Prediction method for belt discharge torque of automobile wet-type double clutches in no-load state

Abstract

The invention belongs to the field of neural network prediction, and particularly relates to a method for predicting belt discharge torque of an automobile wet-type double clutch in a no-load state. The invention provides a hybrid model based on a traditional wet clutch belt discharge torque model and a particle swarm optimization back propagation (PSO-BP) neural network, and the model precision is improved through test data. By applying the wet clutch driving torque prediction method, the rising change of the wet clutch driving torque at the high rotating speed can be accurately predicted, and the prediction result is more accurate compared with a traditional model. Compared with a traditional belt row torque mechanism model, the hybrid prediction model has higher calculation efficiency and more excellent stability.

Description

Technical Field

[0001] The invention belongs to the field of neural network prediction, and in particular relates to a method for predicting the belt torque of a wet dual clutch of an automobile in a no-load state. Background Art

[0002] Wet clutch has the advantages of stable operation, safety and reliability, good heat dissipation performance and large torque transmission, and is widely used in automobile transmission systems. When the wet clutch is in an unloaded state, there is a certain gap between the friction plate and the steel plate, and the friction plate and the steel plate can rotate independently. At this time, if the rotation speed of the friction plate and the steel plate is different, the lubricating oil between the two will be affected by viscous shear to produce a displacement torque. The displacement torque not only consumes the power of the engine, but also is accompanied by a significant temperature rise of the friction pair and the lubricating oil, which endangers the reliability of the vehicle transmission system and reduces the fuel economy of the vehicle. Its negative impact cannot be ignored. Therefore, the study of the displacement torque of the wet clutch is of great significance to improving the performance of the automatic shift transmission device of our vehicle and enhancing the mobility of the vehicle.

[0003] The problem of wet clutch torque has attracted the attention of researchers at home and abroad since the 1970s. Over the past few decades, a lot of research has been carried out in experimental research, analytical modeling and numerical simulation, and remarkable results have been achieved. With the vigorous development of computer technology, the use of data-driven methods for prediction has been widely used in various fields. Recently, related studies have begun to use data-driven methods to study the wet clutch torque. Although domestic and foreign scholars have conducted a lot of research on the wet clutch torque and achieved remarkable results, there are generally problems such as long time consumption and high experimental costs, which have certain limitations for studying the wet clutch torque. At present, there is no technical solution for accurately predicting the wet clutch torque in a wide speed range. The present invention proposes a hybrid prediction method and system that combines the traditional wet clutch torque model with the particle swarm optimization back propagation (PSO-BP) neural network. Through the learning of a large amount of experimental data, it can accurately predict the recovery change of the high-speed torque, providing theoretical support for optimizing the transmission system and improving vehicle reliability. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention proposes a method for predicting the belt torque of a wet dual clutch of an automobile in a no-load state.

[0005] The present invention is achieved through the following technical solutions: A method for predicting the belt torque of a wet dual clutch of an automobile under no-load state comprises the following steps: S1. Low-speed stage: Prediction of viscous shear band torque using low-speed mechanism model The specific operation is: the finite volume method is used to solve the unified control equation of the gap rotating flow field of the wet clutch friction pair, and the density and viscosity of the gap rotating flow field are obtained according to the mass flow conservation principle and the density-pressure and viscosity-pressure equations. Then, according to Newton's internal friction law, the calculation expression of the wet clutch displacement torque is obtained by integration; the displacement torque of the friction pair is obtained using the calculation expression.

[0006] S2, high-speed stage: PSO-BP neural network model is used to predict the recovery of belt torque in high-speed stage The specific operation is as follows: the belt torque of the wet clutch with different design parameters under different working conditions is obtained through experiments, the obtained belt torque data is used as the training set, and the appropriate number of neurons, training times, learning rate and error threshold model parameters are selected to construct a PSO-BP neural network model, and iterate and optimize continuously to achieve accurate prediction of the belt torque in the high-speed stage.

[0007] Furthermore, the calculation expression of the belt torque in step S1 is:

[0008] Where, Z is the number of friction pairs; μ l is the flow field viscosity in the non-slot area; μ g is the flow field viscosity in the slot area; T d is the friction pair belt torque; the number is Ng; θ l is the angle corresponding to the platform area; the inner radius of the friction pair r i and outer radius r o ; w is the angular velocity difference between the friction plate and the steel plate; h0 is the friction pair clearance; θg is the angle corresponding to the groove area; the groove depth is h g .

[0009] Furthermore, the specific operations of step S2 are as follows: (1) Data acquisition: Build a multi-plate wet clutch torque measurement test bench and design a torque measurement test process to obtain the operating data of wet clutches with different design parameters under different working conditions. Use the data as a training set to train the PSO-BP neural network model. (2) Neural network training: normalizing the sample data set, using the normalized sample data as input vectors, using the normalized sample labels as output vectors, training a pre-built PSO-BP neural network model on the normalized sample data set, and using the trained PSO-BP neural network model as a wet clutch displacement torque prediction model after reaching convergence conditions; (3) Prediction output: The main influencing factors of the wet clutch torque to be tested are taken as the input vector to be tested, and the input vector to be tested is input into the wet clutch torque prediction model after normalization to obtain an initial prediction output vector. The initial prediction output vector is denormalized to obtain a final prediction output vector, thereby realizing the wet clutch torque prediction; wherein, the initial prediction output vector is the normalized prediction value of the wet clutch torque, and the final prediction output vector is the denormalized prediction value of the wet clutch torque.

[0010] Furthermore, the algorithm flow of the PSO-BP neural network model is as follows: (1) Input the sample data collected by the wet clutch torque measurement test bench; (2) Establish the BP neural network structure and initialize it; (3) Initialize the position and velocity of the particle swarm, calculate the fitness value, and obtain the individual extreme value and global extreme value of the current population; (4) Iterative optimization allows each particle to continuously adjust its speed and position according to individual extreme values ​​and global extreme values, thereby obtaining the optimal values ​​of the BP neural network weights and thresholds; (5) The BP neural grid after PSO optimization iteratively calculates the output signal according to the input signal, weight and threshold until the minimum error accuracy or the maximum number of iterations is reached to obtain the final result.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a hybrid model that combines the traditional wet clutch displacement torque model and the particle swarm optimization back propagation (PSO-BP) neural network. It can quickly predict the displacement torque of an unloaded wet clutch over a wide speed range based on the clutch operating parameters and design parameters, and improve the model prediction accuracy through a large amount of experimental data.

[0012] 1. The wet clutch torque prediction method of the present invention can accurately predict the recovery change of the wet clutch torque at high speed, and the prediction result is more accurate than the traditional model.

[0013] 2. Compared with the traditional belt torque mechanism model, the hybrid prediction model of the present invention has higher computational efficiency and better stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a schematic diagram of the wet friction pair structure; Figure 2 It is a multi-plate wet clutch with row torque measurement test bench and test process; Figure 3 This is a schematic diagram of the clutch box structure; Figure 4 This is a comparison chart of the theoretical results of the multi-plate wet clutch torque mechanism model and the experimental results with the speed change curve; Figure 5 It is a schematic diagram of the BP neural grid structure; Figure 6 is the average relative error graph for different numbers of hidden layer neurons; Figure 7 It is the flow chart of PSO-BP neural network algorithm; Figure 8 Comparison chart of prediction results for the test set: (a) Comparison curve between the predicted value and the experimental value of the test set; (b) Correlation coefficient plot of the test set; Fig. 9 It is the torque prediction curve of the hybrid model with displacement; Fig.10 Comparison chart of prediction results for different input parameters: (a) Comparison curve between predicted values ​​and experimental values ​​of different input parameters; (b) Correlation coefficient diagram of prediction results with different input parameters; Fig.11 It is the result diagram of evaluation index parameters under different input parameter conditions; Fig.12 Comparison chart of test set prediction results for different prediction methods: (a) Comparison of BP prediction results; (b) Comparison of GA-BP prediction results; (c) Comparison of PSO-BP prediction results; Fig.13 This is a comparison chart of evaluation indicators of different prediction methods; Fig.14 The diagram shows the belt torque prediction results of different prediction methods. DETAILED DESCRIPTION

[0015] The present invention is described in detail below in conjunction with the accompanying drawings. A person skilled in the art can obtain the hybrid prediction model of the present invention and implement its basic functions according to the following method.

[0016] 1. Belt torque mechanism model 1.1. Gap Rotation Flow Field Control Equations Wet friction pair structure such as Figure 1 The friction plate with radial grooves rotates around the z axis with an angular velocity of w (rad / s), while the steel plate remains stationary; the groove width of the friction plate is w g , the number of slots is N g , the groove depth is h g; h is the oil film thickness, h0 is the friction pair clearance, in the groove area h=h0+h g , in the platform area h=h0;θ g is the angle corresponding to the slot area, θ l is the angle corresponding to the platform area, and the circumferential groove platform ratio is δ=θ g / (θ g +θ l ); Inner radius of friction pair r i and outer radius r o The corresponding inlet and outlet oil pressures are p i and p o .

[0017] According to the structural characteristics and actual working conditions of the wet clutch, the influence of the centrifugal force and inertial force generated by the rotating disk on the fluid cannot be ignored. This embodiment assumes that the flow field fluid is an incompressible Newtonian viscous fluid with steady flow, and adopts the steady-state cylindrical coordinate Reynolds equation of ideal fluid under isothermal conditions. According to the core idea of ​​the Elrod method, universal variables and switching functions are introduced to establish a unified control equation that describes both the oil film area and the cavitation area. Steady-state cylindrical coordinate Reynolds equation of ideal fluid under isothermal conditions: (1) In the formula, ρ is the fluid density; h is the oil film thickness; μ is the fluid dynamic viscosity; p is the fluid film pressure; w is the angular velocity difference between the friction plate and the steel plate. Since the wet clutch is in the brake working condition, the angular velocity of the friction plate is the angular velocity difference.

[0018] Introduce the universal variable φ and the switching function F, which are defined as follows: (2) (3) Where, P a is atmospheric pressure; P c is the cavitation pressure; ρ c is the density of lubricating oil; in the complete oil film area, F=1, ϕ is the dimensionless oil film pressure; in the cavitation area, F=0, 1+ϕ is the mixing ratio of oil and gas. That is: (4) Substituting equations (2) and (3) into equation (1) and making them dimensionless, we obtain the unified control equation for the rotating flow field of the friction pair clearance of the wet clutch: (5) Wherein, the dimensionless parameters are defined as follows: (6) Among them, θ0 is the angle corresponding to a calculation area; the dimensionless parameter Re is called the Reynolds number of oil flow; γ is called the bearing number.

[0019] According to the actual working conditions of the wet clutch, the boundary conditions for solving equation (5) are: (7) 1.2. Belt torque calculation model The finite volume method is used to solve the unified control equation of the rotating flow field of the gap of the wet clutch friction pair. According to the mass flow conservation principle and the density-pressure and viscosity-pressure equations, the density and viscosity of the gap rotating flow field are obtained. Then, according to Newton's internal friction law, the calculation expression of the wet clutch belt torque is obtained by integration. In this embodiment, the belt torque of the rotating flow field of the friction pair gap with radial grooves is divided into two parts: the groove area and the non-groove area. The friction pair belt torque expression is: (8) Where, Z is the number of friction pairs; μ l is the flow field viscosity in the non-slot area; μ g is the flow field viscosity in the slot area; T d is the friction pair belt torque.

[0020] 2. Experimental Design In order to better simulate the changes in the belt torque of the wet clutch in actual working conditions, we built a multi-plate wet clutch belt torque measurement test bench and designed a belt torque measurement test process. The data collected by each sensor was used as a training set to train the PSO-BP neural network model.

[0021] 2.1、Wet-disc clutch torque measurement test bench According to the no-load characteristics and assembly structure of the wet clutch, we built a multi-plate wet clutch torque measurement test bench, such as Figure 2 As shown. The test bench is mainly composed of a power drive system, a clutch box, a hydraulic system and a data acquisition system; the power drive system is mainly composed of a 45KW motor (speed range is 0~5000r / min), a 15KW motor (speed range is 0~3000r / min), a transmission belt and a transmission shaft. The two motors drive the friction plate and the steel plate to rotate through the transmission belt and the transmission shaft respectively, so as to realize the independent rotation of the friction plate and the steel plate. In this embodiment, the steel plate remains stationary. The clutch box is mainly composed of a friction plate, a steel plate, an inner and outer drum of the clutch, an adjustment bolt, etc. The number of friction pairs, the clearance of the friction pairs and the test specimen can be adjusted by changing the size of the adjustment bolt. The hydraulic system is mainly composed of a hydraulic pump station, an oil suction pump, an oil supply pump and a circulation pipeline, etc., which can adjust the oil supply flow and the temperature of the lubricating oil. The data acquisition system is composed of a torque sensor, a temperature sensor and a flow sensor, which are used to collect test data. The specifications of each sensor are shown in Table 1.

[0022] Table 1 Sensor parameters

[0023] 2.2 Experimental plan The test process of multi-plate wet clutch with displacement torque measurement is as follows Figure 2 Before the test begins, first check the test equipment, clean the friction plate and steel plate test specimens, remove surface impurities, and then Figure 3 According to the position relationship shown, install the snap ring, friction plate, steel plate, adjustment bolt and limit plate, etc. The lubricating oil used in the test is SAE 15W-40. When measuring the torque under no-load conditions, the hydraulic pump station oil is heated to the required temperature according to the lubricating oil temperature specified in the test, and the hydraulic pump station is turned on to supply oil to the wet clutch package box according to the oil supply flow specified in the test. Finally, adjust the 45KW motor speed according to the wet clutch working condition, increase it slowly to the specified speed, and keep it for 5 seconds to stabilize the speed. After the speed stabilizes, record the measurement results of each sensor, adjust the motor speed to the next test setting value, and when the number of tests reaches 100, change the wet clutch working condition and repeat the above acquisition process to obtain the no-load wet clutch torque test value under different working conditions. The test working condition parameters of the wet clutch friction pair are shown in Table 2.

[0024] Table 2 Multi-plate wet clutch test parameters

[0025] 2.3 Results Analysis Based on the data collected from the wet clutch torque measurement test bench under the working conditions in Table 2, the working conditions of friction pair clearance of 0.5 mm, oil flow rate of 18 L / min, oil temperature of 40 °C, number of grooves of 20, groove width of 12 mm, groove depth of 0.5 mm, and number of friction pairs of 10 were selected, and the theoretical results of the mechanism model and the experimental results were obtained as follows: Figure 4 As shown. Figure 4It can be seen that the theoretical results of the mechanism model and the experimental results are highly consistent before point A. When the friction plate speed exceeds point A, the difference between the theoretical results and the experimental results gradually increases. This is because the traditional model believes that the oil film in the friction pair has been broken at this time, and a large amount of air gradually enters the gap of the friction pair, causing the viscosity of the lubricating medium to gradually decrease. The belt torque decreases with the increase of the speed, but there is a matching gap between the friction plate and the steel plate and the inner and outer hubs of the clutch. When the friction plate reaches a certain speed, under the action of system disturbance and fluid force, the friction plate and the steel plate collide and rub, thereby causing the belt torque to rise. It can be seen that the traditional mechanism model has certain limitations in studying the wet clutch belt torque. In order to solve this problem, this embodiment proposes a data-driven method to improve and correct the model. Point B is the wet clutch belt torque at the highest speed (5000r / min).

[0026] 3. Hybrid Model In order to solve the problem that the traditional mechanism model has certain limitations in predicting the wet clutch torque, this embodiment proposes a PSO-BP neural network model to predict the wet clutch torque after point A, so as to improve the speed and accuracy of the wet clutch torque prediction.

[0027] 3.1 BP Neural Network Model In the BP neural network, each layer of neurons is connected to all neurons in the previous or next layer through weights (v ml and u l ) and threshold (b l The input data is forward propagated through the network to obtain the network output, and the network output is compared with the expected output to calculate the error. Then, the error is back-propagated back to the network layer by layer through the back-propagation algorithm to update the weights of neurons in each layer (v ml and u l ) and threshold (b l and d) to reduce the error until the requirements are met. The input layer nodes should be able to fully reflect the factors affecting the belt torque. The main factors affecting the belt torque during the operation of the wet clutch are the friction plate speed, friction pair clearance, oil flow rate, oil temperature, number of grooves, groove width, groove depth and number of friction pairs. Therefore, this embodiment uses the friction plate speed, friction pair clearance, oil flow rate, oil temperature, number of grooves, groove width, groove depth and number of friction pairs as the input signals of the BP neural network, and the belt torque as the output result of the BP neural network. The schematic diagram of the BP neural network structure is shown in Figure 5 shown.

[0028] Since the wet clutch torque is affected by multiple parameters, and the units and magnitudes of each parameter are different, there will be problems such as reduced prediction model accuracy and non-convergence. Therefore, the data must be normalized for BP neural network training. The normalization formula is:

[0029] (9) In the formula: y is the normalized data; x is the test data; x min and x max are the corresponding maximum and minimum values ​​of a certain parameter respectively.

[0030] In BP neural network, the number of neurons directly affects the learning ability and generalization ability of the neural network. If the number of neurons is too small, underfitting may occur, reducing the learning ability of the neural network; if the number of neurons is too large, the network may be too complex, wasting computing costs and reducing the generalization ability of the network. Therefore, the number of hidden layer neurons should be reasonably selected, and the range of the number of hidden layer neurons can be obtained by the empirical formula: (10) In the formula, l is the number of neurons; m is the number of neurons in the input layer; n is the number of neurons in the output layer; a is an integer between 1 and 10, so the number of neurons in the hidden layer is between 4 and 13.

[0031] This embodiment uses the average relative error as an evaluation index, compares the average relative errors corresponding to different numbers of hidden layer neurons, and further determines the number of hidden layer neurons. The results are as follows: Figure 6 As shown. Figure 6 It can be seen that when the number of hidden layer neurons is 12, the average relative error is the smallest. Therefore, the number of hidden layer neurons is selected as 12, the topology of the BP neural network is 8-12-1, and the number of training times, learning rate and target error are set to 1000, 0.01 and 0.0001 respectively.

[0032] 3.2 PSO Algorithm Particle Swarm Optimization (PSO) is an advanced global optimization intelligent algorithm. Its basic idea is to use the information sharing and collaboration between individuals in a group to find the optimal solution to a problem. When searching for the optimal solution, the algorithm is initialized as a group of random particles, and then finds the optimal solution through iteration. During each iteration, the particles will find the optimal solution according to their own optimal solution (P i ) and the optimal solution of the entire population (P g ) to update its own speed and position, so as to move in the search space and approach the optimal solution. The process is: (11) (12) In the formula, ω is the inertia weight; i=1, 2, ..., n; j=1, 2, ..., D; D is the number of optimized parameters; k is the current number of iterations; c1 and c2 are individual learning factors and social learning factors, c1=c2=1.5; r1 and r2 are random constants between [0,1]. In order to ensure the exploration ability and convergence speed of the algorithm, the particle position range is [-X max , Xmax], take X max =1.5; the particle velocity range is [-V max ,V max ], take V max =1.5, V ij (k) is the velocity of the i-th particle after the k-th iteration, X ij (k) is the position of the i-th particle after the k-th iteration, P ij (k) is the individual optimal solution of the i-th particle, P gj (k) is the optimal solution for the entire population.

[0033] BP neural network has the disadvantage of being easily trapped in local extreme values, and it is difficult to find the global optimal solution. The PSO algorithm has global search capabilities. Combining the two can enable the PSO algorithm to optimize the initial weights and thresholds of the BP neural network, avoiding the problem that the BP neural network is easily trapped in local extreme values ​​during the training process, and enable the PSO-BP neural network to find the optimal solution in the global scope, thereby improving the prediction accuracy. The PSO-BP neural network algorithm process is as follows Figure 7 As shown, the specific process is: (1) Input the sample data collected by the wet clutch torque measurement test bench; (2) Establish the BP neural network structure and initialize it; (3) Initialize the position and velocity of the particle swarm, calculate the fitness value, and obtain the individual extreme value and global extreme value of the current population; (4) Iterative optimization allows each particle to continuously adjust its speed and position according to individual extreme values ​​and global extreme values, thereby obtaining the optimal values ​​of the BP neural network weights and thresholds; (5) The BP neural grid after PSO optimization iteratively calculates the output signal according to the input signal, weight and threshold until the minimum error accuracy or the maximum number of iterations is reached to obtain the final result.

[0034] 3.3 Model Validation In this section, 300 sets of data recorded by the wet clutch torque measurement test bench under the working conditions in Table 2 are selected. The first 200 sets of data are used as the training set of the PSO-BP neural network model, and the last 100 sets of data are used as the test set of the PSO-BP neural network model to verify the accuracy of the model in predicting the wet clutch torque. The prediction results of the test set are shown in Figure 8 As shown. