Self-adaptive correction method for clamping force of motor electric control brake system based on deep learning

Abstract

The invention discloses a motor electric control brake system clamping force self-adaptive correction method based on deep learning, and belongs to the technical field of chassis control, the method fully considers the wear condition of a brake pad in the use process, introduces a deep learning neural network optimization control strategy for the nonlinear change problem caused by the wear of the brake pad, and corrects the clamping force of the motor electric control brake system. The complex nonlinear relation between the clamping force estimated by the rotation angle of the motor and the actual clamping force is modeled as predictable probability distribution, and self-adaptive correction is performed, so that the clamping force control is optimized under the brake pad abrasion condition, the dynamic change caused by the brake pad abrasion is effectively dealt with, the real-time performance of system control is enhanced, and the system reliability is improved. The prediction precision and the response speed of the system are greatly improved, the stability, the reliability and the safety of the brake system are comprehensively enhanced, and meanwhile, the situation that a large number of sensors are adopted for real-time clamping force monitoring is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of chassis control, and in particular relates to a method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning. Background Art

[0002] Electromechanical brakes (EMBs) use electrical signals to achieve fast and precise control of braking force. With their high efficiency, adjustability and precision, they are gradually replacing traditional hydraulic brake systems and becoming an important system and key component in future automotive chassis. Compared with traditional mechanical or hydraulic brake systems, electromechanical brakes can more accurately adjust the brake force distribution, which not only improves the braking response speed, but also greatly improves the stability and safety of the brake system. The main function of the electromechanical brake (EMB) system is to control the target clamping force. The clamping force control is essentially the feedback control of the motor. Common feedback control algorithms include proportional-integral-derivative (PID) control algorithm, linear unconstrained model predictive control (MPC) algorithm, and sliding mode variable structure control (SMC) algorithm. The EMB system consists of a motor, a torque increase and reduction device, a motion conversion device, and a brake assembly. These systems usually use pressure sensors to monitor the brake pressure in real time to ensure precise control of the braking process. However, these systems often require a large number of sensors and often have problems such as short life, high equipment cost and high maintenance cost in high temperature environments, which limits their application in harsh conditions.

[0003] Currently, an approach widely studied by industry and academia is to assume that the key physical components of the brake system, especially the thickness and texture of the brake pad, remain consistent and unchanged throughout the entire service life. This assumption simplifies the control model of the brake force, because the physical properties of the brake pad directly affect the efficiency and accuracy of the brake force transmission. However, in actual applications, the brake pad will undergo physical wear during continuous use, especially the gradual reduction in thickness. The wear of the brake pad not only changes its contact characteristics, but also indirectly increases the overall physical gap in the brake system. The increase in gap leads to a key problem: the clamping force estimation based on the motor rotation angle begins to deviate, because the motor must rotate a larger angle to push the brake pad further toward the brake disc to compensate for the increased gap. In this case, if the motor angle remains unchanged, the actual clamping force applied will decrease due to the increase in physical gap, resulting in the actual clamping force being lower than the clamping force estimated from the motor angle, thus affecting the accuracy of clamping force control.

[0004] In general, although there has been some progress in the research and application of motor-controlled brake systems, the existing brake force control technology still has some shortcomings:

[0005] 1) The performance of existing sensors is limited in high temperature and harsh environments, resulting in shortened life or failure, which affects the precise control of braking force;

[0006] 2) Traditional control methods usually assume that the physical properties of the brake pad remain unchanged, but the brake pad will wear during use, resulting in changes in contact characteristics and physical gaps, which in turn makes the relationship between the clamping force estimated by the motor angle and the actual clamping force complex and nonlinear;

[0007] 3) Traditional linear control models have difficulty coping with the impact of brake pad wear and complex dynamic changes, and cannot accurately predict the clamping force, thus affecting the response speed and stability of the brake system;

[0008] 4) Existing systems usually require a large number of sensors, which not only increases production costs, but also brings difficulties in maintenance and calibration, limiting their widespread application;

[0009] 5) The existing brake system has a long response time and may not be able to generate sufficient braking force in time in emergency braking or quick response scenarios, affecting the braking effect and safety;

[0010] Therefore, how to improve the system's robustness, accuracy and response speed and overcome these challenges has become the key to the development of motor-controlled brake systems. Summary of the invention

