fusion data

[0080] Step 5: Execute steps 1 to 4 on the operating data collected under different fault modes of the gearbox to obtain single sensor weighted fusion data of the operating data under different fault modes.

[0081] Step 3: adaptive weighted fusion processing of multi-sensor data;

[0082] Step 1: Let the observation values ​​of m sensors be X1, X2, ..., X m Independent of each other, they are all unbiased estimates of the true value. The variances corresponding to the observations of each sensor are According to the multi-sensor weighted fusion data expression, the multi-sensor weighted fusion data can be obtained. The corresponding total mean square error is σ 2 , the multi-sensor weighted fusion data expression and the total mean square error expression are as follows:

[0083]

[0084] Among them, W k is the weighting factor of the multi-sensor weighted fusion data of the kth sensor, Multi-sensor weighted fusion data for multiple sensors;

[0085]

[0086] Among them, W k is the multi-sensor fusion weight of the kth sensor, X is the monitoring value of the sensor, is the true value corresponding to the sensor monitoring value, is the variance of the kth sensor;

[0087] Step 2: Calculate the total mean square error σ by Lagrange multiplier method 2 The optimal weight factor corresponding to the kth sensor when it is minimum is The multi-sensor weighted fusion data is obtained based on the optimal weight factor. The optimal weight factor expression is as follows:

[0088]

[0089] Step 3: Execute the second step in step 3 on the operating data collected under different fault modes of the gearbox to obtain multi-sensor weighted fusion data of the operating data under different fault modes;

[0090] Step 4, construct a multi-sensor multi-level data set;

[0091] Step 1: Weighted fusion data of each single sensor under different fault modes obtained in the fourth step of step 2 and the third step of step 3 Weighted fusion of multi-sensor data Perform integration and use the monitored gearbox fault type as label y;

[0092] Step 2: Set the time window length t, divide the continuous time series data into multiple discrete samples, where each sample dimension is (t, m+1), and construct a multi-sensor multi-level dataset P for gearbox fault diagnosis.

[0093] The weighted fusion data of each single sensor in different modes and the weighted fusion data of multiple sensors in the first step of step 4 are obtained through steps S1 and S1;

[0094] Step 5: Build a multi-sensor fault diagnosis model

[0095] Step 1: Build a self-attention network with input and output data dimensions (t, m+1). The expression of the self-attention mechanism is as follows:

[0096]

[0097] Where Q represents the query vector, K represents the key vector, V represents the value vector, and dk represents the dimension of the key vector K;

[0098] Step 2: Construct 1D-CNN. The input data is the output data of the self-attention network. The output data dimension is (c, 1), where c represents the probability of c types of faults output by the current multi-sensor fault diagnosis model. The 1D-CNN network structure includes two convolution blocks. Each convolution block consists of a convolution layer, a batch normalization layer, an activation layer, and a pooling layer in sequence. Finally, it is connected to a fully connected layer and outputs the fault type through a Softmax classifier. The Softmax expression is as follows:

[0099]

[0100] Among them, m is the number of input nodes, that is, the number of categories; z j Output value for the jth node;

[0101] Step 6: Train a multi-sensor fault diagnosis model

[0102] Step 1: Divide the gearbox fault diagnosis multi-sensor multi-level data set P into a training set and a test set in proportion to obtain the training set (X train ,y train ) and the test set (X test ,y test );

[0103] Step 2: Set the batch size and learning rate for model training, use cross entropy as the loss function, and transform the training set (X train ,y train ) are input into the self-attention network and 1D-CNN in turn, and the weights of neurons in different layers of 1D-CNN are updated through error back propagation and gradient descent until the maximum number of iterations is met and the model training is completed. The cross entropy loss function expression is as follows:

[0104]

[0105] Among them, class represents the type of label, x[j] represents the input of the loss function, that is, the characteristics of the network structure output, and N is the dimension of x;

[0106] Step 7: Gearbox fault diagnosis

[0107] Step 1: Set the test set (X test ,y test ) is input into the final network model established, and the prediction results of each fault category are output;

[0108] Step 2: Select the maximum value from the prediction results, take the category corresponding to the maximum value as the category of gearbox fault diagnosis, and output the prediction label;

[0109] Step 3: Calculate the evaluation index (accuracy) based on the prediction results of the test set samples

[0110]

[0111] Among them, N is the number of samples, y i is the true label, To predict the label

[0112] The classification accuracy of the present invention is verified by comparing the method of the present invention, four other methods and the existing method:

[0113] 1. Experimental conditions:

[0114] The hardware platform of the simulation experiment of the present invention is: the central processing unit is Intel (R) Core (TM) i12-12600KF CPU, the main frequency is 3.70GHZ, and the memory is 32G. The software platform of the simulation experiment of the present invention is: WINDOWS11 operating system and Python3.9.

