Abrasion prediction and automatic compensation method and system for braking system of tracked vehicle

Abstract

The invention belongs to the technical field of vehicle braking systems, particularly relates to a tracked vehicle braking system wear prediction and automatic compensation method and system, and aims to solve the problems of low wear detection precision, compensation mechanism lag and the like in the prior art. A multi-source data fusion prediction model is constructed, multi-sensor data and a deep learning algorithm are fused, and the weight is adaptively adjusted in combination with working condition classification, so that real-time accurate prediction of the abrasion loss of the braking system is realized. A high-dynamic electromechanical actuator is used for conducting self-adaptive compensation on braking force, a double-closed-loop control structure is designed, and compensation precision is ensured. A model online updating and fault tolerance mechanism is introduced, the reliability of a braking system is improved, and the adaptability to vehicle complex working conditions and high-load impact is improved.

Description

Technical Field

[0001] This invention belongs to the field of vehicle braking system technology, specifically relating to a method and system for predicting and automatically compensating for wear in tracked vehicle braking systems. It achieves real-time wear prediction of tracked vehicle braking systems through multi-sensor data fusion and artificial intelligence algorithms, and uses a high-dynamic electromechanical actuator for adaptive compensation of braking force, making it particularly suitable for tracked vehicle braking systems under high load and complex working conditions. Background Technology

[0002] In tracked vehicle operation, the braking system is a key component ensuring safe driving and effective operation. Full disc brakes are widely used in tracked vehicles due to their advantages such as high braking force and good heat dissipation. However, with frequent use, components such as brake pads gradually wear down, which not only affects braking performance but may also lead to brake failure, endangering the safety of the vehicle and personnel.

[0003] Traditional methods for detecting and compensating for brake system wear mainly rely on regular manual inspections and adjustments. This approach suffers from problems such as untimely detection and poor accuracy. Furthermore, due to the complex and variable working environment of tracked vehicles, the wear condition of the brake system is affected by various factors, such as braking frequency, braking force, and road conditions, making it extremely difficult to accurately predict the degree of brake system wear.

[0004] With the rapid development of artificial intelligence technology, it has shown great potential in areas such as fault prediction and condition monitoring. Applying artificial intelligence technology to wear prediction and automatic compensation of the full disc brake system of tracked vehicles can achieve real-time monitoring and accurate prediction of the brake system's condition, and timely automatic compensation, thereby improving the reliability and safety of the brake system, which has important practical significance. Summary of the Invention

[0005] (a) Technical problems to be solved The technical problem this invention aims to solve is that tracked vehicles experience heavy braking loads and harsh environments, and the lack of effective models to predict the nonlinear relationship between brake pad wear and braking force loss makes it difficult to accurately predict wear trends, easily leading to decreased braking force or sudden failures. Existing methods for detecting brake pad wear in tracked vehicles mostly rely on periodic manual inspections or simple thickness sensors, resulting in low wear detection accuracy. Furthermore, braking force compensation typically uses fixed threshold adjustments, which cannot adapt to highly dynamically changing wear rates and complex operating conditions (such as mud, high temperatures, and long downhill slopes). The lack of automatic adjustment devices and adaptive compensation closed-loop control mechanisms that dynamically correlate wear and braking force leads to a lag in the compensation mechanism.

[0006] The purpose of this invention is to provide a method and system for predicting and automatically compensating for wear in the braking system of tracked vehicles based on artificial intelligence. By monitoring the operating parameters of the braking system in real time, the method uses artificial intelligence algorithms to accurately predict the wear degree of the braking system and achieve automatic compensation, thereby improving the reliability and safety of the braking system.

