Estimation method for brake clearance identification and self-adaptive clamping force of EMB system

Abstract

The invention belongs to the technical field of EMB braking, and relates to an EMB system brake clearance recognition and self-adaptive clamping force estimation method which comprises the steps that brake clearances are divided into a conventional brake clearance, a large brake clearance and a small brake clearance according to EMB contact time; according to the motor torque balance equation, a braking clamping force dynamic model is established by utilizing the electromagnetic torque, the angle and the angular speed of the motor; according to the braking clamping force dynamic model, respectively constructing an extended state observer of a fusion neural network and extended state observers A and B of angle compensation, and according to the braking clearance, respectively adopting the extended state observer of the fusion neural network and the extended state observer A and B of angle compensation to estimate the braking clamping force; according to the invention, the size of the brake clearance is identified by adopting a non-pressure sensor estimation algorithm, so that the brake clamping force is adaptively estimated under different brake clearances.

Description

Technical Field

[0001] The present invention belongs to the technical field of EMB braking systems, and particularly relates to a method for identifying the braking gap and estimating the adaptive clamping force of an EMB system. Background Art

[0002] Under the general trend of automotive electrification and intelligentization, EMB is driven by an electric motor, with an extremely fast response speed (80 - 100 milliseconds), which has a significant advantage compared to 150 milliseconds of EHB and higher safety. It does not require hydraulic oil, is environmentally friendly and easy to integrate new functions, is convenient for assembly, can simplify the system structure, and improve driving comfort.

[0003] Currently, in the integrated structure of EMB, the layout and application of pressure sensors face many challenges: First, the cost of miniature pressure sensors that meet the accuracy standards is relatively high, and their requirements for assembly accuracy are strict, which further increases the production cost; second, the integrated layout and calibration of pressure sensors are difficult. If it is placed at the front end of the lead screw, the high-temperature environment near the brake pad will interfere with the measurement accuracy; if the sensor is embedded inside the actuator, it may cause friction hysteresis problems; further, it will affect the braking effect of the vehicle and cannot ensure driving safety. At the same time, during the vehicle operation, the braking gap will change, and traditional estimation algorithms cannot ensure accurate estimation of the clamping force under different braking gaps.

[0004] Therefore, in order to avoid the problem of reduced braking force accuracy and affected braking safety caused by the arrangement of pressure sensors and the change of braking gaps; it is necessary to invent a clamping force estimation algorithm that can identify the contact time without pressure sensors and adapt to the braking gap. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method for identifying the braking gap and estimating the adaptive clamping force of an EMB system.

[0006] In order to achieve the purpose of the present invention, the following technical solutions are adopted for implementation.

[0007] A method for identifying the braking gap and estimating the adaptive clamping force of an EMB system includes the following steps:

[0008] S1. Divide the EMB braking gap into a normal braking gap state, a large braking gap state, and a small braking gap state according to the EMB contact time; where: the EMB contact time is determined according to the change of the motor angular velocity;

[0009] S2. According to the motor torque balance equation during the braking process of the EMB, use the electromagnetic torque, motor angle, and motor angular velocity of the motor to establish a dynamic model of the braking clamping force during the braking process of the EMB;

[0010] S3. Construct an extended state observer of the fusion neural network and an extended state observer of angle compensation respectively according to the braking clamping force dynamics model during the braking process of the EMB;

[0011] When the EMB braking gap is in the normal braking gap state, use the extended state observer of the fusion neural network to estimate the braking clamping force;

[0012] When the EMB braking gap is in the large braking gap state, use the extended state observer A of angle compensation to estimate the braking clamping force;

[0013] When the EMB braking gap is in the small braking gap state, use the extended state observer B of angle compensation to estimate the braking clamping force.

[0014] As a preferred solution of the present invention, the specific implementation process of the step S1 includes the following steps:

[0015] S101. Use an angular velocity sensor to collect the motor angular velocity ω m , and obtain the motor angular velocity change rate through calculation;

[0016] S102. Pre-calibrate the motor angular velocity change threshold Q when the EMB contacts. When the motor angular velocity change rate obtained in step S101 is less than or equal to Q, that is, when dω m / dt ≤ Q, collect the EMB contact time t0;

[0017] S103. According to the reasonable braking gap range set by the EMB, pre-calibrate the reasonable range (t min , t max ) of the contact time t0; if t0 ∈ [t min , t max , it is judged as a normal braking gap; if t0 > t max , it is judged that the braking gap is too large; if t0 < t min , it is judged that the braking gap is too small.

[0018] As a preferred solution of the present invention, the specific implementation process of the step S2 includes the following steps:

[0019] S201. Set the motor electromagnetic torque as:

[0020] In the formula, P n is the number of motor pole pairs, i d and i q are the stator-side magnetic flux and d-q coordinate system current respectively;

[0021] S202. Divide the EMB braking process into two processes: clamping and releasing;

[0022] During the clamping process, based on the torque balance equation of the motor, it can be obtained that:

[0023]

[0024] In the formula, ω m is the angular velocity of the motor, t is the time, J is the moment of inertia of the motor, T m is the electromagnetic torque of the motor, A is the dynamic deviation of the inherent parameters of the motor, μ is the Coulomb friction coefficient of the motor, and λ is the amplification coefficient;

[0025] During the release process, it can be obtained that:

[0026]

[0027] In the formula, λ + μ and λ - μ are measured with the motor torque as the abscissa through the EMB clamping force experiment;

[0028] S203. Based on the torque balance equations of the motor during the clamping process and the release process, a dynamic model of the braking clamping force during the clamping process and the release process is established, where:

[0029] The dynamic model of the braking clamping force during the clamping process is:

[0030]

[0031] In the formula, includes the sum of the clamping torque and the frictional torque;

[0032] The dynamic model of the braking clamping force during the release process is:

[0033]

[0034] In the formula,

[0035] As a preferred solution of the present invention, the extended state observer of the fusion neural network includes the extended state observers of the fusion neural network during the clamping process and the release process, where:

[0036] The expression of the extended state observer of the fusion neural network during the clamping process is:

[0037]

[0038] The expression of the extended state observer of the fusion neural network during the release process is:

[0039]

[0040] Among them, is θ mThe estimated value of is ω m The estimated value of is the estimated value of h. β1, β2, and β3 respectively represent gain parameters, T is the time step, and β i (k), i = 1, 2, 3 are the gain parameters output by the neural network during the clamping process. is the gain parameter output by the neural network during the release process. sgn(·) represents the signum function, and the Fal(e, α, δ) function is:

[0041]

[0042] As a preferred solution of the present invention, the process of obtaining the gain parameter: the number of nodes in the input layer of the neural network is n, which are x1, x2,..., x n ;

[0043] The inputs and outputs of the hidden layer are respectively:

[0044] O j = W ij x i ;

[0045]

[0046] In the formula, W ij is the neural network weight, O j and ψ(j) are the input and output of the j-th node in the hidden layer; n and l are the numbers of nodes in the input layer and the hidden layer respectively; a j and b j are the translation factor and the scaling factor respectively;

[0047] The output of the output layer is:

[0048]

[0049] e = y(f) - y d (f);

[0050] In the formula, y(f) and y d (f) are the actual output and the expected output of the f-th node in the output layer respectively; e is the error; W if is the neural network weight; m is the number of nodes in the output layer;

[0051] At the end of each step, the neural network outputs β1(k), β2(k), and β3(k).

[0052] As a preferred embodiment of the present invention, the extended state observer A for angle compensation includes the extended state observer A for angle compensation during the clamping process and the extended state observer A for angle compensation during the releasing process, where:

[0053] The expression of the extended state observer A for angle compensation during the clamping process is:

[0054]

[0055] The expression of the extended state observer A for angle compensation during the releasing process is:

[0056]

[0057] In the formula, is the motor angle corrected by the nonlinear angle compensation curve l1 of the large braking gap estimation model based on the contact time t and the Fal refi , i = 1, 2, …, 6 is the Fal function determined by the reference angle in the large braking gap estimation model, and β i , i = 1, 2,..., 6 are the gain parameters determined by the reference angle in the large braking gap estimation model.

[0058] As a preferred embodiment of the present invention, the establishment process of the extended state observer A for angle compensation includes the following steps:

[0059] S701. Establish estimation models with Δd as the wear interval respectively, and use θ max corresponding to max as the reference angle in the large braking gap estimation model;

[0060] S702. If the braking gap continues to increase, the motor angle θ m and the Fal function in the ESO algorithm should be corrected; through experiments, fix the Fal function and only correct the motor angle to achieve the same estimation accuracy;

[0061] S703. Fit the nonlinear compensation curve l1 of the large braking gap estimation model with the contact time t as the abscissa and the motor angle θ m as the ordinate, and unify the influence parameters generated by different wear amounts into the angle compensation to obtain the extended state observer A for angle compensation.

[0062] As a preferred embodiment of the present invention, the extended state observer B for angle compensation includes the extended state observer B for angle compensation during the clamping process and the extended state observer B for angle compensation during the releasing process, where:

[0063] The expression of the extended state observer B for angle compensation during the clamping process is:

[0064]

[0065] The expression of the extended state observer B for angle compensation during the release process is as follows:

[0066]

[0067] In the formula, is the motor angle corrected by the nonlinear angle compensation curve l2 of the small brake gap estimation model from the contact time t, Fal' refi , i = 1, 2,..., 6 is the Fal function determined by the reference angle in the small brake gap estimation model, β' i , i = 1, 2,..., 6 are the gain parameters determined by the reference angle in the large brake gap estimation model.

[0068] As a preferred solution of the present invention, the establishment process of the extended state observer B for angle compensation includes the following steps:

[0069] S901. Respectively establish estimation models at intervals of Δd, with t min corresponding θ min as the reference angle in the small brake gap estimation model;

[0070] S902. If the brake gap continues to decrease, correct the motor angle θ m and the Fal function in the ESO algorithm; through experiments, fix the Fal function and only correct the motor angle to achieve the same estimation accuracy;

[0071] S903. Fit the nonlinear compensation curve l2 of the small brake gap estimation model with the contact time t as the abscissa and the motor angle θ m as the ordinate, and unify the influence parameters generated by the too small brake gap into the angle compensation to obtain the extended state observer B for angle compensation.

[0072] As a preferred solution of the present invention, the process of updating the training set data of the neural network includes the following steps:

[0073] S1001. When the vehicle brakes each time, the neural network updates the weights in the network by the gradient descent method to obtain the gain parameters; only when parking, the neural network is updated as a whole by updating the training set data;

[0074] S1002. The method for collecting the training set data of the neural network is as follows: Detect the qualification of the clamping force corresponding to each step length during the braking process. The qualification detection standard is:

[0075] Under horizontal road conditions, when the brake pedal is depressed, record the vehicle speed v0 at this moment; when the brake pedal is released and the braking is considered to end, record the vehicle speed as vs During the braking process, the vehicle speed v is recorded every k moments k , until the braking ends; the true clamping force is calculated by the formula: Δk is the data acquisition interval time, where F x is the longitudinal resistance, m is the vehicle mass, z is the number of braking wheels, T b is the braking torque generated by a single wheel, r is the effective radius of the tire, r e is the radius of the brake disc, μ' is the friction coefficient between the brake pad and the brake disc, is the true braking force, f, F air The rolling resistance and air resistance of the vehicle at the current vehicle speed are measured by experiments; if In the formula, D is the specified standard, then the input parameters and output gain parameters corresponding to this step of braking are retained as the data set for neural network update.

[0076] Beneficial effects

[0077] In each braking process of the present invention, the braking gap of the EMB can be detected, and different estimation models are adopted according to the current braking gap, solving the problems that the pressure sensor has a high cost, a high failure rate, and the accuracy of different braking gaps cannot be guaranteed, thus affecting the vehicle braking;

[0078] The present invention first obtains the required braking clamping force F through the position sensor of the brake pedal cl ; then, by identifying the mutation signal of the motor angular velocity ω m , the contact time and braking gap are determined; secondly, considering the torque balance equation of the motor, using the motor electromagnetic torque T m and the motor (angle) θ m , a basic observer model is established, and at the same time, according to the state of the braking gap, the clamping force is estimated by using the conventional gap, large braking gap and small braking gap estimation models respectively Among them, for the conventional gap estimation model, an observer integrating a neural network is built, while the large braking gap estimation model and the small braking gap estimation model adopt non-linear angle compensation:

[0079] The present invention can accurately identify the contact time and braking gap, and respectively establish conventional gap, large braking gap, and small braking gap estimation models based on the extended state observer according to the braking gap, realizing the adaptive clamping force estimation under the current braking gap of the EMB, and solving the problems that may occur when using a pressure sensor and using a single estimation model. Description of the drawings

[0080] Figure 1 is the detailed flowchart of a method for braking gap identification and adaptive clamping force estimation of an EMB system based on a pressure sensorless according to the present invention;

[0081] Figure 2 It is the contact time identification diagram in the embodiment of the present invention;

[0082] Figure 3 It is the topological diagram after the optimization of the wavelet neural network in the embodiment of the present invention;

[0083] Figure 4 It is the ESO flow chart of the fusion wavelet neural network in the conventional braking gap estimation model in the embodiment of the present invention. Detailed implementation manners

[0084] The present invention will be further described in conjunction with the embodiments and the accompanying drawings.

[0085] As an embodiment of the present invention, as Figure 1 shown, a method for identifying the braking gap and estimating the adaptive clamping force of an EMB system based on a pressureless sensor includes the following steps:

[0086] S1: The specific process of identifying the EMB contact time through the motor angular velocity and classifying it into three categories: conventional braking gap, large braking gap, and small braking gap according to the contact time size is as follows:

[0087] S1: The specific process of identifying the EMB contact time through the motor angular velocity and classifying it into three categories: conventional braking gap, large braking gap, and small braking gap according to the contact time size is as follows:

[0088] (1) Collect the motor angular velocity ω through the angular velocity sensor m ;

[0089] (2) Preset the angular velocity change threshold Q when the EMB contacts. When the motor angular velocity change rate is less than or equal to Q, that is, when dω m / dt ≤ Q, collect the EMB contact time t0;

[0090] (3) According to the reasonable braking gap range set by the EMB, preset the reasonable range (t min , t max ) of the contact time t0; if t0 ∈ [t min , t max , it is judged as a conventional braking gap; if t0 > t max , it is judged that the braking gap is too large; if t0 < t min , it is judged that the braking gap is too small;

[0091] S2: The specific process of establishing the EMB dynamic model is as follows:

[0092] (1) Set the motor electromagnetic torque as

[0093] Where P n is the number of pole pairs of the motor, i d and i q are the flux linkage on the stator side and the d-q coordinate system current respectively;

[0094] (2) Divide the EMB braking into two processes: clamping and releasing:

[0095] For the clamping process, according to the torque balance equation of the motor, it can be obtained that:

[0096]

[0097] In the formula, ω m is the angular velocity, t is the time, J is the moment of inertia of the motor, T m is the electromagnetic torque of the motor, A is the dynamic deviation of the inherent parameters of the motor, μ is the Coulomb friction coefficient of the motor, and λ is the amplification coefficient;

[0098] For the releasing process, it can be obtained that:

[0099]

[0100] In the above formula, λ + μ and λ - μ can be measured by the EMB clamping force experiment with the motor torque as the abscissa.

[0101] (3) Build an extended state observer based on the torque balance equation to estimate the clamping force:

[0102] For the clamping process, the dynamic model of the braking clamping force is:

[0103]

[0104] In the formula, includes the sum of the clamping torque and the friction torque;

[0105] For the releasing process, the dynamic model of the braking clamping force is:

[0106]

[0107] In the formula,

[0108] So for the clamping process For the releasing process

[0109] S3: The specific process of estimating the braking clamping force by using the extended state observer with a fusion neural network for the conventional braking gap and the extended state observer with angle compensation for the large and small braking gaps is as follows:

[0110] (1) Based on the EMB dynamic model, an extended state observer for the braking clamping force is built. The observer takes the motor (angle) θ m , the electromagnetic torque T m as inputs, estimates h, and then obtains

[0111] the extended state observer of the braking clamping force during the clamping process:

[0112]

[0113] After discretization:

[0114]

[0115] Where is the estimated value of θ m , is the estimated value of ω m , is the estimated value of h. β1, β2, and β3 respectively represent gain parameters, T is the time step, sgn(·) represents the signum function, and the Fal(e, α, δ) function is:

[0116]

[0117] Similarly, the extended state observer of the braking clamping force during the release process:

[0118]

[0119] After discretization:

[0120]

[0121] Where

[0122] For the clamping process the release process

[0123] (2) If the actual contact t0 is within the specified range, that is, the EMB is in the conventional braking gap, based on the above observer expression, the extended state observer method integrating the updatable neural network is adopted to adapt to small-range deviations. In this embodiment, the wavelet neural network will be specifically used to achieve accurate estimation of the clamping force;

[0124] Among them, the conventional braking gap estimation model uses a neural network to obtain the gain parameters β1, β2, and β3 of the extended state observer: a wavelet neural network is adopted to optimize the tracking effect of the clamping force through training; the hidden layer of the wavelet neural network uses a wavelet function as the activation function, and the learning method is the gradient descent method. The topological map of the wavelet function in this embodiment is as shown in Figure 3 shown:

[0125] The number of nodes in the input layer is 4, which are x1 = e1, x2 = e2, x4 = 1, where e2 = de1 / dt, where F cl is the driver's required braking clamping force;

[0126] The hidden layer is 1 layer, and the number of nodes is l = 5;

[0127] The number of nodes in the output layer is m = 3, and the output is the parameters β1, β2, and β3 required in the ESO;

[0128] The wavelet function selects the Morlet mother wavelet basis function:

[0129]

[0130] Let the objective function of the wavelet neural network be:

[0131]

[0132] where

[0133] The gradient descent method is used to correct the weights W ij and W if , as well as the wavelet basis function coefficients a j , b j ;

[0134]

[0135] where: W ij (k + 1), W jf (k + 1), a j (k + 1), b j (k + 1) are the updated weights and coefficients of the wavelet neural network at the k-th moment; η is the learning rate; when correcting the weights and coefficients, the learning method of gradient descent is needed to deduce successively

[0136]

[0137] where: net j (k), ψ j(k) are the input and output of the j-th node in the hidden layer at the k-th moment; O f (k) is the output of the f-th node in the output layer; Since is unknown, it can be approximated and replaced Finally, we can get:

[0138]

[0139] At the end of each step, the neural network outputs β1(k), β2(k), β3(k);

[0140] The ESO flowchart of the fused wavelet neural network in this embodiment is as Figure 4 shown. The specific process is as follows: After the wavelet neural network is trained, initialize the network, and at this time, it is the moment k = 1; According to the network output, initially set a small learning rate, and then appropriately increase the learning rate when the error curve is relatively flat and the error is decreasing, and vice versa; Obtain the updated neural network weights; Obtain the gain parameters β1(k), β2(k), β3(k) and the system output When e(k + 1) < e(k), increase the learning rate in the way of hierarchical adjustment of the learning rate in this direction until the error no longer decreases, and then record the threshold and weights at this moment; On the contrary, decrease the learning rate until the error decreases, record the threshold and weights; Then enter the next moment.

[0141] The expression of the extended state observer after fusing the neural network, the clamping process is:

[0142]

[0143] The release process is:

[0144]

[0145] Among them, β i (k), i = 1, 2, 3 are the gain parameters output by the neural network during the clamping process, is the gain parameter output by the neural network during the release process;

[0146] For the clamping process The release process

[0147] (3) If the actual contact time t0 is not within the specified range, that is, the EMB is in an abnormal braking gap, based on the above expression of the extended state observer, the estimation model is corrected by using the method of nonlinear angle compensation. Among them, the braking clamping force estimation model established under the large braking gap state is the large braking gap estimation model, and the acquisition method of the nonlinear angle compensation curve in the large braking gap estimation model is as follows:

[0148] ①. Establish estimation models with Δd as the wear interval respectively, and use t max corresponding to θ max as the reference angle in the large braking gap estimation model;

[0149] ②. If the braking gap continues to increase, the motor (angle) θ m and the Fal function in the ESO algorithm should be corrected; through experiments, fix the Fal function and only correct the motor angle to achieve the same estimation accuracy;

[0150] ③. Fit the nonlinear compensation curve l1 of the large braking gap estimation model with the contact time t as the abscissa and the motor angle θ m as the ordinate, and unify the influence parameters generated by different wear amounts into the angle compensation;

[0151] Through angle compensation, adapt to the change of the braking gap and achieve accurate estimation of the clamping force; the observer expression corresponding to the large braking gap estimation model is as follows:

[0152]

[0153] The release process is:

[0154]

[0155] In the formula, is the motor angle corrected by the contact time t and the nonlinear angle compensation curve l1 of the large braking gap estimation model, Fal refi , i = 1, 2,..., 6 is the Fal function determined by the reference angle in the large braking gap estimation model, β i , i = 1, 2,..., 6 is the gain parameter determined by the reference angle in the large braking gap estimation model;

[0156] For the clamping process The release process

[0157] Among them, the braking clamping force estimation model established under the small braking gap state is the small braking gap estimation model, and the acquisition method of the nonlinear angle compensation curve in the small braking gap estimation model is as follows:

[0158] ①. Establish estimation models at intervals of Δd respectively, and use t min corresponding to θ min as the reference angle in the small braking gap estimation model;

[0159] ②. If the braking gap continues to decrease, the motor (angle) θ mAnd correct the Fal function in the ESO algorithm; through experiments, fix the Fal function and only correct the motor angle to achieve the same estimation accuracy;

[0160] ③. Fit the non-linear compensation curve l2 of the small braking gap estimation model with the contact time t as the abscissa and the motor angle θ m as the ordinate, and unify the influencing parameters caused by the too small braking gap into the angle compensation;

[0161] Through angle compensation, adapt to the change of the braking gap and achieve accurate estimation of the clamping force;

[0162] The clamping process is as follows:

[0163]

[0164] The release process is as follows:

[0165]

[0166] In the formula, is the motor angle corrected by the contact time t and the non-linear angle compensation curve l2 of the small braking gap estimation model, Fal' refi , i = 1, 2,..., 6 is the Fal function determined by the reference angle in the small braking gap estimation model, β' i , i = 1, 2,..., 6 is the gain parameter determined by the reference angle in the large braking gap estimation model;

[0167] For the clamping process For the release process

[0168] Therefore, for the case where the braking gap is within the specified range, an extended state observer integrating a neural network is adopted, and for the cases where the braking gap is too large or too small, an extended state observer with angle compensation is adopted to achieve accurate estimation of the clamping force under different braking gap conditions.

[0169] S4: Update the neural network training set data in the conventional gap estimation model according to the vehicle state.

[0170] (1). When the vehicle brakes each time, the wavelet neural network updates the weights in the network through the gradient descent method to obtain the gain parameter; only when parking, the neural network is updated as a whole by updating the training set data to ensure that the network adapts to the current vehicle state.

[0171] (2) The data acquisition method for the training set of the wavelet neural network is as follows: Detect the qualification of the clamping force corresponding to each step length during the braking process. The qualification detection standard is as follows: Under horizontal road conditions, when the brake pedal is depressed, record the vehicle speed v0 at this moment; when the brake pedal is released and the braking is considered ended, record the vehicle speed at this moment as v s ; Among them, at every interval of k moments, record the vehicle speed v k , until the braking ends; Calculate the true clamping force by the following formula:

[0172]

[0173] In the formula, k = 1, 2,..., s, Δk is the data acquisition interval time, where F x is the longitudinal resistance, m is the vehicle mass, z is the number of braking wheels, T b is the braking torque generated by a single wheel, r is the effective radius of the tire, r e is the brake disc radius, μ' is the friction coefficient between the brake pad and the brake disc, is the true braking force, f, F air are the rolling resistance and air resistance suffered by the vehicle at the current vehicle speed, which can be measured by experiments;

[0174] If in the formula D is the specified standard, then retain the input parameters and output gain parameters corresponding to the braking of this step length as the data set for updating the wavelet neural network;

[0175] The above has illustrated the preferred embodiments of the embodiments of the present application with reference to the accompanying drawings, and thus does not limit the scope of the rights of the embodiments of the present application. Any modifications, equivalent replacements, and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the rights of the embodiments of the present application.

Claims

1. A method for identifying the braking gap and estimating the adaptive clamping force of an EMB system, characterized in that: It includes the following steps: S1. Divide the EMB braking gap into a normal braking gap state, a large braking gap state, and a small braking gap state according to the EMB contact time; where: the EMB contact time is determined according to the change of the motor angular velocity; S2. According to the motor torque balance equation during the EMB braking process, use the electromagnetic torque, motor angle, and motor angular velocity of the motor to establish a dynamic model of the braking clamping force during the EMB braking process; S3. Respectively construct an extended state observer of the fusion neural network and an extended state observer of angle compensation according to the dynamic model of the braking clamping force during the EMB braking process; When the EMB braking gap is in the normal braking gap state, use the extended state observer of the fusion neural network to estimate the braking clamping force; When the EMB braking gap is in the large braking gap state, use the extended state observer A of angle compensation to estimate the braking clamping force; When the EMB braking gap is in the small braking gap state, use the extended state observer B of angle compensation to estimate the braking clamping force.

2. The method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 1, wherein: The specific implementation process of step S1 includes the following steps: S101. Collect the angular velocity ω of the motor using an angular velocity sensor m , and obtain the angular velocity change rate of the motor through calculation; S102. Pre-calibrate the threshold value of the motor angular velocity change during EMB contact as Q. When the motor angular velocity change rate obtained in step S101 is less than or equal to Q, that is, when dω m / dt ≤ Q, collect the EMB contact time t0; S103. According to the reasonable braking clearance range set by the EMB, pre-calibrate the reasonable range of the contact time t0 (t min , t max ); if t0 ∈ [t min , t max , it is judged as a normal braking clearance; if t0 > t max , it is judged that the braking clearance is too large; if t0 < t min , it is judged that the braking clearance is too small.

3. The method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 1, wherein: The specific implementation process of step S2 includes the following steps: S201. Set the motor electromagnetic torque to: Wherein, P n is the number of pole pairs of the motor, i d and i q are respectively the magnetic flux on the stator side and the d-q coordinate system current; S202. Divide the EMB braking process into two processes: clamping and releasing: During the clamping process, according to the torque balance equation of the motor, it can be obtained that: where ω m is the angular velocity of the motor, t is the time, J is the moment of inertia of the motor, T m is the electromagnetic torque of the motor, A is the dynamic offset of the inherent parameters of the motor, μ is the Coulomb friction coefficient of the motor, and λ is the amplification coefficient; During the releasing process, it can be obtained that: In the formula, λ + μ and λ - μ are measured with the motor torque as the abscissa through the EMB clamping force experiment; S203. Based on the motor torque balance equations of the clamping process and the releasing process, build a dynamic model of the braking clamping force for the clamping process and the releasing process, where: The dynamic model of the braking clamping force for the clamping process is: In the formula, includes the sum of the clamping moment and the frictional moment; The dynamic model of the braking clamping force for the releasing process is: wherein, 4. The method for identifying braking clearance and estimating adaptive clamping force of an EMB system according to claim 1, characterized in that: The extended state observer of the fusion neural network includes the extended state observer of the fusion neural network for the clamping process and the releasing process, where: The expression of the extended state observer of the fusion neural network for the clamping process is: The expression of the extended state observer of the fusion neural network for the releasing process is: Among them, is the estimated value of θ m , is the estimated value of ω m , is the estimated value of h, β1, β2, and β3 respectively represent gain parameters, T is the time step, and β i (k), i = 1, 2, 3 are the gain parameters output by the neural network during the clamping process. is the gain parameter output by the neural network during the release process, sgn(·) represents the signum function, and the Fal(e, α, δ) function is:

5. A method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 4, characterized in that: The process of obtaining the gain parameter: The number of nodes in the input layer of the neural network is n, which are x1, x2,..., x n ; The input and output of the hidden layer are respectively: O j = W ij x i ; Where, W ij is the neural network weight, O j , ψ(j) are the input and output of the j-th node in the hidden layer; n and l are the nodes of the input layer and the hidden layer respectively; a j , b j are the translation factor and the scaling factor respectively; The output of the output layer is: e = y(f) - y d (f); where y(f) and y d (f) are the actual output and the expected output of the f-th node in the output layer respectively; e is the error; W if is the neural network weight; m is the number of nodes in the output layer; At the end of each step, the neural network outputs β1(k), β2(k), β3(k).

6. The method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 1, wherein: The extended state observer A of angle compensation includes the extended state observer A of angle compensation for the clamping process and the extended state observer A of angle compensation for the releasing process, where: The expression of the extended state observer A of angle compensation for the clamping process is: The expression of the extended state observer A of angle compensation for the releasing process is: In the formula, is the motor angle corrected by the nonlinear angle compensation curve l1 of the contact time t and the large braking gap estimation model, Fal refi , i = 1, 2,..., 6 are the Fal functions determined by the reference angle in the large braking gap estimation model, β i , i = 1, 2,..., 6 are the gain parameters determined by the reference angle in the large braking gap estimation model.

7. The method for identifying braking gap and estimating adaptive clamping force of an EMB system according to claim 6, wherein: The establishment process of the extended state observer A of angle compensation includes the following steps: S701. Establish an estimation model with a wear interval of Δd respectively, and use θ max corresponding to t max as the reference angle in the large braking gap estimation model; S702. If the braking gap continues to increase, the motor angle θ m and the Fal function in the ESO algorithm should be corrected; through experiments, by fixing the Fal function and only correcting the motor angle, the same estimation accuracy can be achieved; S703. Fit out the non-linear compensation curve l1 of the large braking gap estimation model with the contact time t as the abscissa and the motor angle θ m as the ordinate, unify the influence parameters generated by different wear amounts into the angle compensation, so as to obtain the extended state observer A for angle compensation.

8. The method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 1, characterized in that: The extended state observer B of angle compensation includes the extended state observer B of angle compensation for the clamping process and the extended state observer B of angle compensation for the releasing process, where: The expression of the extended state observer B of angle compensation for the clamping process is: The expression of the extended state observer B of angle compensation for the releasing process is: In the formula, is the motor angle corrected by the nonlinear angle compensation curve l2 of the contact time t and the small braking gap estimation model, Fal' refi , i = 1, 2,..., 6 is the Fal function determined by the reference angle in the small braking gap estimation model, β' i , i = 1, 2,..., 6 are the gain parameters determined by the reference angle in the large braking gap estimation model.

9. The method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 8, characterized in that: The establishment process of the extended state observer B of angle compensation includes the following steps: S901. Respectively establish an estimation model at intervals of Δd, with t min corresponding θ min as the reference angle in the small brake clearance estimation model; S902. If the braking gap continues to decrease, correct the motor angle θ m and the Fal function in the ESO algorithm; through experiments, fix the Fal function and only correct the motor angle to achieve the same estimation accuracy; S903. Fit out a non-linear compensation curve l2 of the small braking gap estimation model with the contact time t as the abscissa and the motor angle θ m as the ordinate, and unify the influence parameters generated by the too small braking gap into the angle compensation to obtain an extended state observer B for angle compensation.

10. The method for identifying the braking gap and estimating the adaptive clamping force of an EMB system according to claim 1, characterized in that: The process of updating the training set data of the neural network includes the following steps: S1001. When the vehicle brakes each time, the neural network updates the weights in the network through the gradient descent method to obtain gain parameters; only when parking, the neural network is globally updated by updating the training set data. S1002. The training set data of the neural network is collected as follows: Detect the qualification of the clamping force corresponding to each step length during the braking process. The qualification detection standard is: Under horizontal road conditions, when the brake pedal is depressed, record the vehicle speed v0 at this moment; when the brake pedal is released, it is considered that braking ends, and record the vehicle speed at this moment as v s ; During braking, record the vehicle speed v k every k moments until braking ends; Calculate the true clamping force by the formula: Δk is the data acquisition interval time, where F x is the longitudinal resistance, m is the vehicle mass, z is the number of braking wheels, T b is the braking torque generated by a single wheel, r is the effective radius of the tire, r e is the brake disc radius, μ' is the friction coefficient between the brake pad and the brake disc, is the true braking force, f, F air are the rolling resistance and air resistance of the vehicle at the current vehicle speed, which are measured by experiments; If In the formula, D is the specified standard, then retain the input parameters and output gain parameters corresponding to this step of braking as the dataset for neural network update.

Images