Vehicle brake pad wear degree prediction method and device based on working condition recognition

Abstract

The invention discloses a vehicle brake pad wear degree prediction method and device based on working condition recognition, belongs to the field of brake pad wear degree prediction, and aims at solving the problems that in an existing vehicle brake system, brake pad friction prediction accuracy and robustness are poor; and wear prediction and brake system performance optimization cannot be efficiently and economically completed. The method comprises the following steps: carrying out clustering selection under difference maximization on a vehicle speed sequence of a vehicle to obtain a typical working condition; performing wavelet transform processing on the original vehicle speed sequences under different working conditions to obtain vehicle speed signals after reconstruction and noise reduction; carrying out matrix conversion related to a Gramer angle field on a plurality of subsequences in the vehicle speed signal under the typical working condition to obtain a two-dimensional image; extracting and calculating the spatial features and the time features in the two-dimensional image by using a hybrid neural network model to obtain a predicted value of the abrasion loss of the brake pad in the vehicle braking system at the current moment; and brake gain adjustment is conducted on a vehicle brake system, and the final brake torque is obtained.

Description

Technical Field

[0001] This application relates to the field of predicting the wear degree of brake pads, and particularly to a method and device for predicting the wear degree of vehicle brake pads based on working condition identification. Background Art

[0002] The brake pads (brake linings) in a vehicle braking system are key components related to driving safety. Their wear degree determines the braking performance and service life, and need to be detected during vehicle maintenance or servicing. How to evaluate the wear degree of brake pads during the vehicle use stage is a current difficult problem.

[0003] In recent years, researchers have begun to explore indirectly inferring the wear degree of brake pads based on existing vehicle operation data (such as vehicle speed, acceleration, braking pressure, ambient temperature, etc.), reducing the detection of the brake pad body itself. Existing technologies often require installing sensors to predict the wear of brake pads, but this increases the system cost and complexity. Moreover, the performance degradation of brake pads will affect the performance of the braking system, bringing potential safety hazards to the safe driving of drivers. At the same time, in the existing prediction process of brake pads, it is difficult to improve the accuracy and robustness of predicting the wear degree of brake pads, and it is difficult to realize adjusting and optimizing the braking performance of the braking system by using the prediction results for compensation and braking gain. Summary of the Invention

[0004] Embodiments of this application provide a method and device for predicting the wear degree of vehicle brake pads based on working condition identification, which are used to solve the following technical problems: In the existing vehicle braking system, the friction prediction of brake pads is relatively difficult during the use stage, and it is difficult to improve the accuracy and robustness of predicting the wear degree of brake pads, which is not conducive to efficiently and economically completing the wear prediction and optimizing the performance of the braking system.

[0005] Embodiments of this application adopt the following technical solutions:

[0006] On the one hand, an embodiment of the present application provides a method for predicting the wear degree of vehicle brake pads based on driving condition recognition, including: clustering and selecting the vehicle speed sequence under different types of driving scenarios to maximize the difference, obtaining the original vehicle speed sequence; performing wavelet transform processing on the original vehicle speed sequence under different working conditions to obtain a reconstructed and noise-reduced vehicle speed signal; performing matrix conversion related to the Gram angle field on multiple subsequences in the vehicle speed signal to obtain a two-dimensional image; calculating the output of the feature vector sequence in the two-dimensional image through a pre-trained CNN+LSTM hybrid neural network model to obtain the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment; performing deviation scaling processing related to the load correction weight on the predicted wear amount value to determine the current cumulative wear amount; adjusting the braking gain of the vehicle braking system according to the current cumulative wear amount to obtain the final braking torque, so as to complete the braking control of the vehicle.

[0007] The embodiment of the present application uses the existing vehicle speed data of the vehicle for wear prediction, avoiding additional hardware installation and reducing the system cost and complexity. The density clustering algorithm DBSCAN is used to classify the generated working condition data, and the typical working conditions with the largest difference are screened out to ensure the diversity and representativeness of the data set and improve the generalization ability of the model. Combining the convolutional neural network (CNN) to extract the spatial features of the GASF image and the long short-term memory network (LSTM) to model the time dependence, a hybrid neural network model is formed, significantly improving the accuracy and robustness of wear prediction. Through the data feedback mechanism, the system can continuously collect data in actual use for model retraining and optimization to ensure the long-term effectiveness and adaptability of the prediction system. At the same time, through adaptive gain compensation, in the case of abnormal wear, the deceleration response of the vehicle can still be maintained to the maximum extent, avoiding a sharp decline in braking performance caused by wear.

[0008] In a feasible implementation manner, according to the driving scenarios under different types, clustering selection is performed on the vehicle speed sequence under the maximization of differences to obtain the original vehicle speed sequence, which specifically includes: discretely partitioning the vehicle speed data of the vehicle according to the driving scenarios under different types to obtain a plurality of speed intervals; constructing a speed transition probability matrix based on the actual road test working condition data; performing state transition processing on the plurality of speed intervals from the current moment to the next moment through the speed transition probability matrix to obtain a series of vehicle speed sequences, where the series of vehicle speed sequences represent the speed changes of the vehicle under different working conditions; extracting the speed characteristics in the series of vehicle speed sequences and constructing a speed feature vector based on the speed characteristics, where the speed characteristics at least include: average vehicle speed, vehicle speed variance, and acceleration frequency; calculating the Euclidean distance of the speed feature vectors in the driving scenarios under different types, and performing clustering processing on all the series of vehicle speed sequences according to the DBSCAN parameters defined by the Euclidean distance to obtain a vehicle speed clustering result; selecting a typical working condition type with the strongest representativeness of the cluster center point and the largest difference between clusters, and determining the original vehicle speed sequence corresponding to the typical working condition type, where the typical working condition type covers the driving scenarios under different types.

[0009] In a feasible implementation manner, wavelet transform processing is performed on the original vehicle speed sequence under different working conditions to obtain a vehicle speed signal after reconstruction and noise reduction, which specifically includes: performing soft threshold transformation processing on the original vehicle speed sequence through a preset Daubechies wavelet for the detail coefficients to obtain soft threshold detail coefficients; According to Obtain the vehicle speed signal where c j (t) is the approximation coefficient of the wavelet transform; d′ j (t) is the soft threshold detail coefficient; j represents the decomposition level; J represents the total decomposition level.

[0010] In a feasible implementation manner, matrix conversion related to the Gram angle field is performed on multiple subsequences in the vehicle speed signal to obtain a two-dimensional image, which specifically includes: performing segmented processing on the vehicle speed sequence in the vehicle speed signal with a fixed length to obtain the multiple subsequences, where each subsequence index in the multiple subsequences contains a segmented vehicle speed value; According to Obtain each vehicle speed value V′ k,i The angle φ mapped to the polar coordinate k,i ; where V′ min and V′ max are respectively the minimum value and the maximum value of the subsequence S k ; V′ k,i is the vehicle speed value in each segment; According to G k,ij= cos(φ k,i + φ k,j ), i, j ∈ {1, 2, …, N}, to obtain the relationship between the row and column elements of the GASF matrix G k,ij ; where i is the row element of the GASF matrix, and j is the column element of the GASF matrix; G k,ij also represents the mutual relationship of the vehicle speed sequence at different time steps; through the relationship between the row and column elements, map the GASF matrix corresponding to the vehicle speed value to a grayscale image to obtain a two-dimensional image based on the vehicle speed signal.

[0011] In a feasible implementation manner, before calculating the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment by outputting and calculating the feature vector sequence in the two-dimensional image through a pre-trained CNN + LSTM hybrid neural network model, the method further includes: constructing a comprehensive wear amount evaluation index D according to D = w1·ΔD + w2·Δμ + w3·ΔT; where ΔD is the comprehensive thickness change of the brake pads; Δμ is the change in mechanical properties of the brake pads at different wear stages; w1, w2, and w3 are all weight coefficients; ΔT is the change in heat fade performance due to friction.

[0012] In a feasible implementation manner, calculating the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment by outputting and calculating the feature vector sequence in the two-dimensional image through a pre-trained CNN + LSTM hybrid neural network model specifically includes: extracting the spatial features in the two-dimensional image through a convolutional neural network to obtain a feature vector sequence; inputting and combining the comprehensive wear amount evaluation index into a long short-term memory network; through the long short-term memory network, performing output prediction processing on the feature vector sequence under the time-dependent relationship; where the hybrid neural network model includes: the convolutional neural network and the long short-term memory network; according to to obtain the predicted wear amount of the brake pads where, W fc is the weight of the fully connected layer in the long short-term memory network; b fc is the bias term; h K is the final hidden state of the long short-term memory network; according to

[0013] to obtain the minimized loss function MSE for the brake pad wear amount; where D i is the true wear amount, is the predicted wear amount, and n is the number of samples; according to to obtain the coefficient of determination R 2 ; where, is the average value of the true wear amount; based on the minimized loss function and the coefficient of determination, the parameters of the optimizer in the hybrid neural network model are updated, and based on the optimized long short-term memory network after iterative optimization, the predicted wear amount of the brake pads in the vehicle braking system at the current moment is output.

[0014] In a feasible implementation manner, a deviation scaling process related to the load correction weight is performed on the predicted wear amount value to determine the current cumulative wear amount, which specifically includes: According to Obtain the fusion result of the historical load and the current load where λ is the smoothing coefficient; M(t) is the load; is the smoothed average load in the previous time interval; According to Obtain the load correction weight w L (t); where M full is the full load mass; γ is the sensitivity coefficient; R is a real number; According to Obtain the current cumulative wear amount D current ; where is the predicted wear amount value at the current moment; D previous is the wear amount increment at the historical moment.

[0015] In a feasible implementation manner, according to the current cumulative wear amount, the braking gain of the vehicle braking system is adjusted to obtain the final braking torque, which specifically includes: If the current cumulative wear amount is greater than or equal to the safety threshold, an adaptive compensation coefficient related to the braking torque is constructed based on the ratio between the current cumulative wear amount and the safety threshold; According to T raw (t) = G0·u(t), obtain the original braking torque T raw (t); where u(t) represents the pedal displacement input of the driver to the brake pedal at time t; G0 is the standard braking gain of the vehicle in the normal state of the brake pads; According to β(t)·T raw (t), obtain the adaptive amplified torque where β(t) is the adaptive compensation coefficient; According to Obtain the final braking torque T cmd (t); where T max is the maximum allowable braking torque according to the vehicle's comprehensive hardware factors; Through the final braking torque, the braking force compensation control of the brakes in the vehicle braking system is performed to complete the braking control of the vehicle.

[0016] In a feasible implementation, if the current cumulative wear amount is greater than or equal to the safety threshold, the generated brake pad maintenance information is visually reminded through the in-vehicle display screen; and based on the numerical range interval where the adaptive compensation coefficient is located, the safe driving level of the vehicle is generated.

[0017] On the other hand, an embodiment of the present application also provides a vehicle brake pad wear degree prediction device based on working condition recognition, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute a vehicle brake pad wear degree prediction method based on working condition recognition described in any one of the above embodiments.

[0018] The present application provides a vehicle brake pad wear degree prediction method, device and medium. Compared with the prior art, the embodiments of the present application have the following beneficial technical effects:

[0019] 1. There is no need to install sensors on the brake: By using the existing vehicle speed data for wear prediction, additional hardware installation is avoided, reducing the system cost and complexity.

[0020] 2. Differential working condition screening based on DBSCAN: The density clustering algorithm DBSCAN is used to classify the generated working condition data, and the typical working conditions with the greatest difference are screened out to ensure the diversity and representativeness of the data set and improve the generalization ability of the model.

[0021] 3. The time series is converted into a two-dimensional image through the GASF method, making full use of the advantages of CNN in image feature extraction.

[0022] 4. Combination of deep learning models of CNN and LSTM: Combining the convolutional neural network (CNN) to extract the spatial features of the GASF image and the long short-term memory network (LSTM) to model the time dependence, a hybrid neural network model is formed, significantly improving the accuracy and robustness of wear prediction.

[0023] 5. The system can perform wear prediction in real time during the actual operation of the vehicle, and automatically trigger maintenance suggestions based on the prediction results, realizing intelligent brake system health management and brake performance optimization.

[0024] 6. Continuous learning and model optimization mechanism: Through the data feedback mechanism, the system can continuously collect data in actual use, perform model retraining and optimization, and ensure the long-term effectiveness and adaptability of the prediction system.

[0025] 7. Based on the predicted data, through adaptive gain compensation, in the case of abnormal wear, the vehicle's deceleration response can still be maintained to the maximum extent, avoiding a sharp decline in braking performance caused by wear. Description of the Drawings

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings described below are only some embodiments recorded in the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0027] Figure 1 It is a flowchart of a method for predicting the wear degree of vehicle brake pads based on driving condition recognition provided by an embodiment of the present application;

[0028] Figure 2 It is a schematic diagram of a driving condition speed curve provided by an embodiment of the present application;

[0029] Figure 3 It is a schematic diagram of the wear prediction error effect under various typical driving conditions provided by an embodiment of the present application;

[0030] Figure 4 It is a schematic structural diagram of a device for predicting the wear degree of vehicle brake pads based on driving condition recognition provided by an embodiment of the present application. Detailed Embodiments

[0031] In order to enable those skilled in the art to better understand the technical solutions in the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts should belong to the scope of protection of the present application.

[0032] The embodiment of the present application provides a method for predicting the wear degree of vehicle brake pads based on driving condition recognition, as Figure 1 shown, a method for predicting the wear degree of vehicle brake pads based on driving condition recognition specifically includes steps S101 - S106:

[0033] S101. According to different types of driving scenarios, perform clustering selection on the vehicle speed sequence under the maximization of differences to obtain the original vehicle speed sequence.

[0034] Specifically, it is necessary to first discretize the vehicle speed data according to different types of driving scenarios to obtain multiple speed intervals. And based on the actual road test condition data, construct a speed transition probability matrix.

[0035] Furthermore, use the speed transition probability matrix to perform state transition processing on multiple speed intervals from the current moment to the next moment, obtaining a series of vehicle speed sequences. Among them, the series of vehicle speed sequences represents the speed changes of the vehicle under different working conditions.

[0036] In one embodiment, first discretize the vehicle speed V into S speed intervals {V1, V2, …, V S}, that is, divide the working conditions into sub - segments. Based on the actual road test condition data, construct a speed transition probability matrix P, and its elements represent the probability of transitioning to state j at the next moment when currently in state i. Satisfy: Then use the above - mentioned transition probability matrix to generate a series of vehicle speed sequences Thus, the speed changes of the vehicle under different working conditions are simulated.

[0037] Furthermore, extract the speed characteristics from the series of vehicle speed sequences, and based on the speed characteristics, construct a speed feature vector. Among them, the speed characteristics at least include: average vehicle speed, vehicle speed variance, and acceleration frequency.

[0038] Furthermore, it is also necessary to calculate the Euclidean distance of the speed feature vectors in different types of driving scenarios, and perform clustering processing on all series of vehicle speed sequences according to the DBSCAN parameters defined by the Euclidean distance, obtaining the vehicle speed clustering result.

[0039] Furthermore, select the typical working condition types with the strongest representativeness of the cluster center points and the largest differences between clusters, and determine the original vehicle speed sequences corresponding to the typical working condition types. Among them, the typical working condition types cover different types of driving scenarios.

[0040] In one embodiment, extract features from the generated series of vehicle speed sequences, including but not limited to the average vehicle speed vehicle speed variance acceleration frequency f A etc., to form a feature vector Then calculate the distance metric between samples, usually using the Euclidean distance d(x i , x j ): Among them, x i and x jThey are the feature vectors of the i-th and j-th samples respectively, and K is the feature dimension. Next, the DBSCAN parameters are set: Set the radius parameter ∈: Define the maximum distance of the sample neighborhood. Set the minimum number of samples MinPts: Define the minimum number of samples required to form a cluster. For each sample point, find all its neighbors within the ∈ neighborhood. If the number of points in the neighborhood ≥ MinPts, mark this point as a core point and start forming a cluster. Classify the core point and all points within its neighborhood into the same cluster. Repeat the above process until all points are classified or marked as noise.

[0041] In one embodiment, several working condition categories with the strongest representativeness of the cluster center point and the largest difference between clusters are selected from the DBSCAN clustering results (vehicle speed clustering results) to ensure that the selected working conditions cover different driving scenarios (such as high-speed stable working conditions, frequent start-stop working conditions, etc.). Record the original vehicle speed sequence corresponding to each typical working condition. For subsequent wear tests and model training.

[0042] S102. Perform wavelet transform processing on the original vehicle speed sequences under different working conditions to obtain the vehicle speed signal after reconstruction and noise reduction.

[0043] Specifically, first perform soft threshold transformation processing on the original vehicle speed sequence through a preset Daubechies wavelet to obtain soft threshold detail coefficients.

[0044] Further, according to Obtain the vehicle speed signal where, c j (t) is the approximation coefficient of the wavelet transform. d′ j (t) is the soft threshold detail coefficient; j represents the decomposition level; J represents the total decomposition level.

[0045] In one embodiment, multiple braking tests can be carried out in a laboratory or a closed test site according to the selected typical working condition sequence, and record the vehicle speed V t , braking force F b and data such as the change T of the thermal decay performance, etc. during each braking process. Then, take the cumulative wear amount D corresponding to each working condition sequence as a label to form the corresponding relationship between the working condition data and the wear amount. Next, to remove the high-frequency noise in the vehicle speed data, wavelet transform is used for multi-scale decomposition and threshold processing. Let the original vehicle speed time series be Select Daubechies wavelet for wavelet decomposition, and obtain the approximation coefficient {c j (t)} and detail coefficient {d j (t)} based on the existing classical algorithm, where j represents the decomposition level. Perform threshold processing on the detail coefficient d j (t), and use the soft threshold method to determine the soft threshold detail coefficient d′j (t):

[0046] where λ is the threshold and sgn(□) is the sign function. Finally, calculate the vehicle speed signal after reconstruction and noise reduction

[0047] S103. Perform matrix transformation related to the Gramian Angular Summation Field (GASF) on multiple subsequences in the vehicle speed signal to obtain a two-dimensional image.

[0048] Specifically,[[]] Figure 2 is a schematic diagram of a working condition speed curve provided by an embodiment of the present application. As Figure 2 shown, it is also necessary to perform segmentation processing on the vehicle speed sequence in the vehicle speed signal under a fixed length to obtain multiple subsequences. Among them, each subsequence index in the multiple subsequences contains a segmented vehicle speed value.

[0049] In one embodiment, segment the noise-reduced vehicle speed sequence by a fixed length N to obtain multiple

[0050] subsequences S k =[V' k,1 ,V' k,2 ,…,V' k,N , where k is the subsequence index, and

[0051] Furthermore, according to obtain the angle φ k,i to which each vehicle speed value V' k,i is mapped in polar coordinates. Among them, V' min and V' max are respectively the minimum value and the maximum value of the subsequence S k ; V' k,i is the vehicle speed value in each segment.

[0052] Furthermore, according to G k,ij =cos(φ k,i +φ k,j ), i,j∈{1,2,…,N}, obtain the row-column element relationship G k,ij of the GASF matrix. Among them, i is the row element of the GASF matrix, and j is the column element of the GASF matrix; G k,ij also represents the mutual relationship of the vehicle speed sequence at different time steps.

[0053] Furthermore, through the row-column element relationship, perform gray-scale image mapping processing on the GASF matrix corresponding to the vehicle speed value to obtain a two-dimensional image based on the vehicle speed signal.

[0054] In one embodiment, the GASF matrix G k is regarded as a two-dimensional grayscale image. To meet the input requirements of the CNN, the matrix G k is mapped to a grayscale image, that is, a two-dimensional image, and the pixel value range is [0, 1].

[0055] S104. Through the pre-trained CNN + LSTM hybrid neural network model, perform output calculation on the feature vector sequence in the two-dimensional image to obtain the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment.

[0056] Specifically, it is necessary to first construct a comprehensive wear amount evaluation index D according to D = w1·ΔD + w2·Δμ + w3·ΔT. Among them, ΔD is the comprehensive thickness change of the brake pads; Δμ is the change in the mechanical properties of the brake pads at different wear stages; w1, w2, and w3 are all weight coefficients; ΔT is the change in the heat fade performance due to friction.

[0057] In one embodiment, to achieve a quantitative evaluation of the wear amount of the brake pads, it is necessary to comprehensively consider the thickness change and mechanical property change of the brake pads, as well as the heat fade performance change caused by friction, and construct a quantitative wear amount evaluation index. Thickness change measurement: Regularly measure the thickness D of the brake pads using precision measurement equipment and record the thickness values D t at different time points. Mechanical property measurement: Measure the mechanical property parameters of the brake pads, such as the friction coefficient μ, at different wear stages and record the change amount Δμ.

[0058] Furthermore, through the convolutional neural network, extract and process the spatial features in the two-dimensional image to obtain a feature vector sequence. Input and combine the comprehensive wear amount evaluation index into the long short-term memory network. Through the long short-term memory network, perform output prediction processing on the feature vector sequence under the time-dependent relationship. Among them, the hybrid neural network model includes: a convolutional neural network and a long short-term memory network.

[0059] In one embodiment, the convolutional neural network (CNN) part is responsible for extracting the spatial features of the GASF image. The specific architecture is as follows: Input layer: Accept image data with a size of N×N×C, where C is the number of channels (such as 1 or 3). Convolutional layer: Apply multiple convolutional kernels K l to perform convolution operations and extract the feature map F l : where σ is the activation function (such as ReLU), M is the number of input channels, and b lis the bias term. Pooling layer: Max pooling or average pooling is used to reduce the size of the feature map and retain the main features. Activation function: The ReLU activation function is used: ReLU(x) = max(0, x). Flattening layer: Flatten the multi-dimensional feature map into a one-dimensional vector v k .

[0060] In one embodiment, the long short-term memory network (LSTM) part is responsible for modeling the temporal dependencies of the feature vector sequence, and the specific architecture is as follows:

[0061] ① Input layer: Accept the feature vector sequence {v1, v2,..., v K}} extracted by the CNN, where K is the number of time steps.

[0062] ② LSTM cell: Captures long-term dependencies through the memory gate mechanism, and its core calculation formulas are as follows:

[0063] f t = σ(W f · [h t-1 , v t + b f );

[0064] i t = σ(W i · [h t-1 , v t + b i );

[0065]

[0066] o t = σ(W o · [h t-1 , v t + b o );

[0067] h t = o t ⊙ tanh(C t );

[0068] Where: f t : forget gate; i t : input gate; candidate memory unit; C t : current memory unit; o t : output gate; h t : current hidden state; σ: Sigmoid activation function; ⊙: element-wise product;

[0069] W f , W i , W C , Wo : The weight matrix of each gate of the LSTM; b f , b i , b C , b o : The bias vector of each gate of the LSTM; v t : The input feature vector at the t-th time step.

[0070] ③Fully connected layer: Map the final hidden state h of the LSTM K to the output layer to obtain the predicted value of the brake pad wear

[0071] Furthermore, according to obtain the predicted wear of the brake pad where, W fc is the weight of the fully connected layer in the long short-term memory network. b fc is the bias term. h K is the final hidden state of the long short-term memory network.

[0072] Furthermore, use the Mean Squared Error (MSE) as the loss function: According to obtain the minimized loss function MSE for the brake pad wear. Where, D i is the true wear, is the predicted wear, and n is the number of samples.

[0073] Furthermore, use the Adam optimizer to update the parameters, and the optimization goal is to minimize the loss function. Input the GASF images in the training set into the CNN to extract spatial features, then input the sequence of feature vectors into the LSTM for temporal modeling, and finally output the predicted wear value. Optimize the model parameters through multiple epochs, monitor the model performance on the validation set during this period to prevent overfitting. When the performance on the validation set is the best, save the current model parameters as the optimal model. Evaluate the prediction performance of the model on the test set, and use metrics such as the Mean Squared Error (MSE) and the coefficient of determination (R 2 ) for evaluation: According to obtain the coefficient of determination R of the brake pad wear 2 . Where, is the average value of the true wear.

[0074] Furthermore, based on minimizing the loss function and the coefficient of determination, update the parameters of the optimizer in the hybrid neural network model, and based on the optimized long short-term memory network, output the predicted wear value of the brake pads in the vehicle braking system at the current moment.

[0075] As a feasible implementation, the generated GASF image G tInput the pre-trained CNN model to extract the spatial feature vector v t Input a sequence of K consecutive feature vectors {v t-K+1 , v t-K+2 , …, v t} into the LSTM model to output the predicted value of the brake pad wear at the current moment

[0076] S105. Perform a deviation scaling process on the predicted wear value with respect to the load correction weight to determine the current cumulative wear

[0077] Specifically, first perform a smoothing or weighted fusion on M(t) to obtain According to Obtain the fusion result of the historical load and the current load where λ is the smoothing coefficient, which can be selected according to the need for sensitivity to real-time changes; M(t) is the load; is the average load smoothed in the previous time interval

[0078] Furthermore, since the pre-trained model is based on the full-load condition, when the actual load of the vehicle is lower than the full load, the wear prediction will be appropriately reduced; if the actual load of the vehicle exceeds the full load (overload), the wear will increase accordingly. A correction function with a sensitivity coefficient γ is proposed: Obtain the load correction weight w L (t). Where M full is the full-load mass; γ is the sensitivity coefficient; R is a real number. The linear form can conveniently use γ to control the adjustment strength of the wear when the load changes and will not overshoot. γ is calibrated according to actual experimental data, bench tests, or historical operation data

[0079] Furthermore, since the full-load vehicle is used when pre-training the CNN+LSTM hybrid neural network model based on experiments, when the load is small, the wear amount is relatively reduced. It is necessary to make the model have a certain adjustability to the load sensitivity. In the t-th time interval, the original predicted wear value is denoted as The historical cumulative wear is D previous , then the load correction can be incorporated into the following formula:

[0080] Obtain the current cumulative wear D current . Where is the predicted wear value at the current moment; D previous is the wear increment at the historical moment

[0081] S106. According to the current cumulative wear, adjust the braking gain of the vehicle braking system to obtain the final braking torque to complete the braking control of the vehicle

[0082] Specifically, if the current cumulative wear amount is greater than or equal to the safety threshold, an adaptive compensation coefficient for the braking torque is constructed based on the ratio between the current cumulative wear amount and the safety threshold.

[0083] Furthermore, it is also necessary to obtain the original braking torque T raw (t) = G0 · u(t), where u(t) represents the pedal displacement input of the driver to the brake pedal at time t. G0 is the standard braking gain of the vehicle in the normal state of the brake pads. raw (t).

[0084] Furthermore, it is also necessary to obtain the adaptive amplified torque where β(t) is the adaptive compensation coefficient. where β(t) is the adaptive compensation coefficient.

[0085] Furthermore, by using the final braking torque T cmd (t) is obtained. Where T max is the maximum allowable braking torque based on the comprehensive hardware factors of the vehicle.

[0086] Furthermore, in combination with the final braking torque, braking force compensation control of the brake of the vehicle braking system is carried out to complete the braking control of the vehicle.

[0087] In one embodiment, to maintain the original braking effect within a certain range, when it is detected that D current ≥ D threshold , the constructed adaptive compensation coefficient β(t) is used to amplify and correct the braking gain. The specific definition is as follows: When the degree of wear intensifies and D current is much higher than D threshold , β(t) will exceed 1, thereby appropriately amplifying the brake pedal input. If the vehicle is in an extremely low friction coefficient or severe wear situation, to prevent out-of-control caused by excessive amplification, an upper limit β max is set for β(t), and β nax is calibrated by design or experimental research to ensure that the gain amplification is within a safe and controllable range.

[0088] In one embodiment, let u(t) represent the pedal displacement input of the driver to the brake pedal at time t, and G0 be the nominal braking gain of the vehicle in the normal state of the brake pads. The original braking command (braking torque) can usually be written as: T raw (t) = G0 · u(t). To make up for the insufficient braking force caused by brake wear, when β(t) exceeds 1, the original torque is adaptively amplified: Through the above amplification, the vehicle can still achieve the braking performance close to the original design in the medium or light wear stage. At the same time, when the wear is serious, if the command is only increased blindly, it may cause brake overload or tire locking. Based on the compensated torque, the maximum allowable braking torque T is set. max Limit the amplitude to obtain the final braking torque T cmd (t): Among them, T max It can be set according to comprehensive factors such as vehicle structure, load, thermal decay, etc.

[0089] As a feasible implementation method, if the current accumulated wear is greater than or equal to the safety threshold, the generated brake pad maintenance information is visually reminded through the vehicle display screen, and the vehicle's safe driving level is generated based on the numerical range of the adaptive compensation coefficient.

[0090] In one embodiment, Figure 3 The schematic diagram of wear prediction error effect under various typical working conditions provided in the embodiments of the present application is as follows: Figure 3 As shown, in an example experiment: the experimental data is divided into a training set, a validation set, and a test set. A part of the data (60%) is used as a training set, and the GASF image is extracted and input into CNN. The feature vector sequence output by CNN is then input into LSTM for time-dependent training. Another part of the data (20%) is used as a validation set to adjust the model hyperparameters and monitor the early stopping strategy to prevent overfitting. The remaining data (about 20%) is reserved as a test set to evaluate the final prediction performance of the model. The experiment was conducted on a chassis dynamometer based on the working conditions collected from actual vehicle working conditions (that is, the brake wear conditions were collected while testing the fuel economy of the vehicle). The actual working condition fragments were clustered into 20 typical working conditions through DBSCAN, where each typical working condition in the test set contained 5 to 20 test data, including the working condition data and its corresponding actual brake wear. The method of this patent was applied to a test set consisting of real data for testing, and the results are as follows Figure 3 As shown. The figure shows 20 grouped boxes, each box corresponds to the prediction error distribution under a typical working condition. The upper and lower edges of the box represent the upper quartile and lower quartile of the data group, respectively, and the horizontal line in the box is the median. The vertical dotted line shows the maximum and minimum observed values ​​of the data within the normal range. The results show that the prediction error of vehicle brake wear under various working conditions is within ±30%, and most of them are within ±20%. Among them, the prediction effect for typical working conditions 1, 2, 6, and 8 is the best, and the error is within ±15%. The experiment shows that the prediction results of the method of the present invention are well consistent with the actual results, and the method has a high accuracy in predicting high-wear brakes. The method can achieve the goal of predicting vehicle brake pad wear based on working conditions.

[0091] In addition, the embodiment of the present application also provides a vehicle brake pad wear degree prediction device based on driving condition recognition, as Figure 4 shown. The vehicle brake pad wear degree prediction device 400 based on driving condition recognition specifically includes:

[0092] At least one processor 401. And a memory 402 communicatively connected to the at least one processor 401. Wherein, the memory 402 stores instructions that can be executed by the at least one processor 401, so that the at least one processor 401 can execute:

[0093] According to the driving scenarios of different types, perform clustering selection on the vehicle speed sequence under the maximization of differences to obtain the original vehicle speed sequence;

[0094] Perform wavelet transform processing on the original vehicle speed sequence under different working conditions to obtain a reconstructed and noise-reduced vehicle speed signal;

[0095] Perform matrix conversion related to the Gram angle field on multiple subsequences in the vehicle speed signal to obtain a two-dimensional image;

[0096] Through a pre-trained CNN+LSTM hybrid neural network model, perform output calculation on the feature vector sequence in the two-dimensional image to obtain the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment;

[0097] Perform deviation scaling processing on the predicted wear amount related to the load correction weight to determine the current cumulative wear amount;

[0098] According to the current cumulative wear amount, adjust the braking gain of the vehicle braking system to obtain the final braking torque, so as to complete the braking control of the vehicle.

[0099] The embodiment of the present application also has the following:

[0100] 1. Introduce DBSCAN to cluster a large number of vehicle speed sequences, which can automatically identify and select "typical working conditions" with significant differences between clusters. Compared with the traditional screening method based on manual or a small number of clustering centers, DBSCAN does not require specifying the number of clusters in advance and can better cover diverse driving scenarios (such as high-speed stability, frequent starting and stopping, etc.).

[0101] 2. Not only measure the thickness change of the brake pads, but also consider the influence of the wear process on the braking performance and thermal fade performance; a more quantitative index that can better reflect the true wear degree is comprehensively obtained through the weight coefficient, which is more accurate and comprehensive.

[0102] 3. Samples are taken for each type of "typical working condition" in a laboratory or closed test site and multiple braking experiments are carried out. Information on vehicle speed and wear degree at different time intervals is collected, and the final cumulative wear amount is matched with the working condition sequence. The constructed dataset has the advantages of high quality and being supervised.

[0103] 4. The Gramian Angular Field (GAF) method can graphically represent the global correlation of time series "simultaneously" on a two-dimensional map, providing the possibility for a deep convolutional neural network to extract the spatial structure features therein. Mapping one-dimensional data into a two-dimensional space can capture more internal correlation patterns of time series.

[0104] 5. Since the GAF image matrix captures the mutual relationship of the working condition time series at different time steps, re-extracting features of this time series dependence at the two-dimensional image level, potential feature patterns can be effectively extracted through a convolutional neural network.

[0105] 6. After each GAF image is used to extract features by a convolutional neural network, a set of feature two-dimensional vectors is obtained, and then these vectors are fed into an LSTM network in chronological order. The hybrid model can better balance the local spatial structure and the dependence across time steps, and has higher prediction accuracy for phenomena such as vehicle braking wear that accumulate over time.

[0106] 7. "Quantify" the brake pad wear amount into an input parameter for real-time control, and perform real-time correction of the braking torque according to the wear degree, so as to make up for the impact of wear on vehicle braking performance, effectively delay or alleviate the driving safety risks brought by severe brake pad wear, and at the same time can trigger prompts for maintenance, replacement, etc.

[0107] 8. By using the existing vehicle speed data for wear prediction, additional hardware modification is avoided, and the system cost and complexity are reduced.

[0108] The embodiments in this application are all described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for the relevant parts, reference can be made to the description of the method embodiments.

[0109] The above describes specific embodiments of the present application. In some cases, the actions or steps recorded in the specification can be executed in a different order from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0110] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the embodiments of the present application shall be included in the description of the present application.

[0111] It should be noted that this application is funded by the Shandong Natural Science Foundation (ZR2023QE208), the China Postdoctoral Science Foundation (2024M751577), and the Key Research and Development Plan of Shandong Province (2022CXGC020302) to conduct research on the dynamic performance during vehicle braking drive.

Claims

1. A method for predicting the wear degree of vehicle brake pads based on working condition recognition, characterized in that, The method includes: According to the driving scenarios of different types, perform clustering selection with maximum difference on the vehicle speed sequence to obtain the original vehicle speed sequence; Perform wavelet transform processing on the original vehicle speed sequence under different working conditions to obtain the reconstructed and noise-reduced vehicle speed signal; Perform matrix conversion related to the Gram angle field on multiple subsequences in the vehicle speed signal to obtain a two-dimensional image; Through a pre-trained CNN+LSTM hybrid neural network model, perform output calculation on the feature vector sequence in the two-dimensional image to obtain the predicted wear amount of the brake pads in the vehicle braking system at the current moment; Perform deviation scaling processing related to the load correction weight on the predicted wear amount to determine the current cumulative wear amount; According to the current cumulative wear amount, adjust the braking gain of the vehicle braking system to obtain the final braking torque to complete the braking control of the vehicle.

2. The method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, wherein According to the driving scenarios of different types, perform clustering selection with maximum difference on the vehicle speed sequence to obtain the original vehicle speed sequence, specifically including: According to the driving scenarios of different types, discretize the vehicle speed data to obtain multiple speed intervals; and based on the actual road test working condition data, construct a speed transition probability matrix; Through the speed transition probability matrix, perform state transition processing on the multiple speed intervals from the current moment to the next moment to obtain a series of vehicle speed sequences; wherein, the series of vehicle speed sequences represent the speed changes of the vehicle under different working conditions; Extract the speed features in the series of vehicle speed sequences, and based on the speed features, construct a speed feature vector; wherein, the speed features at least include: average vehicle speed, vehicle speed variance, and acceleration frequency; Calculate the Euclidean distance of the speed feature vectors in the driving scenarios of different types, and according to the DBSCAN parameters defined by the Euclidean distance, perform clustering processing on all the series of vehicle speed sequences to obtain the vehicle speed clustering result; Select the typical working condition type with the strongest representativeness of the cluster center point and the largest difference between clusters, and determine the original vehicle speed sequence corresponding to the typical working condition type; wherein, the typical working condition type covers the driving scenarios of different types.

3. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, Perform wavelet transform processing on the original vehicle speed sequence under different working conditions to obtain the reconstructed and noise-reduced vehicle speed signal, specifically including: Through a preset Daubechies wavelet, perform soft threshold transform processing related to the detail coefficients on the original vehicle speed sequence to obtain soft threshold detail coefficients; According to obtain the vehicle speed signal where c j (t) is the approximation coefficient of the wavelet transform; d′ j (t) is the soft threshold detail coefficient; j represents the decomposition level; J represents the total decomposition level.

4. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, Perform matrix conversion related to the Gram angle field on multiple subsequences in the vehicle speed signal to obtain a two-dimensional image, specifically including: Perform segmented processing with a fixed length on the vehicle speed sequence in the vehicle speed signal to obtain the multiple subsequences; wherein, each subsequence index in the multiple subsequences contains a segmented vehicle speed value; According to each vehicle speed value V′ is obtained k,i and mapped to the angle φ in polar coordinates k,i ; where V′ min and V′ max are respectively the minimum value and the maximum value of the subsequence S k ; and V′ k,i is the vehicle speed value in each segment. According to G k,ij = cos(φ k,i + φ k,j ), i, j ∈ {1, 2, …, N}, the row-column element relationship G of the GASF matrix is obtained as G k,ij ; where i is the row element of the GASF matrix, and j is the column element of the GASF matrix; G k,ij also represents the mutual relationship of the vehicle speed sequences at different time steps; Through the row-column element relationship, perform gray-scale image mapping processing on the GASF matrix corresponding to the vehicle speed value to obtain a two-dimensional image based on the vehicle speed signal.

5. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, Before calculating the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment by outputting and calculating the feature vector sequence in the two-dimensional image through the pre-trained CNN+LSTM hybrid neural network model, the method further includes: Construct a comprehensive wear amount evaluation index D according to D = w1·ΔD + w2·Δμ + w3·ΔT; where ΔD is the comprehensive thickness change of the brake pads; Δμ is the change in mechanical properties of the brake pads at different wear stages; w1, w2, and w3 are all weight coefficients; ΔT is the change in heat fade performance due to friction.

6. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, Calculating the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment by outputting and calculating the feature vector sequence in the two-dimensional image through the pre-trained CNN+LSTM hybrid neural network model specifically includes: Extracting and processing the spatial features in the two-dimensional image through a convolutional neural network to obtain a feature vector sequence; Inputting and combining the comprehensive wear amount evaluation index into a long short-term memory network; Performing output prediction processing on the feature vector sequence through the long short-term memory network under the time-dependent relationship; where the hybrid neural network model includes: the convolutional neural network and the long short-term memory network; According to obtain the predicted wear amount of the brake pad where W fc is the weight of the fully connected layer in the long short-term memory network; b fc is the bias term; h K is the final hidden state of the long short-term memory network; According to the minimum loss function MSE of the brake pad wear is obtained; where D i is the true wear, is the predicted wear, and n is the number of samples; According to obtain the determination coefficient R of the wear amount of the brake pad 2 ; where is the average value of the true wear amount Based on the minimized loss function and the coefficient of determination, updating the parameters of the optimizer in the hybrid neural network model, and outputting the predicted value of the wear amount of the brake pads in the vehicle braking system at the current moment based on the long short-term memory network after optimization iteration.

7. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, Performing deviation scaling processing on the predicted value of the wear amount with respect to the load correction weight to determine the current cumulative wear amount, specifically including: According to obtain the fusion result of historical load and current load where λ is the smoothing coefficient; M(t) is the load; is the average load smoothed in the previous time interval; According to obtain the load correction weight w L (t); where M full is the full load mass; γ is the sensitivity coefficient; R is a real number; According to obtain the current cumulative wear amount D current ; wherein is the wear amount prediction value at the current moment; D previous is the wear amount increment at the historical moment.

8. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, Adjusting the braking gain of the vehicle braking system according to the current cumulative wear amount to obtain the final braking torque, specifically including: If the current cumulative wear amount is greater than or equal to the safety threshold, constructing an adaptive compensation coefficient for the braking torque based on the ratio between the current cumulative wear amount and the safety threshold; According to T raw (t) = G0·u(t), the original braking torque T raw (t) is obtained; where u(t) represents the pedal displacement input of the driver to the brake pedal at time t; G0 is the standard braking gain of the vehicle in the normal state of the brake pads; According to obtain the adaptive amplification torque where β(t) is the adaptive compensation coefficient; According to obtain the final braking torque T cmd (t); wherein, T max is the maximum allowable braking torque under the vehicle's comprehensive hardware factors according to the vehicle Performing braking force compensation control on the brakes of the vehicle braking system through the final braking torque to complete the braking control of the vehicle.

9. A method for predicting the wear degree of vehicle brake pads based on working condition recognition according to claim 1, characterized in that, If the current cumulative wear amount is greater than or equal to the safety threshold, visually reminding the generated brake pad maintenance information through an in-vehicle display screen; and generating the safe driving level of the vehicle based on the numerical range interval where the adaptive compensation coefficient is located.

10. A vehicle brake pad wear degree prediction device based on working condition recognition, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; where, The memory stores instructions executable by the at least one processor, enabling the at least one processor to execute a method for predicting the wear degree of vehicle brake pads based on working condition recognition according to any one of claims 1-9.

Images