[0007] (II) Technical Solution To address the aforementioned technical problems, this invention provides a method for predicting and automatically compensating for wear in the braking system of tracked vehicles, the method comprising the following steps: Step 1: Collect real-time braking system operating data through a multi-sensor module, including brake pad temperature, pressure, displacement, vibration signals, and vehicle status parameters; Step 2: Preprocess the collected braking system operation data, including Kalman filtering for noise reduction, normalization, and multi-source data time-series alignment, to form aligned time-series data; Step 3: Input the preprocessed time series data into the wear prediction model based on deep learning and output the current wear amount; the wear state prediction model extracts local features by a one-dimensional convolutional neural network, captures temporal dependencies by a long short-term memory neural network, and dynamically weights key features through an attention mechanism; Step 4: If the predicted wear exceeds the preset threshold, a compensation command is generated based on the pre-calibrated wear-braking force mapping table. The linear motor stroke of the electromechanical actuator is adjusted through the dual closed-loop PID controller to dynamically adjust the brake pad position. At the same time, the actual braking force is fed back through the pressure sensor to achieve predictive-compensation-feedback coordinated control. Step 5: When the predicted wear amount is lower than the set threshold, switch to the physical model based on the friction coefficient-temperature equation for compensation control and trigger the online model update mechanism.

[0008] In step 3, the wear state prediction model adopts a dynamic weight allocation strategy, which adjusts the fusion weight of sensor data according to the vehicle operating condition classification results. Specifically, it includes: classifying operating condition categories based on vehicle speed, braking frequency and ambient temperature; and differentiating the weights of vibration signals and temperature signals through an attention mechanism for different operating conditions.

[0009] In step 4, the electromechanical actuator is integrated into the brake caliper and includes a linear motor, a ball screw, and a displacement sensor. The compensation range is ±2mm and the response time is ≤50ms.

[0010] The dual closed-loop PID controller in step 4 includes an inner-loop displacement control and an outer-loop braking force control. The inner-loop displacement control adjusts the motor stroke according to the compensation command, and the outer-loop braking force control feeds back the actual braking force through a pressure sensor.

[0011] Furthermore, the present invention also provides a wear prediction and automatic compensation system for the braking system of a tracked vehicle, the system comprising: Data acquisition layer: Collects brake pad temperature, pressure, displacement, vibration signals, and vehicle status information through a sensor array; Artificial Intelligence Processing Layer: After data acquisition and preprocessing, the artificial intelligence processing layer fuses vehicle braking temperature, pressure, vibration, displacement signals, and historical wear data to construct a deep learning-based wear state prediction model, achieving feature extraction and wear prediction. It also introduces a working condition classification module, including off-road / highway modes, to adaptively adjust model weights for different scenarios. Utilizing onboard edge computing devices and real-time incremental data training, the model is updated online to adapt to different driver habits and environmental changes. If the artificial intelligence prediction fails, it switches to a conservative compensation based on a physical model. If the predicted wear exceeds the threshold, a compensation command is triggered; Compensation execution layer: includes electromechanical compensation actuator and human-machine interface; a dual closed-loop control of wear prediction loop + braking force feedback loop is designed, the compensation displacement is calculated according to the wear amount-braking force mapping table, the linear motor stroke is adjusted by the dual closed-loop PID controller, and the actual braking force feedback correction is monitored.

[0012] The physical model is implemented using the friction coefficient-temperature equation.

[0013] In the data acquisition layer, Kalman filtering is used to denoise the vibration signal; Kalman filtering consists of two main steps: prediction and update. 1) Prediction Steps a. State prediction: (1) In the formula, It is a moment k State prediction, It is the system dynamic matrix. It is the optimal estimate of the state at the previous moment. It is a control input matrix. It is a control input; b. Covariance matrix prediction: (2) In the formula, It is the covariance matrix of the state prediction. It is the covariance matrix of the process noise; 2) Update steps a. Calculate the Kalman gain: (3) In the formula, It is Kalman gain. It is the observation matrix. It is the covariance matrix of the measured noise; b. Update state estimate: (4) In the formula, It is the optimal estimate of the updated state. It is the measurement value at the current moment; c. Update the covariance matrix: (5) In the formula, It is the covariance matrix of the updated state estimate. It is an identity matrix.

[0014] In the data acquisition layer, data standardization is performed; All Kalman filtered sensor data x After normalization, sensor data with all values ​​in the range [0, 1] are obtained: (6) In the formula, max ( x ) represents the maximum value of the original data, min( x ) represents the minimum value of the original data.

[0015] In the artificial intelligence processing layer, the wear state prediction model adopts a method combining a one-dimensional convolutional neural network and a long short-term memory neural network. 1) For the input time-series data, local features are first extracted by a one-dimensional convolutional neural network; the expression for the convolution operation is: (7) In the formula, Indicates the first l Layer output; Indicates the first l -1st floor i Output of each channel; express l -1st floor c One channel, and These represent the weights and biases, respectively. Pooling layers perform feature selection after convolutional layers, reducing the number of features; the average pooling calculation expression is: (8) In the formula, Indicates the first l Layer i The first channel tThe value of each neuron; S Indicates the size of the pooled kernel; Indicates the first l Layer i The first channel j The value of each neuron; 2) Secondly, temporal dependencies are captured through a long short-term memory neural network; let... Represents the previous cell state. Represents the output of the previous cell. Represents the input to the current cell; where , , These are the results for the forget gate, input gate, and output gate, respectively; the forget gate consists solely of the sigmoid activation function σ, while the input and output gates consist of σ and the activation function tanh; the input... The forward propagation process in a long short-term memory neural network is as follows: a. Forgot Gate, update calculation formula is: (9) In the formula, Represents the weight matrix, Indicates the bias term; b. Input gate, update the calculation formula as follows: (10) (11) (12) In the formula, W i and W c Represents the weight matrix; b i and b c Indicates the bias term; Indicates will x t and h t-1 State variables that integrate state information; c Indicates the current state; c. Output gate, update calculation formula is: (13) (14) In the formula, W o Represents the weight matrix; b o Indicates the bias term; ht Indicates the output value of the hidden layer; 3) In a Long Short-Term Memory (LSTM) neural network, the hidden layer output weights are dynamically weighted using an attention mechanism, and the calculation expression is as follows: (15) In the formula, a i This represents the weight allocation of the attention mechanism; h i Indicates the output of each hidden layer; 4) Finally, output the current wear amount; update the wear amount prediction value every 5 seconds; if the predicted wear amount exceeds the threshold, trigger the compensation command.

[0016] The compensation execution layer includes a compensation control module. The compensation control module generates compensation commands based on a pre-calibrated wear-braking force mapping table. It dynamically adjusts the linear motor stroke through a dual-closed-loop PID controller, while monitoring the actual braking force for closed-loop correction. The dual-closed-loop PID controller includes an inner-loop displacement control and an outer-loop braking force control. The inner loop adjusts the motor stroke according to the compensation commands, while the outer loop feeds back the actual braking force through a pressure sensor, achieving predictive-compensation-feedback coordinated control.

[0017] (III) Beneficial Effects Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Improved prediction accuracy: Compared with traditional thickness detection, the wear prediction error is reduced from ±15% to +5%, and the life prediction lead time is 200 hours.

[0018] (2) Dynamic compensation response: The range of braking force fluctuation is reduced to +3% (the traditional system is +10%).

[0019] (3) Reduced maintenance costs: Reduced number of unnecessary brake pad replacements and extended maintenance cycle by 30%. Attached Figure Description

[0020] Figure 1 This is a block diagram of the wear prediction and dynamic compensation method for tracked vehicle braking system based on artificial intelligence in this invention; Figure 2 This is the mechanical structure of the electromechanical compensation actuator in this invention; Figure 3 This is a flowchart of the control algorithm for the tracked vehicle braking system based on artificial intelligence in this invention. Detailed Implementation

[0021] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0022] To address the aforementioned technical problems, this invention provides a method for predicting and automatically compensating for wear in the braking system of tracked vehicles, the method comprising the following steps: Step 1: Collect real-time braking system operating data through a multi-sensor module, including brake pad temperature, pressure, displacement, vibration signals, and vehicle status parameters; Step 2: Preprocess the collected braking system operation data, including Kalman filtering for noise reduction, normalization, and multi-source data time-series alignment, to form aligned time-series data; Step 3: Input the preprocessed time series data into the wear prediction model based on deep learning and output the current wear amount; the wear state prediction model extracts local features by a one-dimensional convolutional neural network, captures temporal dependencies by a long short-term memory neural network, and dynamically weights key features through an attention mechanism; Step 4: If the predicted wear exceeds the preset threshold, a compensation command is generated based on the pre-calibrated wear-braking force mapping table. The linear motor stroke of the electromechanical actuator is adjusted through the dual closed-loop PID controller to dynamically adjust the brake pad position. At the same time, the actual braking force is fed back through the pressure sensor to achieve predictive-compensation-feedback coordinated control. Step 5: When the predicted wear amount is lower than the set threshold, switch to the physical model based on the friction coefficient-temperature equation for compensation control and trigger the online model update mechanism.

[0023] In step 3, the wear state prediction model adopts a dynamic weight allocation strategy, which adjusts the fusion weight of sensor data according to the vehicle operating condition classification results. Specifically, it includes: classifying operating condition categories based on vehicle speed, braking frequency and ambient temperature; and differentiating the weights of vibration signals and temperature signals through an attention mechanism for different operating conditions.

[0024] In step 4, the electromechanical actuator is integrated into the brake caliper and includes a linear motor, a ball screw, and a displacement sensor. The compensation range is ±2mm and the response time is ≤50ms.

[0025] The dual closed-loop PID controller in step 4 includes an inner-loop displacement control and an outer-loop braking force control. The inner-loop displacement control adjusts the motor stroke according to the compensation command, and the outer-loop braking force control feeds back the actual braking force through a pressure sensor.

[0026] Furthermore, the present invention also provides a wear prediction and automatic compensation system for the braking system of a tracked vehicle, the system comprising: Data acquisition layer: Collects brake pad temperature, pressure, displacement, vibration signals, and vehicle status information through a sensor array; Artificial Intelligence Processing Layer: After data acquisition and preprocessing, the artificial intelligence processing layer fuses vehicle braking temperature, pressure, vibration, displacement signals, and historical wear data to construct a deep learning-based wear state prediction model, achieving feature extraction and wear prediction. It also introduces a working condition classification module, including off-road / highway modes, to adaptively adjust model weights for different scenarios. Utilizing onboard edge computing devices and real-time incremental data training, the model is updated online to adapt to different driver habits and environmental changes. If the artificial intelligence prediction fails, it switches to a conservative compensation based on a physical model. If the predicted wear exceeds the threshold, a compensation command is triggered; Compensation execution layer: includes electromechanical compensation actuator and human-machine interface; a dual closed-loop control of wear prediction loop + braking force feedback loop is designed, the compensation displacement is calculated according to the wear amount-braking force mapping table, the linear motor stroke is adjusted by the dual closed-loop PID controller, and the actual braking force feedback correction is monitored.

[0027] The physical model is implemented using the friction coefficient-temperature equation.

[0028] In the data acquisition layer, Kalman filtering is used to denoise the vibration signal; Kalman filtering consists of two main steps: prediction and update. 1) Prediction Steps a. State prediction: (1) In the formula, It is a moment k State prediction, It is the system dynamic matrix. It is the optimal estimate of the state at the previous moment. It is a control input matrix. It is a control input; b. Covariance matrix prediction: (2) In the formula, It is the covariance matrix of the state prediction. It is the covariance matrix of the process noise; 2) Update steps a. Calculate the Kalman gain: (3) In the formula, It is Kalman gain. It is the observation matrix. It is the covariance matrix of the measured noise; b. Update state estimate: (4) In the formula, It is the optimal estimate of the updated state. It is the measurement value at the current moment; c. Update the covariance matrix: (5) In the formula, It is the covariance matrix of the updated state estimate. It is an identity matrix.

