Vehicle steering system hydraulic internal leakage fault diagnosis method and system based on Bo-Transformer-LSTM

Abstract

The present invention provides a Bo-Transformer-LSTM-based vehicle steering system hydraulic internal leakage fault diagnosis method and system, relating to the field of fault detection technology. The method comprises steps S1: obtaining fault data of a multi-axle special vehicle hydraulic steering system leaking at a preset fault level; S2: normalizing the fault data and dividing the normalized fault data into a training set and a test set according to a preset ratio; and S3: segmenting the input fault data and inputting it into a Bo-Transformer-LSTM model. The hyperparameters of the Bo-Transformer-LSTM model are optimized using a Bayesian algorithm to obtain the optimal parameter combination, and the trained Bo-Transformer-LSTM model is saved. This method can solve the problem of difficulty in accurately locating faults due to the large number of components and strong nonlinearity in hydraulic systems.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a method and system for diagnosing hydraulic internal leakage faults in a vehicle steering system based on Bo-Transformer-LSTM. Background Art

[0002] Multi-axle special-purpose vehicles, characterized by heavy loads, numerous axles, and long lengths, often utilize hydraulically assisted front-wheel steering and independent rear-axle hydraulic steering systems. Due to the complex and harsh driving environments of these vehicles and the strong nonlinearity of their hydraulic systems, pinpointing the exact location of a fault can be difficult, increasing maintenance complexity and costs.

[0003] In the intelligent fault diagnosis method of vehicle steering system, H.Xiong, Z.Wang, G.Wu, Y.Pan, Z.Yang, and Z.Long, "Steering Actuator Fault Diagnosis for Autonomous Vehicle With anAdaptive Denoising Residual Network," IEEE Trans.Instrum.Meas., 71, pp.1-13, (2022) proposed an adaptive denoising residual network for steering actuator fault diagnosis, reducing noise interference. S.He, X.Xu, J.Xie, F.Wang, Z.Liu, and F.Zhao, "Fault detection and fault-tolerant control of autonomous steering system for intelligent vehicles combining Bi-LSTM and SPRT," Measurement, 212, p.112708, (2023) designed a fault detection and fault-tolerant control strategy based on a bidirectional long short-term memory frame displacement estimator combined with a sequential probability ratio test to handle displacement sensor faults. Although deep learning can solve the fault diagnosis problem of nonlinear systems, a large amount of data is required to ensure the diagnostic accuracy of the neural network. At the same time, actual vehicle experiments under hydraulic steering system failure conditions cannot guarantee vehicle safety, making it difficult to collect sufficient failure data and unable to effectively solve the problem of difficulty in accurately locating hydraulic system components when they fail. Summary of the Invention

[0004] The problem solved by the present invention is how to solve the problem that the hydraulic system has many components and strong nonlinearity, making it difficult to accurately locate the fault when it occurs.

[0005] To solve the above problems, the present invention provides a vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM, comprising the steps of:

[0006] S1: Obtaining leakage fault data of a hydraulic steering system of a multi-axle special vehicle under a preset fault level;

[0007] S2: Normalize the fault data and divide the normalized fault data into a training set and a test set according to a preset ratio;

[0008] S3: Segment the input fault data and input it into the Bo-Transformer-LSTM model. Use the Bayesian algorithm to optimize the hyperparameters of the Bo-Transformer-LSTM model to obtain the optimal parameter combination, and save the trained Bo-Transformer-LSTM model.

[0009] S4: The test set data is split and input into the trained Bo-Transformer-LSTM model for prediction. The output of the prediction result is denormalized to obtain the actual prediction result. The actual prediction result is compared with the data in the test set to evaluate the prediction accuracy of the Bo-Transformer-LSTM model.

[0010] Furthermore, the acquisition of the fault data includes:

[0011] S11: Use a steering angle tester to measure the steering wheel input angle and the steering torque fed back to the steering wheel by the steering system. Simultaneously, the vehicle's steering wheel angle, steering wheel torque, and travel speed are collected.

[0012] S12: Testing the vehicle at different speeds and recording the steering angle and steering torque input by the steering wheel during the test according to step S11. When the vehicle returns to its original position after one circle, the test is considered completed.

[0013] S13: The steering angle, steering torque, and steering cylinder inlet and outlet pressures are collected respectively, and combined with the location information of the fault that occurred during the test to generate a fault data set.

[0014] Furthermore, the step S2 includes:

[0015] S21: Set the ratio of the training set to the test set, as well as the optimization range of the number of LSTM units, learning rate, and L2 regularization parameter, and set the maximum number of iterations and maximum number of training times of the Bayesian network.

[0016] Furthermore, the composition of the Bo-Transformer-LSTM model in step S3 includes:

[0017] Transformer encoding layer: Provides attentive data to the Transformer-LSTM network. Data processed by the Transformer encoding layer provides data features for fault diagnosis. The Transformer encoding layer's mechanisms include self-attention, multi-head attention, position encoding, residual connections, and layer normalization.

[0018] LSTM decoding layer: It consists of an input gate, a forget gate, an output gate, and a cell state. LSTM takes the Transformer encoding layer data, the previous hidden layer data, and the previous cell state output as common inputs, and processes the data serially through a gating mechanism.

[0019] Bayesian optimization is used to find the optimal hyperparameter configuration of the model, and the algorithm is optimized by optimizing the parameters of the learning rate, L2 regularization, and the number of LSTM units.

[0020] Furthermore, the step S4 includes:

[0021] S41: The prediction accuracy of the model was evaluated based on the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM as algorithm evaluation indicators;

[0022] S42: Based on the evaluation indicators, the collected fault data set is diagnosed using the Bo-Transformer-LSTM algorithm to obtain a prediction result.

[0023] Furthermore, the calculation formulas for the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM in step S41 are as follows:

[0024]

[0025] Among them, TP represents the number of samples correctly identified as faulty; TN represents the number of samples correctly identified as non-faulty; FN represents the number of samples incorrectly identified as faulty; FP represents the number of samples incorrectly identified as non-faulty; f(x) represents the receiver operating characteristic curve.

[0026] A vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM, comprising:

[0027] Acquisition module: used to obtain leakage fault data of the hydraulic steering system of a multi-axle special vehicle under a preset fault degree;

[0028] Processing module: used to normalize the fault data and divide the normalized fault data into training set and test set according to the preset ratio;

[0029] Segmentation module: It is used to segment the input fault data and input it into the Bo-Transformer-LSTM model. It uses the Bayesian algorithm to optimize the hyperparameters of the Bo-Transformer-LSTM model to obtain the optimal parameter combination and save the trained Bo-Transformer-LSTM model.

[0030] Fault prediction module: This module is used to segment the test set data and input it into the trained Bo-Transformer-LSTM model for prediction. The output of the prediction result is denormalized to obtain the actual prediction result. The actual prediction result is compared with the data in the test set to evaluate the prediction accuracy of the Bo-Transformer-LSTM model.

[0031] Furthermore, the acquisition of the fault data includes:

[0032] Acquisition unit: used to measure the steering wheel input angle and the steering torque fed back to the steering wheel by the steering system through a steering force angle tester, and simultaneously collect the vehicle's steering wheel angle, steering wheel steering torque and travel speed;

[0033] Test unit: used to test the vehicle at different speeds and record the steering angle and steering torque input by the steering wheel during the test through the acquisition unit. When the vehicle returns to its original position after driving a circle, the experiment is considered completed.

[0034] Generation unit: collects steering angle, steering torque, and steering cylinder inlet and outlet pressures respectively, and generates a fault data set based on the location information of the fault that occurred during the test.

[0035] Furthermore, the processing module includes:

[0036] Setting unit: used to set the ratio of training set to test set, as well as the optimization range of LSTM unit number, learning rate, L2 regularization parameter, and the maximum number of iterations and training times of the Bayesian network.

[0037] Furthermore, the structure of the Bo-Transformer-LSTM model in the segmentation module includes:

[0038] Transformer encoding layer: Provides attentive data to the Transformer-LSTM network. Data processed by the Transformer encoding layer provides data features for fault diagnosis. The Transformer encoding layer's mechanisms include self-attention, multi-head attention, position encoding, residual connections, and layer normalization.

[0039] LSTM decoding layer: It consists of an input gate, a forget gate, an output gate, and a cell state. LSTM takes the Transformer encoding layer data, the previous hidden layer data, and the previous cell state output as common inputs, and processes the data serially through a gating mechanism.

[0040] Bayesian optimization is used to find the optimal hyperparameter configuration of the model, and the algorithm is optimized by optimizing the parameters of the learning rate, L2 regularization, and the number of LSTM units.

[0041] Furthermore, the fault prediction module includes:

[0042] Evaluation unit: Based on the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM as algorithm evaluation indicators, the prediction accuracy of the model is evaluated;

[0043] Fault diagnosis unit: Based on the evaluation indicators, the collected fault data set is diagnosed through the Bo-Transformer-LSTM algorithm to obtain the prediction results.

[0044] Furthermore, the calculation formulas for the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM in the evaluation unit are as follows:

[0045]

[0046] Among them, TP represents the number of samples correctly identified as faulty; TN represents the number of samples correctly identified as non-faulty; FN represents the number of samples incorrectly identified as faulty; FP represents the number of samples incorrectly identified as non-faulty; f(x) represents the receiver operating characteristic curve.

[0047] The present invention adopts the above technical solution to achieve the following beneficial effects:

[0048] This invention utilizes a Bo-Transformer-LSTM neural network. The Transformer decoding layer is reconstructed and integrated with a long-short-term memory (LSTM) neural network structure. Bayesian optimization is used to optimize the network parameters to identify and process different types of faults. The Bo-Transformer-LSTM neural network accurately identifies different types of faults, effectively reducing the maintenance difficulty of the hydraulic steering system of multi-axle special vehicles. Actual vehicle test data under normal driving conditions is collected and compared with an AMESim simulation model of the hydraulic steering system to verify model accuracy. The simulation model is then used to generate a fault dataset for the hydraulic steering system. A Bo-Transformer-LSTM algorithm is also proposed for fault diagnosis of the hydraulic steering system of multi-axle special vehicles. This network extracts data features by combining serial and parallel data. In the Bo-Transformer-LSTM network, the Transformer handles parallel data processing, focusing on specific data components. This ensures that features are not forgotten when analyzing specific faults, improving fault diagnosis accuracy. The LSTM handles serial data processing, focusing on overall data trends. This ensures sufficient data features when analyzing multiple types of faults, improving fault classification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The process of the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in the first embodiment of the present invention Figure 1 ;

[0050] Figure 2 The process of the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in the first embodiment of the present invention Figure 2 ;

[0051] Figure 3 A schematic diagram of the LSTM algorithm structure in the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in Example 1 of the present invention;

[0052] Figure 4 Flowchart of a vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in Example 1 of the present invention;

[0053] Figure 5 A line graph of steering wheel angle data in the Bo-Transformer-LSTM-based vehicle steering system hydraulic internal leakage fault diagnosis method provided in Example 1 of the present invention;

[0054] Figure 6A line graph of steering wheel torque data in the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in Example 1 of the present invention;

[0055] Figure 7 A schematic diagram of a simulation model of fault categories in the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in the first embodiment of the present invention;

[0056] Figure 8a , Figure 8b , Figure 8c and Figure 8d A comparison chart of prediction results of four different algorithm test sets in the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in Example 1 of the present invention;

[0057] Figure 9a , Figure 9b , Figure 9c and Figure 9d A comparison chart of confusion matrices of four different algorithms in the Bo-Transformer-LSTM-based vehicle steering system hydraulic internal leakage fault diagnosis method provided in Example 1 of the present invention;

[0058] Figure 10a , Figure 10b , Figure 10c and Figure 10d A comparison chart of the areas under the curves of four different algorithms in the Bo-Transformer-LSTM-based vehicle steering system hydraulic internal leakage fault diagnosis method provided in Example 1 of the present invention;

[0059] Figure 11 A comparison chart of evaluation indicators of four different algorithms in the vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM provided in Example 1 of the present invention;

[0060] Figure 12a and Figure 12b A schematic diagram of the network structure within the Transformer-LSTM unit in the Bo-Transformer-LSTM-based vehicle steering system hydraulic internal leakage fault diagnosis method provided in Example 1 of the present invention;

[0061] Figure 13 The structure of the vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM provided in the second embodiment of the present invention Figure 1 ;

[0062] Figure 14The structure of the vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM provided in the second embodiment of the present invention Figure 2 . DETAILED DESCRIPTION

[0063] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0064] The following are specific embodiments of the present invention and the accompanying drawings to further describe the technical solutions of the present invention, but the present invention is not limited to these embodiments.

[0065] Example 1: This embodiment provides a method for diagnosing hydraulic internal leakage faults in a vehicle steering system based on Bo-Transformer-LSTM. Figure 1 and Figure 2 As shown, the method includes the steps of:

[0066] S1: Obtaining leakage fault data of a hydraulic steering system of a multi-axle special vehicle under a preset fault level;

[0067] S2: Normalize the fault data and divide the normalized fault data into a training set and a test set according to a preset ratio;

[0068] S3: Segment the input fault data and input it into the Bo-Transformer-LSTM model. Use the Bayesian algorithm to optimize the hyperparameters of the Bo-Transformer-LSTM model to obtain the optimal parameter combination, and save the trained Bo-Transformer-LSTM model.

[0069] S4: The test set data is split and input into the trained Bo-Transformer-LSTM model for prediction. The output of the prediction result is denormalized to obtain the actual prediction result. The actual prediction result is compared with the data in the test set to evaluate the prediction accuracy of the Bo-Transformer-LSTM model.

[0070] Specifically, real-world test data from a multi-axle special vehicle under normal driving conditions was collected and compared with an AMESim simulation model of the hydraulic steering system to verify model accuracy. The simulation model was then used to generate a fault dataset for the hydraulic steering system. A Bo-Transformer-LSTM algorithm was proposed for fault diagnosis of the hydraulic steering system of the multi-axle special vehicle. This network extracts data features by combining serial and parallel processing. In the Bo-Transformer-LSTM network, the Transformer handles parallel data processing, focusing on specific data components. This ensures that features are not forgotten when analyzing specific faults, improving fault diagnosis accuracy. The LSTM handles serial data processing, focusing on overall data trends. This ensures sufficient data features when analyzing multiple fault types, improving fault classification accuracy. Bayesian optimization optimizes network hyperparameters, dynamically adjusting input data processing during network training to optimize model performance.

[0071] The acquisition of fault data includes:

[0072] S11: Use a steering angle tester to measure the steering wheel input angle and the steering torque fed back to the steering wheel by the steering system. Simultaneously, the vehicle's steering wheel angle, steering wheel torque, and travel speed are collected.

[0073] S12: Testing the vehicle at different speeds and recording the steering angle and steering torque input by the steering wheel during the test according to step S11. When the vehicle returns to its original position after one circle, the test is considered completed.

[0074] S13: The steering angle, steering torque, and steering cylinder inlet and outlet pressures are collected respectively, and combined with the location information of the fault that occurred during the test to generate a fault data set.

[0075] Wherein, step S2 includes:

[0076] S21: Set the ratio of the training set to the test set, as well as the optimization range of the number of LSTM units, learning rate, and L2 regularization parameter, and set the maximum number of iterations and maximum number of training times of the Bayesian network.

[0077] The structure of the Bo-Transformer-LSTM model in step S3 includes:

[0078] Transformer encoding layer: Provides attentive data to the Transformer-LSTM network. Data processed by the Transformer encoding layer provides data features for fault diagnosis. The Transformer encoding layer's mechanisms include self-attention, multi-head attention, position encoding, residual connections, and layer normalization.

[0079] LSTM decoding layer: It consists of an input gate, a forget gate, an output gate, and a cell state. LSTM takes the Transformer encoding layer data, the previous hidden layer data, and the previous cell state output as common inputs, and processes the data serially through a gating mechanism.

[0080] Bayesian optimization is used to find the optimal hyperparameter configuration of the model, and the algorithm is optimized by optimizing the parameters of the learning rate, L2 regularization, and the number of LSTM units.

[0081] See Figure 12a and Figure 12b , modify the Decoder layer in the Transformer model to a fully connected layer, and replace the original attention layer with an LSTM layer.

[0082] Specifically, the self-attention mechanism is a key feature of the Transformer model, which enables the model to perform weighted aggregation on parts of the input sequence. By calculating the attention score between the query Q, key K and value V, d K Represents the number of dimensions of K. The model can determine the importance of each position to other positions. After the attention score is scaled, it is fused into the value matrix by weighted summation to produce the final output. The calculation formula is shown in formula (1):

[0083]

[0084] Specifically, the multi-head attention mechanism uses multiple parallel self-attention mechanisms to capture different types of relationships. Each head learns different weights, so that it can capture information in different representation subspaces. Using H parallel attention calculations, In this case, W A Represents the learning parameters, and the calculation formula of the multi-head attention mechanism is shown in formula (2):

[0085]

[0086] Specifically, position encoding is used to add some information to each position in the input sequence because the Transformer has no built-in sequence order information so that the model can distinguish between tokens at different positions. The calculation formulas are shown in formulas (3) and (4):

[0087]

[0088] Among them, pos represents the sequence position, i represents the dimension, and d represents the size of the embedding space dimension.

[0089] Specifically, the LSTM decoding layer consists of an input gate, a forget gate, an output gate, and a cell state. LSTM needs to transform the Transformer encoding layer data and the previous hidden layer data h t and the cell state output C at the previous moment t As a common input, data is processed serially through the gating mechanism to achieve data processing. The LSTM network structure is as follows Figure 3 As shown, σ represents the sigmoid activation function, h represents the hidden state, x represents the input, and C represents the cell state.

[0090] Specifically, in the application of the Transformer-LSTM model, Bayesian optimization can be used to automatically find the optimal hyperparameter configuration of the model and optimize the algorithm by optimizing the following parameters:

[0091] Learning rate: controls the step size of the model's weight update in each iteration, which has a significant impact on the model's convergence speed and final performance;

[0092] L2 regularization: A regularization technique used to prevent overfitting by randomly dropping some neurons during training to improve the generalization ability of the model;

[0093] Number of LSTM units: The number of neurons in the LSTM layer determines the capacity and complexity of the model.

[0094] Through Bayesian optimization, the way input data is processed can be dynamically adjusted during the training process to optimize the performance of the model.

[0095] See Figure 4 The fault diagnosis process of hydraulic internal leakage of multi-axis special vehicle steering system based on Bo-Transformer-LSTM network is as follows: obtain fault data of different degrees of internal leakage of multi-axis special vehicle hydraulic steering system; normalize the processed data and divide it into training set and test set according to the proportion; segment the input data, input the segmented data into the model, and optimize the hyperparameters through the Bayesian algorithm to obtain the optimal parameter combination and save the trained model; segment the test set data and input it into the model for prediction, denormalize the output to obtain the prediction result, compare it with the prediction set data, and evaluate the model accuracy.

[0096] Wherein, step S4 includes:

[0097] S41: The prediction accuracy of the model was evaluated based on the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM as algorithm evaluation indicators;

[0098] S42: Based on the evaluation indicators, the collected fault data set is diagnosed using the Bo-Transformer-LSTM algorithm to obtain a prediction result.

[0099] Specifically, the fault dataset is constructed:

[0100] The AM-2012 steering force angle tester is used to measure the steering wheel input angle θ and the steering torque T fed back to the steering wheel by the steering system. The vehicle's steering wheel angle θ, steering wheel steering torque T, and travel speed are also collected.

[0101] Before the experiment began, the driver drove the vehicle several times along a pre-set experimental path similar to the shape of the letter "∞" to prepare for the experiment. At the start of the experiment, the driver first returned the steering wheel to the center of the experimental vehicle and began driving from the symmetrical center of the "∞" experimental path. After completing one lap along the pre-set experimental path, the driver stopped the vehicle and returned it to its original position. After the vehicle stopped, the steering wheel was returned to its original position.

[0102] During the test, the vehicle drove along a pre-set "∞" path at speeds of 10 km / h, 20 km / h, and 30 km / h. When the vehicle's speed stabilized, the AM-2012 tester began recording the steering wheel angle θ and steering torque T in real time. A full test was completed when the vehicle returned to its starting position after a complete circle from the center of the test path. The vehicle verification test was completed after three full tests at each speed.

[0103] The experiment obtained 9 data sets at three different vehicle speeds, using the representative experimental data at 20km / h, totaling 1800 sets. The experimental filtered data was plotted as a line graph. The results are as follows Figure 5 and Figure 6 As shown in the figure. By comparing the simulation results of the all-wheel steering system model with the actual experimental data, the correctness of the model can be verified. Figure 5 and Figure 6 shown.

[0104] Figure 5 By comparing the model input angle signal with the experimental angle data, it is found that the change of the model input angle signal can roughly fit the change trend of the input angle data image during the experiment; Figure 6 Comparing the model's spring steering torque with experimental steering torque data also reveals a close fit between the image trends. This comparison demonstrates that, given a close fit between the input angles, the steering torque parameter curves also roughly match the experimental results. The simulation and experiment are largely consistent, essentially achieving the experimental objective and validating the overall model.

[0105] Specifically, hydraulic steering system failure classification:

[0106] For special vehicles, internal leakage in the steering hydraulic system typically occurs in the rotary valve, front axle hydraulic steering, 4-axis hydraulic steering, and 5-axis hydraulic steering, a total of four fault locations. These faults are categorized into 15 categories based on their location and number, as shown in Table 1, where 1 represents the fault location.

[0107] Table 1

[0108]

[0109] According to the degree of internal leakage, the preset fault level is divided into three levels of fault:

[0110] Slight internal leakage, specifically manifested as the vehicle being able to complete the expected steering performance, but with a certain steering delay;

[0111] Moderate internal leakage: The vehicle still has the steering function, but cannot reach the expected steering angle;

[0112] Severe internal leakage, manifested as a complete loss of the vehicle's steering function.

[0113] Therefore, according to the location of the fault and the severity of the internal leakage, combined with the loss of vehicle steering function in engineering practice when severe internal leakage occurs, it is impossible to accurately determine the location of the severe internal leakage, and a total of 30 types are divided.

[0114] Specifically, according to the classification of leakage faults in the hydraulic steering system in Table 1, the fault data set is collected through the following simulation model. Taking the category 15 fault model as an example, the remaining 14 categories are permutations and combinations of the four fault locations. Category 15 is as follows: Figure 7 shown.

[0115] Figure 5 The data shown here is generated by the fault model under steering conditions. The fault dataset was generated based on the 15 fault categories listed above. The orifice diameter was set in 0.1mm increments from 0 to the corresponding pipe diameter. Seven types of data were collected, including steering angle, steering torque, front axle steering cylinder inlet and outlet pressures, and 4-axis and 5-axis steering cylinder inlet and outlet pressures, to generate the fault dataset.

[0116] Specifically, the data set is set up to collect 30 types of data for slight internal leakage and moderate internal leakage corresponding to 15 types of faults, and 500 groups of data are selected from each type of data to generate the fault data set.

[0117] Model parameter settings: In terms of model, the ratio of training set to test set is 7:3, the optimization range of LSTM unit number parameter is 10-50; the optimization range of learning rate parameter is 10 -3-1; L2 regularization parameter optimization interval is 10 -10 -10 -2 ; The maximum number of iterations of the Bayesian network is 30, and the maximum number of training times is 200.

[0118] Evaluation metrics use a method called polygon area metric to assess classifier performance. PAM uses classification accuracy (CA), sensitivity (SE), specificity (SP), area under the receiver operating characteristic (ROC) curve (AUC), Jaccard index (JI), and F-value (FM) as algorithm evaluation metrics. Their mathematical definitions are shown in Formulas (5) to (10).

[0119]

[0120] Where TP is the number of true positives, or samples correctly identified as faulty. TN is the number of true negatives, or samples correctly identified as non-faulty. FN is the number of false negatives, or samples incorrectly identified as faulty. FP is the number of false positives, or samples incorrectly identified as faulty. f(x) is the receiver operating characteristic curve, where the true positive rate (SE) is a function of the false positive rate (1-SP) at different cutoff points.

[0121] PAM is calculated using the area of ​​the polygon created by the points CA, SE, SP, AUC, JI, and FM in a regular hexagon. The regular hexagon is composed of six equilateral triangles with each side length equal to 1. where represent the values ​​of CA, SE, SP, AUC, JI, and FM, respectively. The calculation formula for PAM is as shown in Equation (11):

[0122]

[0123] Where PA is the area of ​​the polygon. To normalize PAM to the interval [0,1], the PA value is divided by 2.59807.

[0124] This experiment verifies the performance of the network through multiple sets of data. Four algorithms, LSTM, Transformer, Transformer-LSTM, and Bo-Transformer-LSTM, are used to diagnose the collected fault data sets. The results are shown in Figures 8 to 9. Figure 11 results.

[0125] Figure 8a , Figure 8b , Figure 8c and Figure 8d Represents the prediction results of the test set of 4 different algorithms, Figure 9a , Figure 9b , Figure 9c and Figure 9d Represents the confusion matrix of 4 different algorithms; Figure 10a , Figure 10b , Figure 10c and Figure 10d Represents the area under the curve of 4 different algorithms; Figure 11 Represents 4 different algorithm evaluation indicators.

[0126] Table 2

[0127]

[0128] Specifically, the comparison results of the four algorithms are shown in Figures 8 to Figure 11 As shown in Table 2, the Bo-Transformer-LSTM algorithm demonstrates significant advantages over the other three algorithms. Its classification accuracy reaches 98.0556%, demonstrating its effective fault classification. A sensitivity of 1 indicates strong fault detection capabilities, while a specificity of 1 indicates that the algorithm can reduce unnecessary repairs and inspections. The area under the curve (AUC) reaches 0.9997, demonstrating good overall performance across all possible classification thresholds. Its value ranges from 0 to 1, with 1 representing a perfect classifier, 0.5 an invalid classifier equivalent to random guessing, and values ​​below 0.5 worse than random guessing, indicating negative classifier performance. A Jaccard index of 1 indicates high algorithm consistency, and an F-value of 1 indicates a good balance between precision and recall, with a low percentage of false positives and negatives. The PAM value, derived from comprehensive analysis of all data, is 0.99, the best among the four fault diagnosis algorithms.

[0129] Specifically, a simulation model of a hydraulic steering system was established and compared with real-vehicle experiments to verify its correctness. A fault dataset was generated using the simulation model of a hydraulic steering system internal leakage fault. This dataset was then used to perform fault diagnosis and analysis on the hydraulic steering system of a special vehicle. A Bo-Transformer-LSTM-based multi-axis special vehicle hydraulic steering system fault diagnosis algorithm was employed, and its effectiveness was verified using classification accuracy (CA), sensitivity (SE), specificity (SP), area under the receiver operating characteristic (ROC) curve (AUC), Jaccard index (TI), and F-value (FM) as evaluation metrics.

[0130] This method utilizes a Bo-Transformer-LSTM neural network. The Transformer decoding layer is reconstructed and integrated with a long-short-term memory (LSTM) neural network structure. Bayesian optimization is used to optimize the network parameters to identify and process different types of faults. The Bo-Transformer-LSTM neural network accurately identifies different types of faults, effectively reducing the maintenance difficulty of the hydraulic steering system of multi-axle special vehicles. Real vehicle test data under normal driving conditions is collected and compared with an AMESim simulation model of the hydraulic steering system to verify the model's accuracy. The simulation model is then used to generate a fault dataset for the hydraulic steering system. A Bo-Transformer-LSTM algorithm is proposed for fault diagnosis of the hydraulic steering system of multi-axle special vehicles. This network extracts data features by combining serial and parallel data processing. In the Bo-Transformer-LSTM network, the Transformer handles parallel data processing, focusing on specific data components. This ensures that features are not forgotten when analyzing specific faults, improving fault diagnosis accuracy. The LSTM handles serial data processing, focusing on overall data trends. This ensures sufficient data features when analyzing multiple types of faults, improving fault classification accuracy.

[0131] Example 2: This embodiment provides a vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM, such as Figure 13 and Figure 14 As shown, this system includes:

[0132] Acquisition module: used to obtain leakage fault data of the hydraulic steering system of a multi-axle special vehicle under a preset fault degree;

[0133] Processing module: used to normalize the fault data and divide the normalized fault data into training set and test set according to the preset ratio;

[0134] Segmentation module: It is used to segment the input fault data and input it into the Bo-Transformer-LSTM model. It uses the Bayesian algorithm to optimize the hyperparameters of the Bo-Transformer-LSTM model to obtain the optimal parameter combination and save the trained Bo-Transformer-LSTM model.

[0135] Fault prediction module: This module is used to segment the test set data and input it into the trained Bo-Transformer-LSTM model for prediction. The output of the prediction result is denormalized to obtain the actual prediction result. The actual prediction result is compared with the data in the test set to evaluate the prediction accuracy of the Bo-Transformer-LSTM model.

[0136] See Figure 14 , where the acquisition of fault data includes:

[0137] Acquisition unit: used to measure the steering wheel input angle and the steering torque fed back to the steering wheel by the steering system through a steering force angle tester, and simultaneously collect the vehicle's steering wheel angle, steering wheel steering torque and travel speed;

[0138] Test unit: used to test the vehicle at different speeds and record the steering angle and steering torque input by the steering wheel during the test through the acquisition unit. When the vehicle returns to its original position after driving a circle, the experiment is considered completed.

[0139] Generation unit: collects steering angle, steering torque, and steering cylinder inlet and outlet pressures respectively, and generates a fault data set based on the location information of the fault that occurred during the test.

[0140] The processing module includes:

[0141] Setting unit: used to set the ratio of training set to test set, as well as the optimization range of LSTM unit number, learning rate, L2 regularization parameter, and the maximum number of iterations and training times of the Bayesian network.

[0142] The Bo-Transformer-LSTM model in the segmentation module consists of:

[0143] Transformer encoding layer: Provides attentive data to the Transformer-LSTM network. Data processed by the Transformer encoding layer provides data features for fault diagnosis. The Transformer encoding layer's mechanisms include self-attention, multi-head attention, position encoding, residual connections, and layer normalization.

[0144] LSTM decoding layer: It consists of an input gate, a forget gate, an output gate, and a cell state. LSTM takes the Transformer encoding layer data, the previous hidden layer data, and the previous cell state output as common inputs, and processes the data serially through a gating mechanism.

[0145] Bayesian optimization is used to find the optimal hyperparameter configuration of the model, and the algorithm is optimized by optimizing the parameters of the learning rate, L2 regularization, and the number of LSTM units.

[0146] Among them, the fault prediction module includes:

[0147] Evaluation unit: Based on the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM as algorithm evaluation indicators, the prediction accuracy of the model is evaluated;

[0148] Fault diagnosis unit: Based on the evaluation indicators, the collected fault data set is diagnosed through the Bo-Transformer-LSTM algorithm to obtain the prediction results.

[0149] Among them, the calculation formulas for classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM in the evaluation unit are:

[0150]

[0151] Among them, TP represents the number of samples correctly identified as faulty; TN represents the number of samples correctly identified as non-faulty; FN represents the number of samples incorrectly identified as faulty; FP represents the number of samples incorrectly identified as non-faulty; f(x) represents the receiver operating characteristic curve.

[0152] This system utilizes a Bo-Transformer-LSTM neural network. The Transformer decoding layer is restructured and integrated with a long-short-term memory (LSTM) neural network. Bayesian optimization is used to optimize the network parameters to identify and process different types of faults. The Bo-Transformer-LSTM neural network accurately identifies different types of faults, effectively reducing the maintenance difficulty of the hydraulic steering system of multi-axle special vehicles. Real vehicle test data from a multi-axle special vehicle under normal driving conditions is collected and compared with an AMESim simulation model of the hydraulic steering system to verify the model's accuracy. The simulation model is then used to generate a fault dataset for the hydraulic steering system. A Bo-Transformer-LSTM algorithm is proposed for fault diagnosis of the hydraulic steering system of multi-axle special vehicles. This network extracts data features by combining serial and parallel data processing. In the Bo-Transformer-LSTM network, the Transformer handles parallel data processing, focusing on specific data components. This ensures that features are not forgotten when analyzing specific faults, improving fault diagnosis accuracy. The LSTM handles serial data processing, focusing on overall data trends. This ensures sufficient data features when analyzing multiple types of faults, improving fault classification accuracy.

[0153] Although the present disclosure is disclosed as above, the protection scope of the present disclosure is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present disclosure, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM, characterized in that: Including steps: S1: Obtaining leakage fault data of a hydraulic steering system of a multi-axle special vehicle under a preset fault level; S2: Normalize the fault data and divide the normalized fault data into a training set and a test set according to a preset ratio; The step S2 comprises: S21: Set the ratio of training set to test set, as well as the optimization range of LSTM unit number, learning rate, L2 regularization parameter, and set the maximum number of iterations and maximum number of training times of the Bayesian network; S3: Segment the input fault data and input it into the Bo-Transformer-LSTM model. Use the Bayesian algorithm to optimize the hyperparameters of the Bo-Transformer-LSTM model to obtain the optimal parameter combination, and save the trained Bo-Transformer-LSTM model. The structure of the Bo-Transformer-LSTM model in step S3 includes: Transformer encoding layer: Provides attentive data to the Transformer-LSTM network. Data processed by the Transformer encoding layer provides data features for fault diagnosis. The Transformer encoding layer's mechanisms include self-attention, multi-head attention, position encoding, residual connections, and layer normalization. LSTM decoding layer: It consists of an input gate, a forget gate, an output gate, and a cell state. LSTM takes the Transformer encoding layer data, the previous hidden layer data, and the previous cell state output as common inputs, and processes the data serially through a gating mechanism. Bayesian optimization is used to find the optimal hyperparameter configuration of the model. The algorithm is optimized by optimizing the parameters of learning rate, L2 regularization, and number of LSTM units. S4: The test set data is split and input into the trained Bo-Transformer-LSTM model for prediction. The output of the prediction result is denormalized to obtain the actual prediction result. The actual prediction result is compared with the data in the test set to evaluate the prediction accuracy of the Bo-Transformer-LSTM model.

2. The vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM according to claim 1 is characterized in that: The acquisition of the fault data includes: S11: Use a steering angle tester to measure the steering wheel input angle and the steering torque fed back to the steering wheel by the steering system. Simultaneously, the vehicle's steering wheel angle, steering wheel torque, and travel speed are collected. S12: Testing the vehicle at different speeds and recording the steering angle and steering torque input by the steering wheel during the test according to step S11. When the vehicle returns to its original position after one circle, the test is considered completed. S13: The steering angle, steering torque, and steering cylinder inlet and outlet pressures are collected respectively, and combined with the location information of the fault that occurred during the test to generate a fault data set.

3. The vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM according to claim 1 is characterized in that: The step S4 comprises: S41: The prediction accuracy of the model was evaluated based on the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM as algorithm evaluation indicators; S42: Based on the evaluation indicators, the collected fault data set is diagnosed using the Bo-Transformer-LSTM algorithm to obtain a prediction result.

4. The vehicle steering system hydraulic internal leakage fault diagnosis method based on Bo-Transformer-LSTM according to claim 3 is characterized in that: The calculation formulas for the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM in step S41 are: Among them, TP represents the number of samples correctly identified as faulty; TN represents the number of samples correctly identified as non-faulty; FN represents the number of samples incorrectly identified as faulty; FP represents the number of samples incorrectly identified as non-faulty; f(x) represents the receiver operating characteristic curve.

5. A vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM, characterized in that: include: Acquisition module: used to obtain leakage fault data of the hydraulic steering system of a multi-axle special vehicle under a preset fault degree; Processing module: used to normalize the fault data and divide the normalized fault data into training set and test set according to the preset ratio; The processing module includes: Setting unit: used to set the ratio of training set to test set, as well as the optimization range of LSTM unit number, learning rate, L2 regularization parameter, and set the maximum number of iterations and maximum number of training times of Bayesian network; Segmentation module: It is used to segment the input fault data and input it into the Bo-Transformer-LSTM model. It uses the Bayesian algorithm to optimize the hyperparameters of the Bo-Transformer-LSTM model to obtain the optimal parameter combination and save the trained Bo-Transformer-LSTM model. The structure of the Bo-Transformer-LSTM model in the segmentation module includes: Transformer encoding layer: Provides attentive data to the Transformer-LSTM network. Data processed by the Transformer encoding layer provides data features for fault diagnosis. The Transformer encoding layer's mechanisms include self-attention, multi-head attention, position encoding, residual connections, and layer normalization. LSTM decoding layer: It consists of an input gate, a forget gate, an output gate, and a cell state. LSTM takes the Transformer encoding layer data, the previous hidden layer data, and the previous cell state output as common inputs, and processes the data serially through a gating mechanism. Bayesian optimization is used to find the optimal hyperparameter configuration of the model. The algorithm is optimized by optimizing the parameters of learning rate, L2 regularization, and number of LSTM units. Fault prediction module: This module is used to segment the test set data and input it into the trained Bo-Transformer-LSTM model for prediction. The output of the prediction result is denormalized to obtain the actual prediction result. The actual prediction result is compared with the data in the test set to evaluate the prediction accuracy of the Bo-Transformer-LSTM model.

6. The vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM according to claim 5 is characterized in that: The acquisition of the fault data includes: Acquisition unit: used to measure the steering wheel input angle and the steering torque fed back to the steering wheel by the steering system through a steering force angle tester, and simultaneously collect the vehicle's steering wheel angle, steering wheel steering torque and travel speed; Test unit: used to test the vehicle at different speeds and record the steering angle and steering torque input by the steering wheel during the test through the acquisition unit. When the vehicle returns to its original position after driving a circle, the experiment is considered completed. Generation unit: collects steering angle, steering torque, and steering cylinder inlet and outlet pressures respectively, and generates a fault data set based on the location information of the fault that occurred during the test.

7. The vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM according to claim 5 is characterized in that: The fault prediction module includes: Evaluation unit: Based on the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM as algorithm evaluation indicators, the prediction accuracy of the model is evaluated; Fault diagnosis unit: Based on the evaluation indicators, the collected fault data set is diagnosed through the Bo-Transformer-LSTM algorithm to obtain the prediction results.

8. The vehicle steering system hydraulic internal leakage fault diagnosis system based on Bo-Transformer-LSTM according to claim 7 is characterized in that: The calculation formulas for the classification accuracy CA, sensitivity SE, specificity SP, area under the ROC curve AUC, Jaccard index TI and F value FM in the evaluation unit are as follows: Among them, TP represents the number of samples correctly identified as faulty; TN represents the number of samples correctly identified as non-faulty; FN represents the number of samples incorrectly identified as faulty; FP represents the number of samples incorrectly identified as non-faulty. f(x) represents the receiver operating characteristic curve.

Images