[0115] The hardware platform of the experiment of the present invention consists of a 2.2kW three-phase asynchronous motor, a torque sensor, a two-stage parallel gearbox, a magnetic powder brake as a torque generator, and a measurement and control system. The experimental platform is equipped with two triaxial vibration acceleration sensors (model TES001V, sensitivity 100mv / g), which measure the triaxial vibration signals of the motor and the gearbox intermediate shaft along the x, y, and z axes at a sampling frequency of 12.8kHz. The measurement positions and types are shown in Table 1.

[0116] Table 1 Sensor measurement position and type

[0117]

[0118]

[0119] The input experimental data used in the simulation experiment of the present invention is experimental data from the gearbox multi-mode fault diagnosis test bench of Tsinghua University. The data is vibration data collected under a constant working condition of a speed of 1000r / min and a load of 10Nm, including gear pitting, gear wear, gear missing teeth, gear broken teeth, gear cracks, and no defects, a total of 6 gearbox states, each state includes 250 samples, each sample is a collected vibration signal segment, consisting of 1024 sampling data points. 200 samples are randomly selected from each state as training data to maintain universality, and the remaining 50 samples are used for testing. Detailed information about the 6 fault types is listed in Table 2.

[0120] Table 2 Gearbox fault type information

[0121]

[0122] 2. Simulation content and results analysis:

[0123] By adopting the method of the present invention, the raw data of the above working conditions are subjected to multi-sensor adaptive fusion processing to obtain single sensor fusion signals and multi-sensor weighted fusion data of multiple sensors under different fault modes. Figure 2 The time domain waveform diagram before and after the weighted fusion data of multiple sensors under the gear pitting state is shown in the present invention.

[0124] The experiment of the present invention adopts the method of the present invention, four other methods and existing methods, which include four ablation variants, wherein the comparison method 1 removes the multi-sensor data-level fusion module and the attention network, and directly uses the original data as the input of the 1D-CNN model; the comparison method 2 omits the multi-sensor data-level fusion module, but retains the attention network, and inputs the original data into the attention network and 1D-CNN in sequence for fault diagnosis training; the comparison method 3 does not include the attention network, and directly uses the multi-sensor multi-level data set as the input of the 1D-CNN model; the comparison method 4 only uses the multi-sensor weighted fusion data as the model input, and the existing method is an existing multi-sensor weighted fusion data fault diagnosis method, and all comparison method settings are shown in Table 3, wherein the 1D-CNN structural parameters are shown in Table 4. Based on the above experimental conditions, the experimental data from the gearbox multi-mode fault diagnosis test bench of Tsinghua University are simulated and analyzed to obtain experimental results, and the gearbox fault types are further classified according to the experimental results, and the classification results are compared with the classification results of the prior art.

[0125] The prior art refers to the method proposed by Xie Jionghong et al. in "Gearbox fault diagnosis based on multi-sensor signal fusion and residual neural network%J Journal of Chongqing University of Technology (Natural Science) [J]. 2023, 37(07): 144-52."

[0126] Table 3 Comparison method settings

[0127]

[0128] Table 41 Structural parameters of D-CNN

[0129]

[0130]

[0131] The fault diagnosis model is set to use the Adam optimizer, the learning rate is 10-3, the training batch size of the data set is set to 64, and a total of 60 epochs are trained. The training set is input into the fault diagnosis models of different methods for training. After the model training is completed, the model is tested with the test set, and the prediction results are output. The classification accuracy Acc is used to evaluate the diagnostic accuracy of different methods. The Acc expression is as follows:

[0132]

[0133] Among them, N is the number of samples, y i is the true label, is the predicted label.

[0134] The fault diagnosis results of the method of the present invention are compared with those of different ablation variants and the prior art. The test accuracy is shown in Table 3. The accuracy curves of the various methods are shown in Figure 3 shown.

[0135] Table 3 Test accuracy of the present invention and other methods

[0136]

[0137] It can be seen from Table 3 that the classification accuracy of the gearbox fault diagnosis method based on multi-sensor adaptive fusion proposed in the present invention is 93.66%, which is significantly higher than the classification accuracy of the four ablation variants. Compared with the prior art method, it can be seen that the classification accuracy of the method of the present invention is improved by 4.66%.

[0138] At the same time, in order to further evaluate the diagnostic accuracy of the method of the present invention on each fault type, the following Figure 3 The schematic diagram of the gearbox fault diagnosis confusion matrix is ​​shown. It can be seen from the figure that the method of the present invention has a good diagnostic effect on each fault mode.

[0139] In summary, the present invention can not only effectively identify and remove signal outliers, but also significantly reduce the interference of noise on fault feature analysis, ensuring the reliability of multi-sensor weighted fusion data. In addition, by integrating multi-sensor weighted fusion data with single-sensor fusion signals after consistency preprocessing, the fault feature information is effectively supplemented and enhanced, thereby significantly improving the accuracy of the gearbox fault diagnosis model.

[0140] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A gearbox fault diagnosis method based on multi-sensor adaptive fusion, characterized in that: The following steps are involved: Acquire monitoring data of multiple sensors in the gearbox, and perform adaptive weighted fusion processing based on the multi-sensor data of the gearbox to obtain multi-sensor weighted fusion data under different fault modes; A multi-sensor multi-level dataset is constructed based on weighted fusion data to train a multi-sensor fault diagnosis model for fault diagnosis of gearbox faults.

2. The gearbox fault diagnosis method based on multi-sensor adaptive fusion according to claim 1 is characterized in that: Gearbox multi-sensor data acquisition includes the following steps: Collect multi-sensor data of gearbox; A1. Multiple sensors are pre-arranged at key positions of the gearbox to synchronously collect gearbox operation data; A2. Execute step A1 for different gearbox failure mode operation states; Weighted fusion processing of single sensor data; B1. Use the consistency test theory based on eccentric distance to pre-process the measurement data of each sensor and eliminate the data with large measurement deviation in the multi-sensor system; B2. Eliminate c observations from the calculated eccentricity distances in descending order, and perform single sensor weighted fusion on the remaining nc observations of the kth sensor. The smaller the observation value deviates from the center, the higher the consistency with other observation values, and the larger the corresponding weighting factor; B3. After the kth sensor is processed by outlier removal and weighted fusion, the single sensor weighted fusion data of the remaining n–c observations can be obtained; B4. Execute steps B1-B3 on the remaining sensor data under the fault mode to obtain multiple single sensor weighted fusion data; B5. Execute steps B1-B4 on the operating data collected under different fault modes of the gearbox to obtain single sensor weighted fusion data of the operating data under different fault modes.

3. The gearbox fault diagnosis method based on multi-sensor adaptive fusion according to claim 1 is characterized in that: After constructing the multi-sensor multi-level dataset, the following steps are also included: Construct a multi-sensor fault diagnosis model; C1. Build a self-attention network, input and output data dimensions; C2. Construct 1D-CNN. The input data is the output data of the self-attention network. The output data dimension is 1. The 1D-CNN network structure includes two convolution blocks. Each convolution block is composed of a convolution layer, a batch normalization layer, an activation layer, and a pooling layer in sequence. Finally, it is connected to a fully connected layer and the fault type is output through a Softmax classifier. Training multi-sensor fault diagnosis models; D1. Divide the gearbox fault diagnosis multi-sensor integrated data set into a training set and a test set in proportion to obtain a training set and a test set; D2. Set the batch size and learning rate for model training, use cross entropy as the loss function, input the training set into the self-attention network and 1D-CNN in turn, and update the weights of neurons in different layers of 1D-CNN through error back propagation and gradient descent until the stopping condition is met and the model completes training; Gearbox failure diagnosis; E1. Input the test set into the final network model established and output the prediction results for each fault category; E2. Select the maximum value from the prediction results, use the category corresponding to the maximum value as the category of gearbox fault diagnosis, and output the prediction label; E3. Calculate the accuracy index based on the prediction results of the test set samples.

4. The gearbox fault diagnosis method based on multi-sensor adaptive fusion according to claim 1 is characterized in that: The multi-sensor data adaptive weighted fusion processing specifically includes the following steps: F1. Calculate the variance corresponding to the observation value of each sensor, and obtain the total mean square error corresponding to the multi-sensor weighted fusion data according to the multi-sensor weighted fusion data formula; F2. Calculate the optimal weight factor corresponding to the kth sensor when the total mean square error is minimized by the Lagrange multiplier method, and obtain the multi-sensor weighted fusion data under the fault mode based on the optimal weight factor; F3. Execute step F2 on the operating data collected under different fault modes of the gearbox to obtain multi-sensor weighted fusion data of the operating data under different fault modes.

5. The gearbox fault diagnosis method based on multi-sensor adaptive fusion according to claim 1 is characterized in that: Constructing a multi-sensor multi-level dataset includes the following steps: G1. Integrate the weighted fusion data of each single sensor under different fault modes with the weighted fusion data of multiple sensors, and use the monitored gearbox fault type as a label; G2. Divide the fused data into multiple discrete samples and construct a multi-sensor and multi-level dataset for gearbox fault diagnosis.

6. The gearbox fault diagnosis method based on multi-sensor adaptive fusion according to claim 4 is characterized in that: The expression of multi-sensor weighted fusion data is as follows: Among them, W k is the weighting factor of the multi-sensor weighted fusion data of the kth sensor, is the fusion data of multiple sensors, m is the number of sensors, is the single sensor weighted fusion data of the kth sensor when the measurement data sample length is n; The mean square error expression of multi-sensor weighted fusion data is as follows: Among them, W k is the multi-sensor fusion weight of the kth sensor, X is the monitoring value of the sensor, is the true value corresponding to the sensor monitoring value, is the variance of the kth sensor; The calculation expression of the optimal weight factor is as follows:

Images