[0029] In the data acquisition layer, data standardization is performed; All Kalman filtered sensor data x After normalization, sensor data with all values ​​in the range [0, 1] are obtained: (6) In the formula, max ( x ) represents the maximum value of the original data, min( x ) represents the minimum value of the original data.

[0030] In the artificial intelligence processing layer, the wear state prediction model adopts a method combining a one-dimensional convolutional neural network and a long short-term memory neural network. 1) For the input time-series data, local features are first extracted by a one-dimensional convolutional neural network; the expression for the convolution operation is: (7) In the formula, Indicates the first l Layer output; Indicates the first l -1st floor i Output of each channel; express l -1st floor c One channel, and These represent the weights and biases, respectively. Pooling layers perform feature selection after convolutional layers, reducing the number of features; the average pooling calculation expression is: (8) In the formula, Indicates the first l Layer i The first channel t The value of each neuron; S Indicates the size of the pooled kernel; Indicates the first l Layer i The first channel j The value of each neuron; 2) Secondly, temporal dependencies are captured through a long short-term memory neural network; let... Represents the previous cell state. Represents the output of the previous cell. Represents the input to the current cell; where , , These are the results for the forget gate, input gate, and output gate, respectively; the forget gate consists solely of the sigmoid activation function σ, while the input and output gates consist of σ and the activation function tanh; the input... The forward propagation process in a long short-term memory neural network is as follows: a. Forgot Gate, update calculation formula is: (9) In the formula, Represents the weight matrix, Indicates the bias term; b. Input gate, update the calculation formula as follows: (10) (11) (12) In the formula, W i and W c Represents the weight matrix; b i and b c Indicates the bias term; Indicates will x t and h t-1 State variables that integrate state information; c Indicates the current state; c. Output gate, update calculation formula is: (13) (14) In the formula, W o Represents the weight matrix; b o Indicates the bias term; h t Indicates the output value of the hidden layer; 3) In a Long Short-Term Memory (LSTM) neural network, the hidden layer output weights are dynamically weighted using an attention mechanism, and the calculation expression is as follows: (15) In the formula, a i This represents the weight allocation of the attention mechanism; h i Indicates the output of each hidden layer; 4) Finally, output the current wear amount; update the wear amount prediction value every 5 seconds; if the predicted wear amount exceeds the threshold, trigger the compensation command.

[0031] The compensation execution layer includes a compensation control module. The compensation control module generates compensation commands based on a pre-calibrated wear-braking force mapping table. It dynamically adjusts the linear motor stroke through a dual-closed-loop PID controller, while monitoring the actual braking force for closed-loop correction. The dual-closed-loop PID controller includes an inner-loop displacement control and an outer-loop braking force control. The inner loop adjusts the motor stroke according to the compensation commands, while the outer loop feeds back the actual braking force through a pressure sensor, achieving predictive-compensation-feedback coordinated control.