[0011] In order to solve the shortcomings of the prior art, effectively deal with the dynamic changes caused by brake pad wear, significantly improve the prediction accuracy and response speed of the brake system, and comprehensively enhance the stability, reliability and safety of the brake system, the present invention adopts the following technical solutions:

[0012] The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning includes the following steps:

[0013] Step 1: Construct a mathematical model of the brake system;

[0014] Based on the brake signal, the motor rotates and outputs torque, converting the rotational motion into the linear motion of the brake pad, so that the brake pad clamps the brake disc to brake; the displacement of the linear motion is obtained according to the rotation angle of the motor, and the deformation of the caliper of the brake pad clamping the brake disc is calculated through the initial position and the displacement, and a nonlinear relationship between the deformation and the clamping force is constructed to establish the relationship between the motor rotation angle and the clamping force;

[0015] Step 2: Build a neural network model;

[0016] The clamping force is predicted by a mathematical model, the target clamping force is generated by a neural network model, and the actual clamping force is corrected by the target clamping force.

[0017] Furthermore, in step 1, the model is built by a planetary gear reducer and a ball screw, and the relationship between the motor rotation angle and the displacement of the ball screw nut is:

[0018]

[0019] Wherein, P represents the pitch, θ represents the motor rotation angle (unit: radian), and i represents the transmission ratio of the reducer.

[0020] Further, the deformation is obtained according to the difference between the nut displacement s and the initial brake clearance s0;

[0021] Δs=s-s0

[0022] Build up clamping force F n The nonlinear relationship between the deformation Δs:

[0023] F n =K f1 Δs 3 +K f2 Δs 2 +K f3 Δs+K f4

[0024] Among them, K f1 , K f2 , K f3 , K f4 A coefficient that indicates the clamping force.

[0025] Furthermore, the motor rotates and outputs torque, and the torque multiplier converts the low torque output by the motor into high torque to establish a relationship between the clamping force and the motor rotation torque;

[0026]

[0027] Among them, T L Indicates the motor rotation torque, F n represents the clamping force, L represents the lead of the ball screw, η s Indicates the transmission efficiency of the ball screw, i p Indicates the transmission ratio of the planetary gear reducer, η p Indicates the efficiency of the transmission system.

[0028] Furthermore, the loss function of the neural network model in step 2 adopts mean square error (MSE), and the calculation formula is as follows:

[0029]

[0030] Among them, y represents the true value, m represents the number of samples, Represents the predicted value.

[0031] Furthermore, the neural network model of step 2 adopts a multilayer perceptron and uses a ReLU activation function to introduce nonlinear characteristics, thereby enhancing the learning ability of the model. The ReLU function is defined as follows:

[0032] σ(z)=max(0,z)

[0033] Where z represents the neuron input of the multilayer perceptron, σ(z) represents the ReLU activation function, which is used to set all negative values ​​to zero while keeping all non-negative values ​​unchanged; the output of each layer is calculated as follows:

[0034] z (i) =W (i) a (i) +b (i)

[0035] a (i+1) =σ(z (i) )

[0036] Among them, W (i) and b (i) Represent the weight and bias of the previous layer, a (i) represents the output of the previous layer, a (i+1) Represents the output of the current layer, which is used as the input of the next layer, z (i) Represents the neuron input of the previous layer.

[0037] Furthermore, the neural network model trains a multi-layer perceptron, adjusts the weight W and the bias b to minimize the loss function, and uses stochastic gradient descent (SGD) as an optimization algorithm. The update rule of stochastic gradient descent (SGD) is as follows:

[0038]

[0039] Among them, θ represents the parameters of the neural network model, including weights and biases, η represents the learning rate, represents the gradient of the loss function L with respect to the parameter θ.

[0040] The motor-controlled brake system based on deep learning includes physical hardware components and a feedback control system. The feedback control system includes a clamping force loop. The clamping force loop adopts the clamping force adaptive correction method of the motor-controlled brake system based on deep learning to correct the clamping force in real time to improve the accuracy of clamping force control.

[0041] Furthermore, the feedback control system also includes a speed loop and a current loop. The speed loop obtains speed data through an encoder installed on the motor for speed feedback control; the current loop obtains current data through the encoder for current regulation.

[0042] Furthermore, the physical hardware components include a permanent magnet synchronous motor, a planetary gear set, a ball screw and a caliper brake disc; the permanent magnet synchronous motor is used to provide a power source; the planetary gear set is used to increase the output torque; the ball screw converts rotational motion into linear motion, thereby completing the braking process through the caliper brake disc.

[0043] The advantages and beneficial effects of the present invention are:

[0044] The present invention uses a neural network model to replace the traditional clamping force sensor, which reduces the dependence of the brake system on hardware sensors and has stronger adaptability. At the same time, the loss of clamping force caused by brake pad wear after long-term use is taken into account in the modeling process; the present invention proposes an adaptive clamping force correction strategy, which utilizes a multi-layer perceptron network, takes the clamping force estimated by the motor rotation angle as input, and the actual clamping force as a training target, predicts the probability distribution of the clamping force, and estimates the actual clamping force by calculating its expected value. By compensating for the clamping force estimation error, the clamping force estimation deviation caused by brake pad wear can be effectively compensated, significantly improving the control performance of the brake system and ensuring the reliability and accuracy of the system during long-term operation; the present invention provides a neural network-based feedback control method, which successfully captures the complex nonlinear relationship between the motor rotation angle and the actual clamping force. The network model predicts the probability distribution of the clamping force and calculates the expected value, thereby realizing real-time correction of the clamping force and ensuring that the brake system can quickly respond to the target value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of a method in an embodiment of the present invention.

[0046] Figure 2 1 is a flow chart of the braking process in an embodiment of the present invention.

[0047] Figure 3 Schematic diagram of the structure of the brake system in the embodiment of the present invention.

[0048] Figure 4 It is a diagram of the clamping force estimated by the motor and the actual clamping force when the target braking force is 8000N in the embodiment of the present invention.

[0049] Figure 5 It is a diagram of the clamping force estimated by the motor and the actual clamping force when the target braking force is 6000N in an embodiment of the present invention.

[0050] Figure 6 It is a diagram of the clamping force estimated by the motor and the actual clamping force when the target braking force is 4000N in an embodiment of the present invention.

[0051] Figure 7It is a comparison diagram between the estimated clamping force corrected by the neural network and the actual clamping force when the target clamping force is 8000N in an embodiment of the present invention.

[0052] Figure 8 It is a comparison diagram between the estimated clamping force corrected by the neural network and the actual clamping force when the target clamping force is 6000N in an embodiment of the present invention.

[0053] Fig. 9 It is a comparison diagram between the estimated clamping force corrected by the neural network and the actual clamping force when the target clamping force is 4000N in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0055] like Figure 1 As shown, the present invention's method for adaptively correcting the clamping force of an electric motor and electronically controlled brake system based on deep learning first mathematically models the electronic mechanical brake system model, and then proposes an innovative adaptive clamping force correction strategy based on a neural network to deeply optimize the performance of the electric motor and electronically controlled brake system. Through a carefully constructed multi-layer perceptron network model, combined with multiple hidden layer structures, the clamping force estimated by the motor rotation angle is used as input, and the actual measured clamping force is used as a training label, thereby achieving efficient modeling and accurate prediction of the complex nonlinear relationship between the clamping force and the motor angle. The model generates a probability distribution in the output layer using the Softmax function, thereby effectively capturing the potential change trend of the clamping force under different input conditions. By accurately calculating the expected value of the probability distribution, the method can accurately predict the actual clamping force in a complex dynamic environment, surpassing the limitations of traditional control methods. This innovative method based on probability prediction not only greatly improves the accuracy of the prediction, but also enhances the reliability and accuracy of the brake force control. Specifically includes the following steps:

[0056] Step 1: Electromechanical brake selection and mathematical modeling;

[0057] like Figure 2 As shown in the figure, after receiving the brake signal, the electromechanical brake (EMB) system will apply brake pressure. The motor starts to rotate and output torque; the torque multiplier converts the low torque output by the motor into a sufficiently high torque; the motion converter converts the rotational motion of the motor into the linear motion of the brake pad, which clamps the brake disc to achieve braking.

[0058] The planetary gear reducer and ball screw are modeled. During braking, the deformation of the brake pad when clamping the brake disc generates a clamping force. The deformation of the caliper Δs is defined as the difference between the nut displacement s and the initial brake gap s0:

[0059] Δs=s-s0 (1)

[0060] The unit of deformation is millimeter (mm). There is a cubic nonlinear relationship between the clamping force and deformation:

[0061] F n =K f1 Δs 3 +K f2 Δs 2 +K f3 Δs+K f4 (2)

[0062] Among them, K f1 , K f2 , K f3 , K f4 The coefficient representing the clamping force. The load torque on the EMB motor can be calculated by the clamping force:

[0063]

[0064] Among them, the transmission ratio of the planetary gear reducer is i p , the transmission efficiency of the ball screw is η s , L is the lead of the ball screw (unit: mm), and the efficiency of the transmission system is η p .

[0065] The relationship between the motor rotation angle and the nut displacement is:

[0066]

[0067] Where P is the pitch, θ is the motor rotation angle (unit: radians), and i is the transmission ratio of the reducer.

[0068] According to formulas (1), (2) and (4), the current clamping force can be estimated. By fitting the curve with the test bench data, the characteristic curve between the caliper deformation and the clamping force is obtained. Furthermore, using formula (3), a function describing the relationship between the motor rotation angle and the clamping force is established (the angle unit is radians). Based on the existing test bench data and literature modeling, the clamping force F N There is a nonlinear relationship between the displacement Δs:

[0069]

[0070] Step 2: Neural network model construction;

[0071] The caliper deformation variable Δs is the difference between the nut displacement s and the brake clearance s0. With continued use, the brake clearance s0 will gradually increase, causing Δs to decrease. This difference causes a deviation between the estimated clamping force and the actual clamping force. In addition, the relationship between the estimated clamping force and the actual clamping force of the initial calibration is complex and nonlinear.

[0072] The present invention adopts a multi-layer perceptron to solve the regression problem, including network architecture selection, activation function application, loss function determination and optimization algorithm selection.

[0073] Input layer: Receives one input feature, which is the preprocessed predicted clamping force value.

[0074] Hidden layer: It consists of three layers of multilayer perceptrons. The first layer contains 64 neurons, the second layer also contains 64 neurons, and the third layer contains 128 neurons. The ReLU activation function σ(z) is used to introduce nonlinear characteristics to enhance the learning ability of the model. The ReLU function is defined as follows:

[0075] σ(z)=max(0,z)(6)

[0076] Where z is the input to the neuron. The ReLU function sets all negative values ​​to zero while keeping all non-negative values ​​unchanged. The output of each layer is calculated as follows:

[0077] z (i) =W (i) a (i) +b (i) (7)

[0078] a (i+1) =σ(z (i) ) (8)

[0079] Among them, W (i) and b (i) are the weight and bias of the previous layer, a (i) is the output of the previous layer, a (i+1) is the output of the current layer and serves as the input of the next layer.

[0080] Output layer: The output layer consists of 2000 neurons, which use the softmax function to represent the distribution of the target clamping force within the specified range. Then, the expected value of this probability distribution is calculated to obtain the corrected actual clamping force.

[0081] Loss function: In order to train the network, the mean square error (MSE) is used as the loss function. The calculation formula is as follows:

[0082]

[0083] Among them, y represents the true value, m is the number of samples, is the predicted value.

[0084] Training Optimization: Training an MLP involves adjusting the weights W and biases b to minimize the loss function. Stochastic Gradient Descent (SGD) is used as the optimization algorithm. The update rule of SGD is as follows:

[0085]

[0086] Among them, θ represents the parameters of the model, including weights and biases, η is the learning rate, is the gradient of the loss function with respect to the parameters.

[0087] like Figure 3 As shown, the electromechanical brake system and its control architecture include physical hardware components and feedback control systems.

[0088] The physical hardware components mainly include a permanent magnet synchronous motor that provides the main power, a planetary gear set that increases the output torque, a ball screw that converts rotational motion into linear motion, and a caliper brake disc that effectively performs the braking action. This configuration achieves efficient power transmission and precise control of braking force. The permanent magnet synchronous motor is the main power source, the planetary gear set is used to increase the output torque, and the ball screw converts the motor's rotational motion into the linear motion of the brake pad, thereby completing the braking process through the caliper brake disc.

[0089] The feedback control system adopts a three-loop feedback control structure, including a clamping force loop, a speed loop, and a current loop.

[0090] The clamping force loop accurately calculates the motor rotation angle and uses a neural network built based on deep learning to build an adaptive clamping force correction method for the motor electronically controlled brake system to make real-time corrections to improve the accuracy of clamping force control.

[0091] The speed loop obtains speed data through the encoder installed on the motor for speed feedback control;

[0092] The current loop obtains current data through the encoder for current regulation.

[0093] In terms of control strategy, the clamping force loop and speed loop mainly adopt PID feedback control technology; 2) The current loop adopts the model predictive control (MPC) algorithm to improve the performance verification effect.

[0094] Model training and performance evaluation: In model training, the cross-validation method is used to evaluate the accuracy and generalization ability of the model. The evaluation indicators include: 1) mean square error (MSE); 2) root mean square error (RMSE); 3) mean absolute error (MAE); 4) coefficient of determination (R2 );

[0095] System performance evaluation. The main evaluation indicators of system performance include: 1) Steady-state error: measures the deviation between the output value and the target setting value after the system reaches a stable state, reflecting the accuracy and reliability of the system in long-term operation; 2) Transient response: describes the process of the system from the initial state to the stable state, including the response speed and dynamic characteristics during the transition process.

[0096] This architecture combines efficient hardware design with intelligent feedback control strategies to not only improve the control accuracy of the braking system, but also improve its operating stability and reliability in complex environments.

[0097] Example:

[0098] The purpose of this experiment is to verify the effectiveness of the adaptive clamping force correction strategy based on a multi-layer perceptron (MLP), especially its performance under simulated brake wear conditions. In the experiment, in order to build and test the electro-mechanical brake system and its control strategy, we chose to use the SIMULINK 2024a platform to create a comprehensive model of the brake system. The model includes components such as motors, sensors, and control algorithms, which can intuitively demonstrate the real-time response and adjustment process of the system. In addition, in order to implement the adaptive clamping force correction strategy based on neural networks, we developed and trained a model on the PyTorch platform. After the training phase is completed, the neural network model is exported to the ONNX (Open Neural Network Exchange) format and imported into the SIMULINK environment. This integration helps to realize the real-time application and verification of the model in the experimental setting.

[0099] Data Collection: Brake Actuator Wear Simulation. We simulated the wear of the brake actuator and collected data on the clamping force estimated by the motor shaft angle and the actual clamping force. Assume that the initial brake clearance is 0.7mm and the brake pad wear increases to 0.75mm.

[0100] Network training: The collected data is used to train the neural network. The collected data is used to train the neural network. The input of the network is the motor shaft angle, and the label is the actual measured clamping force. The network architecture consists of an input layer, several hidden layers, and an output layer. The output layer uses a softmax function to predict the probability distribution of the clamping force and determines the clamping force by calculating the expected value of this distribution.

[0101] Experimental results:

[0102] The influence of brake pad wear on clamping force control: The influence of brake clearance increase. Assuming that the initial brake clearance is 0.7mm, the clearance increases to 0.75mm after the brake pad wears. When the target braking force is set to 4000N, 6000N and 8000N respectively, the clamping force estimated by the motor rotation angle and the actual clamping force are as follows Figures 4 to 6 The red line indicates the clamping force estimated by the motor, and the yellow line indicates the actual clamping force. Although the clamping force estimated by the motor rotation angle can reach the target value, the actual clamping force has a lag and is lower than the estimated value.

[0103] Model training and regression performance evaluation. The model performance is shown in Table 1.

[0104] Table 1 Model fitting results table

[0105]

[0106] The model is further integrated into the brake control system to adapt and accurately predict the wear state of the brake pads. The predicted clamping force and the actual clamping force are as follows Figures 7 to 9 As shown in the figure, it can be clearly seen that the model has successfully learned the relationship between the clamping force estimated by the motor rotation angle and the actual clamping force, and has shown good prediction accuracy. After adjustment, the clamping force curve estimated by the motor rotation angle is highly consistent with the actual clamping force curve, and can quickly and stably approach the desired target value.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based adaptive correction method for the clamping force of an electric motor and electronically controlled brake system, characterized in that The steps include: Step 1: Construct a mathematical model of the brake system; Based on the brake signal, the motor rotates and outputs torque, converting the rotational motion into the linear motion of the brake pad, so that the brake pad clamps the brake disc to brake; The displacement of the linear motion is obtained according to the rotation angle of the motor, the deformation of the caliper of the brake pad clamping the brake disc is calculated through the initial position and the displacement, and a nonlinear relationship between the deformation and the clamping force is constructed to establish the relationship between the rotation angle of the motor and the clamping force; Step 2: Build a neural network model; The clamping force is predicted by a mathematical model, the target clamping force is generated by a neural network model, and the actual clamping force is corrected by the target clamping force.

2. The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning according to claim 1 is characterized in that: In step 1, the model is built by the planetary gear reducer and the ball screw, and the relationship between the motor rotation angle and the displacement of the ball screw nut is: Among them, P represents the pitch, θ represents the motor rotation angle, and i represents the transmission ratio of the reducer.

3. The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning according to claim 2 is characterized in that: The deformation is obtained according to the difference between the nut displacement s and the initial brake clearance s0; Δs=s-s0 Build-up clamping force F n The nonlinear relationship between the deformation Δs: F n =K f1 Δs 3 +K f2 Δs 2 +K f3 Δs+K f4 Among them, K f1 , K f2 , K f3 , K f4 A coefficient that indicates the clamping force.

4. The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning according to claim 3 is characterized in that: The motor rotates and outputs torque, and then the torque multiplier converts the low torque output by the motor into high torque to establish a relationship between the clamping force and the motor rotation torque; Among them, T L Indicates the motor rotation torque, F n represents the clamping force, L represents the lead of the ball screw, η s Indicates the transmission efficiency of the ball screw, i p Indicates the transmission ratio of the planetary gear reducer, η p Indicates the efficiency of the transmission system.

5. The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning according to claim 1 is characterized in that: The loss function of the neural network model in step 2 adopts mean square error, and the calculation formula is as follows: Among them, y represents the true value, m represents the number of samples, Represents the predicted value.

6. The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning according to claim 1 is characterized in that: The neural network model in step 2 adopts a multilayer perceptron and uses an activation function to introduce nonlinear characteristics. The activation function is defined as follows: σ(z)=max(0,z) Where z represents the neuron input of the multilayer perceptron, σ(z) represents the activation function, which is used to set all negative values ​​to zero and keep all non-negative values ​​unchanged; the output of each layer is calculated as follows: z (i) =W (i) a (i) +b (i) a (i+1) =σ(z (i) ) Among them, W (i) and b (i) Represent the weight and bias of the previous layer, a (i) represents the output of the previous layer, a (i+1) Represents the output of the current layer, which is used as the input of the next layer, z (i) Represents the neuron input of the previous layer.

7. The method for adaptively correcting the clamping force of a motor-controlled brake system based on deep learning according to claim 1 is characterized in that: The neural network model trains a multi-layer perceptron, adjusts weights and biases to minimize the loss function, and uses stochastic gradient descent as the optimization algorithm. The update rule of stochastic gradient descent is as follows: Among them, θ represents the parameters of the neural network model, including weights and biases, η represents the learning rate, represents the gradient of the loss function L with respect to the parameter θ.

8. A motor-controlled brake system based on deep learning, including physical hardware components and a feedback control system, characterized in that: The feedback control system includes a clamping force loop, which adopts the deep learning-based motor electronically controlled brake system clamping force adaptive correction method described in claim 1 to correct the clamping force in real time.

9. The motor-controlled brake system based on deep learning according to claim 8, characterized in that: The feedback control system also includes a speed loop and a current loop. The speed loop obtains speed data through an encoder installed on the motor for speed feedback control; the current loop obtains current data through the encoder for current regulation.

10. The motor-controlled brake system based on deep learning according to claim 8, characterized in that: The physical hardware components include a permanent magnet synchronous motor, a planetary gear set, a ball screw and a caliper brake disc; the permanent magnet synchronous motor is used to provide a power source; the planetary gear set is used to increase the output torque; the ball screw converts rotational motion into linear motion, thereby completing the braking process through the caliper brake disc.

Images