Tractor clutch test bench cooperative control method based on MPC-BP neural network PID

Abstract

The invention relates to a tractor clutch test bench cooperative control method based on MPC-BP neural network PID, and belongs to the technical field of tractor clutch performance testing. The method specifically comprises the steps of establishing an enhanced mathematical model of a tractor clutch test bed system; an MPC prediction controller is designed, and an optimal control sequence is generated through rolling optimization of an objective function; a BP neural network PID controller is improved; a multi-unit cooperative control strategy is designed, MPC prediction and BP neural network PID control are combined, and cooperative adjustment of the torque of the loading unit, the rotating speed of the driving unit and the displacement of the clutch separation mechanism unit is achieved; a self-adaptive adjustment and load compensation mechanism is introduced, and the adaptability of the system to different working conditions and clutch models is improved through online model correction, heat fading modeling and fault diagnosis. The control precision, the dynamic response speed and the robustness of the test bed are remarkably improved, and the test bed can be widely applied to research, development and performance testing of the tractor clutch.

Description

Technical Field

[0001] The present invention relates to the technical field of tractor clutch test bench control, and particularly to a cooperative control method for a tractor clutch test bench based on MPC-BP neural network PID. Background Technique

[0002] In the process of research and development and production of tractors, the performance test of the clutch is crucial. There are many deficiencies in the control of existing tractor clutch test benches. Traditional control methods are difficult to achieve precise cooperative control of the loading unit, driving unit, and clutch separation mechanism unit of the test bench, resulting in poor accuracy and reliability of test results. During the loading process, it is impossible to accurately adjust the loading torque according to the real-time state of the clutch, resulting in a large deviation between the test conditions and the actual working conditions; the rotational speed control accuracy of the driving unit is not high, affecting the test effect of the dynamic performance of the clutch; the action response of the clutch separation mechanism unit is slow and cannot meet the requirements of fast and precise control. In addition, the existing control methods lack the adaptive ability to complex working conditions and are difficult to meet the diverse test requirements of different types of clutches.

[0003] In the prior art, although a single BP neural network PID controller has self-adaptability, it lacks the ability to predict the future state of the system, resulting in limited control effects under rapidly changing dynamic conditions. With the continuous development of tractor technology, higher requirements are put forward for the control accuracy and adaptability of the power shift clutch test bench. Therefore, it is urgent to develop an innovative control method to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a cooperative control method for a tractor clutch test bench based on MPC-BP neural network PID to solve the problems of low control accuracy, poor cooperativity, and insufficient adaptive ability of existing tractor clutch test benches, realize precise cooperative control of each unit of the tractor clutch test bench, improve the accuracy and reliability of test results, meet the diverse test requirements of different types of clutches, and have a strong adaptive load compensation function. By introducing an MPC predictive controller, the present invention can predict the state changes in the future time domain, optimize the control instructions in advance, and significantly improve the control robustness under complex working conditions.

[0005] The present invention is specifically realized through the following technical solutions. A cooperative control method for a tractor clutch test bench based on MPC-BP neural network PID proposed according to the present invention includes the following steps:

[0006] Step (1), establish an enhanced mathematical model of the tractor clutch test bench system, including the discretized state space models of the loading unit, driving unit, and clutch separation mechanism unit;

[0007] Step (2): Design an MPC predictive controller based on the enhanced mathematical model in step (1), generate an optimal control sequence through rolling optimization of the objective function, and impose system constraint conditions;

[0008] Step (3): Improve the BP neural network PID controller, use the future system state output by the MPC predictive controller, the real-time feedback error, and the error change rate as inputs, and dynamically adjust the control parameters of the PID controller;

[0009] Step (4): Design a multi-unit cooperative control strategy, combine MPC prediction and BP neural network PID control to achieve cooperative adjustment of the loading unit torque, the driving unit speed, and the displacement of the clutch release mechanism unit;

[0010] Step (5): Introduce an adaptive adjustment mechanism, through thermal fade modeling and fault diagnosis, to improve the adaptability of the system to different working conditions and clutch models.

[0011] Furthermore, in step (1), perform mathematical modeling on the loading unit, driving unit, and clutch release mechanism unit of the tractor clutch test bench, and analyze the input-output relationship and dynamic characteristics of each unit.

[0012] Among them, the loading unit establishes a relationship model between the loading torque T and the excitation current I based on the principle of an eddy current retarder: c The relationship model:

[0013]

[0014] In Equation (1), B is the magnetic induction intensity, ρ is the resistivity of the rotor disk, μ0 is the magnetic permeability of vacuum, N is the number of turns of the excitation winding, I c is the excitation current, I g is the air gap, k e is the conversion coefficient, Δ h is the skin depth of the eddy current, S p is the eddy current area, ω c is the angular velocity of the magnetic field change. Among them, N, I c , ρ, Δ h , S p can all obtain data from the nameplate of the eddy current retarder. The value of μ0 is generally 4π×10 -7 H / m, the value of I g is generally negligible, and the value of k e is generally 2.

[0015] In Equation (2), N p is the number of pole pairs and can obtain data from the nameplate of the eddy current retarder; n is the rotational speed and is obtained through a rotational speed sensor.

[0016]

[0017] In Equation (3), T is the loading torque, N is the number of turns of the exciting winding, g is the acceleration due to gravity, B is the magnetic induction intensity, r is the radius of the eddy current ring, d is the diameter of the yoke, ω is the rotational angular velocity of the rotor disc, and Δ h is the skin depth of the eddy current, and ρ is the resistivity of the rotor disc. Among them, N, ω, r, d, Δ h , and ρ can all obtain data from the nameplate of the eddy current retarder.

[0018] The drive unit constructs a dynamic relationship model between the motor speed n and the input signal f of the frequency converter based on the characteristics of the variable-frequency speed-regulating motor;

[0019]

[0020] In Equation (4), s is the slip ratio, and p is the number of pole pairs of the motor (data obtained from the motor nameplate); in Equation (5), n s is the synchronous speed (data obtained through the speed sensor).

[0021] The clutch release mechanism unit establishes a dynamic response model between the cylinder displacement d and the control pressure P based on the working principle of the hydraulic system;

[0022] In Equation (6), m is the equivalent mass, c is the damping coefficient, k is the spring stiffness, and A is the effective area of the piston, all of which can obtain data from the nameplates of the cylinder and the piston.

[0023] Based on the mathematical models of the loading unit, the drive unit, and the clutch release mechanism unit, through the integration of the mathematical models of each unit, a discretized state-space model (the overall system model of the test bench) suitable for MPC is constructed, providing a basis for the subsequent control algorithm design. The discretized state-space model includes the system state equation and the output equation:

[0024]

[0025] In Equation (7), x(k) is the system state vector, including state variables such as the torque of the loading unit, the speed of the drive unit, and the displacement of the clutch release mechanism unit; u(k) is the control input vector, including control quantities such as the input signal of the frequency converter and the control pressure; y(k) is the output vector, which can be determined according to the actual measurement requirements; A, B, and C are state matrices, and their elements are determined according to the mathematical models of each unit.

[0026] Furthermore, in step (2), the objective function of the MPC predictive controller is:

[0027]

[0028] In Equation (8), y ref is the reference trajectory, which is set according to test requirements, such as the desired loading torque, rotational speed, displacement, etc.; y(k+i) represents the system vector in the next i steps; Q and R are weight matrices used to adjust the weights of the system output tracking the reference trajectory and the change weights of the control input, and can be adjusted according to actual control requirements.

[0029] Solve the optimal control sequence U by rolling optimization of the objective function * =[u(k), u(k + 1),....., u(k + N C -1)], and take the first term u(k) as the current control quantity. During the optimization process, consider the constraints of the system, such as the torque limit of the loading unit, the rotational speed limit of the driving unit, the displacement limit of the clutch disengagement mechanism unit, etc., to ensure the feasibility of the prediction results.

[0030] Furthermore, in step (3), in view of the unique property that multiple variables of the clutch test bench are closely coupled, a dedicated algorithm controller - BP neural network PID controller is specially customized. The BP neural network has strong self-learning and adaptive capabilities and can automatically adjust the parameters of the PID controller according to the real-time data during the test process.

[0031] Take the prediction values of the system output at multiple future moments output by the MPC predictive controller, the errors between the set values of the rotational speed, torque, and displacement of each unit, the change rate of the rotational speed error, and the friction plate temperature as the inputs of the BP neural network. Based on these inputs, the BP neural network outputs the proportional coefficient, integral coefficient, and differential coefficient of the PID controller after network learning and training. The two complement each other, enabling the entire control system to have both the flexible characteristics of intelligent learning and the reliable quality of classical control, and achieving a high degree of adaptation to the complex and variable working conditions during clutch testing.

[0032] Among them, the input layer of the BP neural network includes 5 nodes: rotational speed error, change rate of rotational speed error, torque error, displacement error, and friction plate temperature. The hidden layer uses the Sigmoid activation function, and the output layer generates the PID parameters K p 、K i 、K d and updates the weights online through the backpropagation algorithm. The specific training process is as follows:

[0033] The 5 nodes in the input layer of the BP neural network include:

[0034] Rotational speed error e n =n 设定 -n 预测

[0035] Change rate of rotational speed error Δe n =e n(t)-e n (t - 1)

[0036] Torque error e T = T 设定 - T 预测

[0037] Displacement error e d = d 设定 - d 预测

[0038] Friction plate temperature T (0 - 400 °C)

[0039] The hidden layer uses 15 nodes, and the activation function is the Sigmoid function:

[0040] The calculation of the hidden layer output is:

[0041] The output layer generates PID parameters K p 、K i 、K d , and the calculation formula is:

[0042]

[0043] In equations (11) to (13), h j is the output of the hidden layer, ω and b are the weights and bias terms.

[0044] The calculation and training process of the BP neural network includes:

[0045] (1) The objective function uses the mean square error:

[0046]

[0047] In equation (14), y k is the expected output; is the actual output; m is the number of samples.

[0048] (2) The weight update uses the backpropagation algorithm:

[0049] (2.1) Weight correction of the output layer:

[0050]

[0051] (2.2) Weight correction of the hidden layer

[0052]

[0053] In equations (15) and (16), the learning rate η = 0.01; the value range is usually in [0, 1] and can be adjusted according to the training effect.

[0054] Further, in step (4), a cooperative control strategy for the test bench based on MPC-BP neural network PID is formulated. During the test, parameters such as the torque of the loading unit, the rotational speed of the driving unit, and the displacement of the clutch release mechanism unit are collected in real time, and these parameters are used as feedback signals and input into the MPC predictive controller and the BP neural network PID controller. The MPC predictive controller predicts the future system output based on the current system state, and the BP neural network PID controller calculates the control signals of each unit according to the input feedback signals and the MPC predictive output, realizing the cooperative control of the loading torque of the loading unit, the rotational speed of the driving unit, and the action of the clutch release mechanism unit.

[0055] The cooperative control strategy includes: the driving unit eliminates the rotational speed error by controlling the frequency of the frequency converter through PID; the loading unit dynamically compensates the load inertia and generates a compensation current to adjust the loading torque; the clutch release mechanism unit accurately controls the cylinder pressure according to the displacement error to ensure the stability of the engagement speed.

[0056] The cooperative control strategy is executed hierarchically:

[0057] (1) The driving unit eliminates the rotational speed error by controlling the frequency of the frequency converter through PID, and the rotational speed error Δn ≤ ±1%;

[0058] (2) The loading unit dynamically compensates the load inertia J and generates a compensation current I 补偿 to adjust the loading torque:

[0059]

[0060] The compensation current is: I 补偿 = K p ·ΔJ + K i ·∫ΔJdt (18)

[0061] In formula (17), J is the total load inertia, J0 is the initial load inertia, given according to the actual situation of the test bench, m i is the weight of the i-th additional mass block, r i is the equivalent radius of the i-th additional mass block, and N is the number of additional mass blocks;

[0062] In formula (18), I 补偿 is the compensation current; K p and K i are the proportional coefficient and the integral coefficient respectively, both from the BP neural network PID controller; ΔJ is the load inertia error.

[0063] (3) The clutch release mechanism unit accurately controls the cylinder pressure according to the displacement error to ensure the stability of the engagement speed; the displacement control accuracy Δd ≤ ±0.1 mm.

[0064] When the clutch is in the engagement process, according to the rotational speed of the drive unit and the torque transmission characteristics of the clutch, the loading torque of the loading unit is automatically adjusted by the MPC-BP neural network PID controller to simulate the load change in actual operation during the test process. At the same time, according to the displacement feedback of the clutch separation mechanism unit, the engagement speed of the clutch is accurately controlled to ensure the stability and accuracy of the test process.

[0065] Furthermore, in step (5), the adaptive adjustment mechanism includes: incorporating the MPC prediction error into the BP neural network training data to enhance the model's self-learning ability; real-time correcting the friction coefficient model based on the change in the friction plate temperature to suppress the torque fluctuation caused by thermal fade; in the case of sensor failure or torque mutation, switching to the current estimation mode and triggering the safety protection mechanism.

[0066] To improve the adaptability of the test bench to different types of clutches and complex test conditions, an adaptive adjustment mechanism is introduced into the control method. The MPC prediction error is incorporated into the BP neural network training data to enhance the model's self-learning ability. Through the real-time analysis of test data, the BP neural network can automatically identify the characteristics of the current test conditions and adjust its own structure and parameters according to the preset rules and experience to better adapt to different test requirements.

[0067] Meanwhile, based on the BP neural network, a model of the thermal fade characteristics of the clutch friction plate is established to predict the temperature change trend, and the control parameters are adjusted in advance to suppress the torque fluctuation in the heat load test. When testing clutches with different friction coefficients, the BP neural network can automatically adjust the parameters of the PID controller according to the characteristic parameters of the clutch to achieve the best control effect. In addition, the control strategy can be adjusted in a timely manner according to abnormal situations during the test, such as sensor failure and equipment overload, to ensure the safe progress of the test.

[0068] The detection process of the adaptive adjustment mechanism includes:

[0069] (1) Thermal fade compensation: The friction coefficient model is:

[0070]

[0071] In Equation (19), μ0 is the friction coefficient at normal temperature, λ is the attenuation coefficient, T is the real-time temperature, and T0 is the reference temperature.

[0072] (2) Fault diagnosis:

[0073] When the temperature of the friction plate exceeds the threshold (T > 350 °C), an alarm is triggered and the speed is automatically reduced to the safe speed, n safe = 0.5n 设定 .

[0074] When a torque mutation is detected (|ΔT>10%T 设定 |), switch to the current estimation mode and isolate the abnormal signal. The current estimation mode: T 估算 =K t ·I 电机 , where K t is the motor torque constant, usually obtained from the technical manual of the motor factory.

[0075] Compared with the prior art, the present invention has obvious advantages and beneficial effects. By means of the above technical solution, the present invention can achieve quite high technical progressiveness and practicability, and has wide utilization value. It has at least the following advantages:

[0076] (1) Improve control accuracy: Through the automatic optimization of the PID controller parameters by the BP neural network and the prediction of the future system state by the MPC predictive controller, it is possible to achieve precise control of the loading unit, driving unit, and clutch separation mechanism unit of the tractor clutch test bench, effectively improving the control accuracy.

[0077] (2) Enhance coordination: Based on the coordinated control strategy of MPC-BP neural network PID, it is possible to more comprehensively adjust the torque of the loading unit, the speed of the driving unit, and the displacement of the clutch separation mechanism unit according to the real-time state and future predicted state of each unit during the test, further improving the coordination between the units of the clutch test bench. This coordinated control method makes the test process closer to the actual working conditions, effectively improving the reliability and effectiveness of the test.

[0078] (3) Improve adaptability: Introduce an adaptive adjustment mechanism and combine the adaptability of the MPC predictive controller to different working conditions, so that the test bench can automatically adapt to the diverse test requirements of different types of clutches and the changes in complex test working conditions. Whether it is a clutch with different friction coefficients, different sizes and specifications, or in tests simulating various complex working environments, the coordinated control method of the present invention can ensure the stable operation of the test bench and obtain accurate test results.

[0079] (4) Enhance predictive control ability and robustness: The MPC predictive controller can optimize the control command in advance, significantly reducing the overshoot under dynamic working conditions, improving the response speed, and importing the predicted output into the BP-PID algorithm for real-time adjustment. The system has stronger adaptability to model errors and external disturbances. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 is the architecture diagram of the coordinated adaptive control system.

[0081] Figure 2 is the flowchart of the self-tuning of BP neural network PID parameters.

[0082] Figure 3 It is a flow chart of a multi - unit collaborative control strategy.

[0083] Figure 4 It is a schematic diagram of the MPC predictive control process and optimization mechanism.

[0084] Figure 5 It is a flow chart of the MPC online model correction and adaptive update. Specific implementation manners

[0085] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will, in conjunction with specific embodiments, clearly and completely describe the technical solutions of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0086] A collaborative control method for a tractor clutch test bench based on MPC - BP neural network PID includes the following steps:

[0087] Step (1): Establish an enhanced mathematical model of the tractor clutch test bench system, including the discretized state - space models of the loading unit, the driving unit, and the clutch separation mechanism unit;

[0088] Step (2): Design an MPC predictive controller based on the enhanced mathematical model in step (1), generate an optimal control sequence through rolling optimization of the objective function, and impose system constraint conditions;

[0089] Step (3): Improve the BP neural network PID controller, with the future system state output by the MPC predictive controller, the real - time feedback error, and the error change rate as inputs, and dynamically adjust the control parameters of the PID controller;

[0090] Step (4): Design a multi - unit collaborative control strategy, combine MPC prediction and BP neural network PID control to achieve coordinated adjustment of the torque of the loading unit, the speed of the driving unit, and the displacement of the clutch separation mechanism unit;

[0091] Step (5): Introduce an adaptive adjustment mechanism, through thermal fade modeling and fault diagnosis, to improve the adaptability of the system to different working conditions and clutch models.

[0092] Further, in step (1), mathematical modeling is carried out on the loading unit, the driving unit, and the clutch separation mechanism unit of the tractor clutch test bench, and the input - output relationships and dynamic characteristics of each unit are analyzed.

[0093] Among them, the loading unit establishes a relationship model between the loading torque T and the exciting current I based on the principle of an eddy current retarder: c

[0094]

[0095] In Equation (1), B is the magnetic induction intensity, ρ is the resistivity of the rotor disk, μ0 is the magnetic permeability of vacuum, N is the number of turns of the exciting winding, I c is the exciting current, I g is the air gap, k e is the conversion coefficient, Δ h is the skin depth of the eddy current, S p is the eddy current area, ω c is the angular velocity of the magnetic field change. Among them, N, I c , ρ, Δ h , S p can all obtain data from the nameplate of the eddy current retarder. The value of μ0 is generally 4π×10 -7 H / m, the value of I g is generally negligible, and the value of k e is generally 2.

[0096] In Equation (2), N p is the number of pole pairs and can obtain data from the nameplate of the eddy current retarder; n is the rotational speed and is obtained through a rotational speed sensor.

[0097]

[0098] In Equation (3), T is the loading torque, N is the number of turns of the exciting winding, g is the acceleration due to gravity, B is the magnetic induction intensity, r is the radius of the eddy current ring, d is the diameter of the yoke, ω is the rotational angular velocity of the rotor disk, Δ h is the skin depth of the eddy current, and ρ is the resistivity of the rotor disk. Among them, N, ω, r, d, Δ h , ρ can all obtain data from the nameplate of the eddy current retarder.

[0099] The drive unit constructs a dynamic relationship model between the motor rotational speed n and the frequency converter input signal f based on the characteristics of a variable-frequency speed-regulating motor;

[0100]

[0101] In Equation (4), s is the slip ratio, p is the number of pole pairs of the motor (obtain data from the motor nameplate); in Equation (5), n s is the synchronous rotational speed (obtain data through a rotational speed sensor).

[0102] The clutch disengagement mechanism unit establishes a dynamic response model between the cylinder displacement d and the control pressure P based on the working principle of a hydraulic system;​

[0103] In Equation (6), m is the equivalent mass, c is the damping coefficient, k is the spring stiffness, and A is the effective area of the piston. The data of all these parameters can be obtained from the nameplates of the oil cylinder and the piston.

[0104] Based on the mathematical models of the loading unit, the driving unit, and the clutch disengagement mechanism unit, a discretized state - space model (the overall test bench system model) suitable for MPC is constructed by integrating the mathematical models of each unit, providing a basis for the subsequent design of the control algorithm. The discretized state - space model includes the system state equation and the output equation:

[0105]

[0106] In Equation (7), x(k) is the system state vector, which includes state variables such as the torque of the loading unit, the rotational speed of the driving unit, and the displacement of the clutch disengagement mechanism unit; u(k) is the control input vector, which includes control variables such as the input signal of the frequency converter and the control pressure; y(k) is the output vector, which can be determined according to the actual measurement requirements; A, B, and C are state matrices, and their elements are determined according to the mathematical models of each unit.

[0107] Furthermore, in step (2), the objective function of the MPC predictive controller is:

[0108]

[0109] In Equation (8), y ref is the reference trajectory, which is set according to the test requirements, such as the desired loading torque, rotational speed, displacement, etc.; y(k + i) represents the system vector in the future i steps; Q and R are weight matrices, which are used to adjust the weight of the system output tracking the reference trajectory and the change weight of the control input, and can be adjusted according to the actual control requirements.

[0110] The optimal control sequence U * =[u(k), u(k + 1),....., u(k + N C -1)] is obtained by rolling - optimizing the objective function, and the first term u(k) is taken as the current control quantity. During the optimization process, the constraint conditions of the system are considered, such as the torque limit of the loading unit, the rotational speed limit of the driving unit, the displacement limit of the clutch disengagement mechanism unit, etc., to ensure the feasibility of the prediction results.

[0111] Furthermore, in step (3), in view of the unique property that multiple variables of the clutch test bench are closely coupled, a dedicated algorithm controller - the BP neural network PID controller is specifically tailored. The BP neural network has strong self - learning and adaptive capabilities and can automatically adjust the parameters of the PID controller according to the real - time data during the test process.

[0112] The predicted values of the system output at multiple future moments output by the MPC predictive controller, the errors between the rotational speed, torque, and displacement set values of each unit, the change rate of the rotational speed error, and the friction plate temperature are used as the inputs of the BP neural network. Based on these inputs, the BP neural network outputs the proportional coefficient, integral coefficient, and differential coefficient of the PID controller through network learning and training. The two complement each other, enabling the entire control system to have both the flexible characteristics of intelligent learning and the reliable quality of classical control, achieving a high degree of adaptation to the complex and variable working conditions during clutch testing.

[0113] Among them, the input layer of the BP neural network includes 5 nodes: rotational speed error, change rate of rotational speed error, torque error, displacement error, and friction plate temperature. The hidden layer uses the Sigmoid activation function, and the output layer generates the PID parameters K p 、K i 、K d And the weights are updated online through the backpropagation algorithm. The specific training process is as follows:

[0114] The 5 nodes in the input layer of the BP neural network include:

[0115] Rotational speed error e n =n 设定 -n 预测

[0116] Change rate of rotational speed error Δe n =e n (t)-e n (t - 1)

[0117] Torque error e T =T 设定 -T 预测

[0118] Displacement error e d =d 设定 -d 预测

[0119] Friction plate temperature T (0 - 400 °C)

[0120] The hidden layer uses 15 nodes, and the activation function is the Sigmoid function:

[0121] The calculation of the output of the hidden layer is:

[0122] The output layer generates the PID parameters K p 、K i 、K d The calculation formula is:

[0123]

[0124] In Equations (11) to (13), h j is the output of the hidden layer, and ω and b are the weights and bias terms.

[0125] The calculation and training process of the BP neural network includes:

[0126] (1) The objective function uses the mean square error:

[0127]

[0128] In Equation (14), y k is the expected output; is the actual output; m is the number of samples.

[0129] (2) The weight update uses the backpropagation algorithm:

[0130] (2.1) Output layer weight correction:

[0131]

[0132] (2.2) Hidden layer weight correction

[0133]

[0134] In Equations (15) and (16), the learning rate η = 0.01; the value range is usually in [0, 1] and can be adjusted according to the training effect.

[0135] Furthermore, in step (4), a collaborative control strategy based on the MPC - BP neural network PID is formulated. During the experiment, parameters such as the torque of the loading unit, the rotational speed of the driving unit, and the displacement of the clutch separation mechanism unit are collected in real time, and these parameters are used as feedback signals and input into the MPC predictive controller and the BP neural network PID controller. The MPC predictive controller predicts the future system output according to the current system state, and the BP neural network PID controller calculates the control signals of each unit according to the input feedback signals and the MPC predictive output, realizing the collaborative control of the loading torque of the loading unit, the rotational speed of the driving unit, and the action of the clutch separation mechanism unit.

[0136] The collaborative control strategy includes: the driving unit eliminates the rotational speed error by controlling the frequency of the frequency converter through PID; the loading unit dynamically compensates the load inertia and generates a compensation current to adjust the loading torque; the clutch separation mechanism unit accurately controls the cylinder pressure according to the displacement error to ensure the stability of the engagement speed.

[0137] The collaborative control strategy is executed hierarchically:

[0138] (1) The drive unit eliminates the rotational speed error by PID controlling the frequency of the frequency converter, and the rotational speed error Δn ≤ ±1%.

[0139] (2) The loading unit dynamically compensates the load inertia J to generate a compensation current I 补偿 to adjust the loading torque:

[0140]

[0141] The compensation current is: I 补偿 = K p ·ΔJ + K i ·∫ΔJdt (18)

[0142] In formula (17), J is the total load inertia, J0 is the initial load inertia given according to the actual situation of the test bench, m i is the weight of the i-th additional mass block, r i is the equivalent radius of the i-th additional mass block, and N is the number of additional mass blocks;

[0143] In formula (18), I 补偿 is the compensation current; K p and K i are the proportional coefficient and the integral coefficient respectively, both from the MPC - BP neural network PID controller; ΔJ is the load inertia error.

[0144] (3) The clutch separation mechanism unit accurately controls the oil cylinder pressure according to the displacement error to ensure the stable engagement speed; the displacement control accuracy Δd ≤ ±0.1 mm.

[0145] When the clutch is in the engagement process, according to the rotational speed of the drive unit and the torque transmission characteristics of the clutch, the MPC - BP neural network PID controller automatically adjusts the loading torque of the loading unit to simulate the load change in actual operation during the test process; at the same time, according to the displacement feedback of the clutch separation mechanism unit, the engagement speed of the clutch is accurately controlled to ensure the stability and accuracy of the test process.

[0146] Furthermore, in step (5), the adaptive adjustment mechanism includes: incorporating the MPC prediction error into the BP neural network training data to enhance the self - learning ability of the model; real - time correcting the friction coefficient model based on the change in the friction plate temperature to suppress the torque fluctuation caused by heat fade; in case of sensor failure or torque mutation, switching to the current estimation mode and triggering the safety protection mechanism.

[0147] To improve the adaptability of the test bench to different types of clutches and complex test conditions, an adaptive adjustment mechanism is introduced into the control method. The MPC prediction error is incorporated into the BP neural network training data to enhance the model's self-learning ability. Through real-time analysis of the test data, the BP neural network can automatically identify the characteristics of the current test conditions and adjust its own structure and parameters according to pre-set rules and experience to better meet different test requirements.

[0148] Meanwhile, based on the BP neural network, a model of the thermal fade characteristics of the clutch friction plate is established to predict the temperature change trend, and the control parameters are adjusted in advance to suppress the torque fluctuation in the thermal load test. When testing clutches with different friction coefficients, the BP neural network can automatically adjust the parameters of the PID controller according to the characteristic parameters of the clutch to achieve the best control effect. In addition, according to abnormal situations during the test, such as sensor failures and equipment overload, the control strategy can be adjusted in a timely manner to ensure the safe conduct of the test.

[0149] The detection process of the adaptive adjustment mechanism includes:

[0150] (1) Thermal fade compensation: The friction coefficient model is:

[0151]

[0152] In Equation (19), μ0 is the friction coefficient at room temperature, λ is the attenuation coefficient, T is the real-time temperature, and T0 is the reference temperature.

[0153] (2) Fault diagnosis:

[0154] When the temperature of the friction plate exceeds the threshold (T > 350 °C), an alarm is triggered and the speed is automatically reduced to a safe speed, n safe = 0.5n 设定 .

[0155] When a torque mutation is detected (|ΔT > 10%T 设定 |), switch to the current estimation mode and isolate the abnormal signal. The current estimation mode: T 估算 = K t ·I 电机 , where K t is the motor torque constant, usually obtained from the technical manual of the motor factory.

[0156] When using the above collaborative control method to actually evaluate the performance of the tractor clutch, the main steps include:

[0157] Test bench construction: According to the design requirements of the tractor clutch test bench, build the test bench hardware system, including the drive unit, loading unit, and clutch release mechanism unit.

[0158] System initialization: Before the experiment starts, initialize the collaborative adaptive control system based on the MPC-BP neural network PID. Set the structure of the BP neural network, including the number of nodes in the input layer, hidden layer, and output layer; initialize the weights of the BP neural network as random values in the range of [-0.5, 0.5], and the bias term initial value is 0. Initialize the parameters of the MPC predictive controller, set the prediction horizon N p , control horizon N c and other parameters, and check and debug each unit of the test bench (loading unit, driving unit, clutch separation mechanism unit) to ensure the normal operation of the equipment.

[0159] The described collaborative adaptive control system is as Figure 1 shown, including a clutch test bench, speed sensor, torque sensor, displacement sensor, temperature sensor, MPC predictive controller, BP neural network PID controller, drive system, hydraulic system, loading system; the clutch test bench is connected to the speed sensor, torque sensor, displacement sensor, temperature sensor, the speed sensor, torque sensor, displacement sensor, temperature sensor are also connected to the MPC predictive controller, the MPC predictive controller is connected to the BP neural network PID controller, the BP neural network PID controller is connected to the drive system, hydraulic system, loading system, and the drive system, hydraulic system, loading system are connected to the clutch test bench.

[0160] Data acquisition and processing: During the experiment, measure the parameters such as the torque of the loading unit, the speed of the driving unit, the displacement of the clutch separation mechanism unit, and the temperature in real time through the measurement and control system (the measurement and control system includes a speed sensor, torque sensor, displacement sensor, temperature sensor, MPC predictive controller, BP neural network PID controller), and perform preprocessing such as filtering and denoising on the acquired data to improve the accuracy and reliability of the data. Use digital filtering algorithms, such as mean filtering, median filtering, etc., to filter the acquired data and remove noise interference. At the same time, perform normalization processing on the data and map it to the range of [-1, 1] to improve the learning efficiency and control accuracy of the BP neural network. Input the processed data as a feedback signal into the MPC predictive controller and the BP neural network PID controller.

[0161] Model calculation and control signal output: The MPC predictive controller predicts the future system output based on the input feedback signal, and the BP neural network PID controller calculates control signals such as the control current of the loading unit, the inverter input signal of the driving unit, and the control pressure of the clutch separation mechanism unit according to the feedback signal and the MPC prediction output parameters, and outputs these signals to the corresponding actuators (Figure 1 in the drive system, hydraulic system and loading system).

[0162] Test process monitoring and adjustment: During the test process, the operating status and test data of each unit of the test bench (loading unit, driving unit, clutch release mechanism unit) are monitored in real time. If abnormal test data or unstable conditions occur during the test process, the parameters or control strategies of the MPC-BP neural network PID controller are adjusted through an adaptive adjustment mechanism. At the same time, all data during the test process are recorded for subsequent analysis and processing.

[0163] Analysis of test results: After the test is completed, the test data are analyzed in depth to evaluate the performance of the clutch. According to the test results, the collaborative control method based on MPC-BP neural network PID is further optimized to improve the control effect and test quality.

[0164] The above are only embodiments of the present invention, and do not impose any form of limitation on the present invention. The present invention may also have other forms of embodiments based on the above structure and function, which will not be listed one by one. Therefore, any person skilled in the art, without departing from the scope of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A tractor clutch test bench collaborative control method based on MPC-BP neural network PID, characterized in that The following steps are involved: Step (1), establishing an enhanced mathematical model of the tractor clutch test bench system, including a discretized state space model of a loading unit, a driving unit and a clutch release mechanism unit; Step (2), designing an MPC predictive controller based on the enhanced mathematical model described in step (1), generating an optimal control sequence by rolling optimization of the objective function, and applying system constraints; Step (3), improving the BP neural network PID controller, taking the future system state output by the MPC prediction controller, the real-time feedback error and the error change rate as input, and dynamically adjusting the control parameters of the PID controller; Step (4), designing a multi-unit coordinated control strategy, combining MPC prediction with BP neural network PID control, and realizing coordinated adjustment of loading unit torque, driving unit speed and clutch release mechanism unit displacement; Step (5) introduces an adaptive adjustment mechanism to improve the system's adaptability to different working conditions and clutch models through thermal decay modeling and fault diagnosis.

2. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 1, characterized in that: Step (1) mathematically modeling the loading unit, driving unit and clutch release mechanism unit of the tractor clutch test bench, and analyzing the input-output relationship and dynamic characteristics of each unit; wherein the loading unit establishes the loading torque T and the excitation current I based on the eddy current retarder principle. c The driving unit builds a dynamic relationship model between the motor speed n and the inverter input signal f based on the characteristics of the variable frequency speed regulation motor; the clutch release mechanism unit builds a dynamic response model of the cylinder displacement d and the control pressure P based on the working principle of the hydraulic system; and, based on the mathematical models of the loading unit, the driving unit and the clutch release mechanism unit, the mathematical models of each unit are integrated to build a discrete state space model suitable for MPC, which provides a basis for the subsequent control algorithm design.

3. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 1, characterized in that: The objective function in step (2) is: In formula (8), y ref is the reference trajectory, which is set according to the test requirements; y(k+i) represents the system vector for the next i steps; Q and R are weight matrices, which are used to adjust the weight of the system output tracking the reference trajectory and the change weight of the control input; Solve the optimal control sequence U by rolling optimization objective function * =[u(k),u(k+1),.....,u(k+N C -1)], taking the first term u(k) as the current control variable. During the optimization process, the system constraints are considered to ensure the feasibility of the prediction results.

4. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 1, characterized in that: In step (3), the system output prediction values ​​at multiple future moments output by the MPC prediction controller and the errors between the speed, torque, displacement setting values ​​of each unit, the speed error change rate, and the friction plate temperature are used as inputs of the BP neural network. Based on these inputs, the BP neural network outputs the proportional coefficient, integral coefficient, and differential coefficient of the PID controller after network learning and training.

5. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 4, characterized in that: The BP neural network input layer includes five nodes: speed error, speed error change rate, torque error, displacement error, and friction plate temperature, which are: Speed ​​error e n =n 设定 -n 预测 ; Speed ​​error change rate Δe n =e n (t)-e n (t-1); Torque error e T =T 设定 -T 预测 ; Displacement error e d =d 设定 -d 预测 ; Friction plate temperature T (0~400℃); The hidden layer uses 15 nodes, and the activation function is the Sigmoid function: The hidden layer output is calculated as: The output layer generates PID parameters K p , K i , K d , the calculation formula is: In formula (11) to formula (13), h j is the hidden layer output, ω and b are the weight and bias terms.

6. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 1, characterized in that: The coordinated control strategy in step (4) includes: the driving unit controls the frequency of the inverter through PID to eliminate the speed error; the loading unit dynamically compensates for the load inertia and generates a compensation current to adjust the loading torque; the clutch release mechanism unit accurately controls the cylinder pressure according to the displacement error to ensure the stability of the engagement speed; When the clutch is in the engagement process, the loading torque of the loading unit is automatically adjusted through the MPC-BP neural network PID controller according to the speed of the drive unit and the torque transmission characteristics of the clutch, so that the test process simulates the load changes in actual work; at the same time, according to the displacement feedback of the clutch separation mechanism unit, the engagement speed of the clutch is accurately controlled to ensure the stability and accuracy of the test process.

7. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 1, characterized in that: The adaptive adjustment mechanism in step (5) includes: incorporating the MPC prediction error into the BP neural network training data to enhance the model's self-learning ability; correcting the friction coefficient model in real time based on the temperature change of the friction plate to suppress torque fluctuations caused by thermal decay; and switching to the current estimation mode and triggering the safety protection mechanism in response to sensor failure or torque mutation.

8. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 7, characterized in that: The friction coefficient model is: In formula (19), μ0 is the friction coefficient at room temperature, λ is the attenuation coefficient, T is the real-time temperature, and T0 is the reference temperature.

9. The tractor clutch test bench coordinated control method based on MPC-BP neural network PID as claimed in claim 7 or 8, characterized in that: When the friction plate temperature exceeds the threshold, the speed will automatically decrease to a safe speed and trigger an alarm. The threshold is T>350℃; when a torque mutation is detected, that is, |ΔT>10%T 设定 |, switch to current estimation mode and isolate abnormal signals, current estimation mode: T 估算 =K t I 电机 , where K t is the motor torque constant.

Images