[0032] Example 1 In this embodiment, the system specifically includes the following: 1. Data Acquisition Layer (1) Vehicle status input Vehicle speed (r / min), brake pedal travel (mm), temperature (°C); (2) Sensor group Brake pad temperature sensor (°C), pressure sensor (°C), displacement sensor (mm), vibration sensor (m / s) 2 )wait; (3) Filtering and noise reduction Kalman filtering is used for noise reduction of vibration signals; Kalman filtering consists of two main steps: prediction and update. 1) Prediction Steps a. State prediction: (1) In the formula, It is a moment k State prediction, It is the system dynamic matrix. It is the optimal estimate of the state at the previous moment. It is a control input matrix. It is a control input; b. Covariance matrix prediction: (2) In the formula, It is the covariance matrix of the state prediction. It is the covariance matrix of the process noise; 2) Update steps a. Calculate the Kalman gain: (3) In the formula, It is Kalman gain. It is the observation matrix. It is the covariance matrix of the measured noise; b. Update state estimate: (4) In the formula, It is the optimal estimate of the updated state. It is the measurement value at the current moment; c. Update the covariance matrix: (5) In the formula, It is the covariance matrix of the updated state estimate. It is the identity matrix; (4) Data standardization All Kalman filtered sensor data x After normalization, sensor data with all values ​​in the range [0, 1] are obtained: (6) In the formula, max ( x ) represents the maximum value of the original data, min( x () represents the minimum value of the original data; 2. Artificial Intelligence Processing Layer (1) Working condition classification module Operating conditions are categorized based on vehicle speed, braking frequency, and ambient temperature; for different operating conditions, the weights of vibration and temperature signals are differentiated through an attention mechanism. (2) Wear condition prediction model The wear condition prediction model uses a combination of a one-dimensional convolutional neural network and a long short-term memory neural network. 1) For the input time-series data, local features are first extracted by a one-dimensional convolutional neural network; the expression for the convolution operation is: (7) In the formula, Indicates the first l Layer output; Indicates the first l -1st floor i Output of each channel; express l -1st floor c One channel, and These represent the weights and biases, respectively. Pooling layers perform feature selection after convolutional layers, reducing the number of features; the average pooling calculation expression is: (8) In the formula, Indicates the first l Layer i The first channel t The value of each neuron; S Indicates the size of the pooled kernel; Indicates the first l Layer i The first channel j The value of each neuron; 2) Secondly, temporal dependencies are captured through a long short-term memory neural network; let... Represents the previous cell state. Represents the output of the previous cell. Represents the input to the current cell; where , , These are the results for the forget gate, input gate, and output gate, respectively; the forget gate consists solely of the sigmoid activation function σ, while the input and output gates consist of σ and the activation function tanh; the input... The forward propagation process in a long short-term memory neural network is as follows: a. Forgot Gate, update calculation formula is: (9) In the formula, Represents the weight matrix, Indicates the bias term; b. Input gate, update the calculation formula as follows: (10) (11) (12) In the formula, W i and W c Represents the weight matrix; b i and b c Indicates the bias term; Indicates will x t and h t-1 State variables that integrate state information; c Indicates the current state; c. Output gate, update calculation formula is: (13) (14) In the formula, W o Represents the weight matrix; b o Indicates the bias term; h t Indicates the output value of the hidden layer; 3) In a Long Short-Term Memory (LSTM) neural network, the hidden layer output weights are dynamically weighted using an attention mechanism, and the calculation expression is as follows: (15) In the formula, a i This represents the weight allocation of the attention mechanism; h i Indicates the output of each hidden layer; 4) Finally, output the current wear amount; update the wear amount prediction value every 5 seconds; if the predicted wear amount exceeds the threshold (such as 30% of the initial thickness), trigger the compensation command; (3) Redundancy prediction module When the prediction accuracy of the artificial intelligence model falls below a set threshold, the system switches to a physical model based on the friction coefficient-temperature equation for compensation control and triggers an online model update mechanism. 3. Compensation Execution Layer (1) Human-computer interaction module Real-time display of wear status and alarm information; prompts for replacement when wear exceeds a set threshold. (2) Compensation control module The compensation command is generated based on the pre-calibrated wear-braking force mapping table. The linear motor stroke is dynamically adjusted via a dual-closed-loop PID controller, while the actual braking force is monitored for closed-loop correction. (See appendix) Figure 2 The dual-loop PID controller includes an inner-loop displacement control and an outer-loop braking force control. The inner loop adjusts the motor stroke according to the compensation command, while the outer loop feeds back the actual braking force through a pressure sensor, thus achieving predictive-compensation-feedback coordinated control. (3) Electromechanical compensation actuation module The linear motor integrated into the brake caliper includes a linear motor, a ball screw, and a displacement sensor; it displaces the brake pads via the ball screw, with a compensation range of ±2mm and a response time ≤50ms. (See attached image) Figure 3 ; Calculate the required compensation displacement based on the pre-calibrated wear amount-braking force mapping table.

[0033] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting and automatically compensating for wear in the braking system of a tracked vehicle, characterized in that, The method includes the following steps: Step 1: Collect real-time braking system operating data through a multi-sensor module, including brake pad temperature, pressure, displacement, vibration signals, and vehicle status parameters; Step 2: Preprocess the collected braking system operation data, including Kalman filtering for noise reduction, normalization, and multi-source data time-series alignment, to form aligned time-series data; Step 3: Input the preprocessed time series data into the wear prediction model based on deep learning and output the current wear amount; the wear state prediction model extracts local features by a one-dimensional convolutional neural network, captures temporal dependencies by a long short-term memory neural network, and dynamically weights key features through an attention mechanism; Step 4: If the predicted wear exceeds the preset threshold, a compensation command is generated based on the pre-calibrated wear-braking force mapping table. The linear motor stroke of the electromechanical actuator is adjusted through the dual closed-loop PID controller to dynamically adjust the brake pad position. At the same time, the actual braking force is fed back through the pressure sensor to achieve predictive-compensation-feedback coordinated control. Step 5: When the predicted wear amount is lower than the set threshold, switch to the physical model based on the friction coefficient-temperature equation for compensation control and trigger the online model update mechanism.

2. The method for predicting and automatically compensating for wear in the braking system of a tracked vehicle as described in claim 1, characterized in that, The wear state prediction model in step 3 adopts a dynamic weight allocation strategy, which adjusts the fusion weight of sensor data according to the vehicle operating condition classification results. Specifically, it includes: classifying operating condition categories based on vehicle speed, braking frequency and ambient temperature; and differentiating the weights of vibration signals and temperature signals through an attention mechanism for different operating conditions.

3. The method for predicting and automatically compensating for wear in the braking system of a tracked vehicle as described in claim 1, characterized in that, The electromechanical actuator in step 4 is integrated into the brake caliper and includes a linear motor, a ball screw, and a displacement sensor. The compensation range is ±2mm and the response time is ≤50ms.

4. The method for predicting and automatically compensating for wear in the braking system of a tracked vehicle as described in claim 1, characterized in that, The dual closed-loop PID controller in step 4 includes an inner loop displacement control and an outer loop braking force control. The inner loop displacement control adjusts the motor stroke according to the compensation command, and the outer loop braking force control feeds back the actual braking force through a pressure sensor.

5. A wear prediction and automatic compensation system for the braking system of a tracked vehicle, characterized in that, The system includes: Data acquisition layer: Collects brake pad temperature, pressure, displacement, vibration signals, and vehicle status information through a sensor array; Artificial Intelligence Processing Layer: After data acquisition and preprocessing, the artificial intelligence processing layer fuses vehicle braking temperature, pressure, vibration, displacement signals, and historical wear data to construct a deep learning-based wear state prediction model, achieving feature extraction and wear prediction. It also introduces a working condition classification module, including off-road / highway modes, to adaptively adjust model weights for different scenarios. Utilizing onboard edge computing devices and real-time incremental data training, the model is updated online to adapt to different driver habits and environmental changes. If the artificial intelligence prediction fails, it switches to a conservative compensation based on a physical model. If the predicted wear exceeds the threshold, a compensation command is triggered; Compensation execution layer: includes electromechanical compensation actuator and human-machine interface; a dual closed-loop control of wear prediction loop + braking force feedback loop is designed, the compensation displacement is calculated according to the wear amount-braking force mapping table, the linear motor stroke is adjusted by the dual closed-loop PID controller, and the actual braking force feedback correction is monitored.

6. The tracked vehicle braking system wear prediction and automatic compensation system as described in claim 5, characterized in that, The physical model is implemented using the friction coefficient-temperature equation.

7. The tracked vehicle braking system wear prediction and automatic compensation system as described in claim 5, characterized in that, In the data acquisition layer, Kalman filtering is used to denoise the vibration signal; Kalman filtering consists of two main steps: prediction and update. 1) Prediction Steps a. State prediction: （1） In the formula, It is a moment k State prediction, It is the system dynamic matrix. It is the optimal estimate of the state at the previous moment. It is a control input matrix. It is a control input; b. Covariance matrix prediction: （2） In the formula, It is the covariance matrix of the state prediction. It is the covariance matrix of the process noise; 2) Update steps a. Calculate the Kalman gain: （3） In the formula, It is Kalman gain. It is the observation matrix. It is the covariance matrix of the measured noise; b. Update state estimates: （4） In the formula, It is the optimal estimate of the updated state. It is the measurement value at the current moment; c. Update the covariance matrix: （5） In the formula, It is the covariance matrix of the updated state estimate. It is an identity matrix.

8. The tracked vehicle braking system wear prediction and automatic compensation system as described in claim 7, characterized in that, In the data acquisition layer, data standardization is performed; All Kalman filtered sensor data x After normalization, sensor data with all values ​​in the range [0, 1] are obtained: （6） In the formula, max ( x ) represents the maximum value of the original data, min( x ) represents the minimum value of the original data.

9. The tracked vehicle braking system wear prediction and automatic compensation system as described in claim 8, characterized in that, In the artificial intelligence processing layer, the wear state prediction model adopts a method combining a one-dimensional convolutional neural network and a long short-term memory neural network. 1) For the input time-series data, local features are first extracted by a one-dimensional convolutional neural network; the expression for the convolution operation is: （7） In the formula, Indicates the first l Layer output; Indicates the first l -1st floor i Output of each channel; express l -1st floor c One channel, and These represent the weights and biases, respectively. Pooling layers perform feature selection after convolutional layers, reducing the number of features; average The pooling calculation expression is as follows （8） In the formula, Indicates the first l Layer i The first channel t The value of each neuron; S Indicates the size of the pooled kernel; Indicates the first l Layer i The first channel j The value of each neuron; 2) Secondly, temporal dependencies are captured through a long short-term memory neural network; let... Represents the previous cell state. Represents the output of the previous cell. Represents the input to the current cell; where , , These are the results for the forget gate, input gate, and output gate, respectively; the forget gate consists solely of the sigmoid activation function σ, while the input and output gates consist of σ and the activation function tanh; the input... The forward propagation process in a long short-term memory neural network is as follows: a. Forgot Gate, update calculation formula is: （9） In the formula, Represents the weight matrix, Indicates the bias term; b. Input gate, update the calculation formula as follows: （10） （11） （12） In the formula, W i and W c Represents the weight matrix; b i and b c Indicates the bias term; Indicates will x t and h t-1 State variables that integrate state information; c Indicates the current state; c. Output gate, update calculation formula is: （13） （14） In the formula, W o Represents the weight matrix; b o Indicates the bias term; h t Indicates the output value of the hidden layer; 3) In a Long Short-Term Memory (LSTM) neural network, the hidden layer output weights are dynamically weighted using an attention mechanism, and the calculation expression is as follows: （15） In the formula, a i This represents the weight allocation of the attention mechanism; h i Indicates the output of each hidden layer; 4) Finally, output the current wear amount; update the wear amount prediction value every 5 seconds; if the predicted wear amount exceeds the threshold, trigger the compensation command.

10. The tracked vehicle braking system wear prediction and automatic compensation system as described in claim 9, characterized in that, The compensation execution layer includes a compensation control module; The compensation control module generates compensation commands based on a pre-calibrated wear-braking force mapping table. It dynamically adjusts the linear motor stroke through a dual-closed-loop PID controller, while monitoring the actual braking force for closed-loop correction. The dual-closed-loop PID controller includes an inner-loop displacement control and an outer-loop braking force control. The inner loop adjusts the motor stroke according to the compensation commands, while the outer loop feeds back the actual braking force through a pressure sensor, achieving predictive-compensation-feedback coordinated control.

Images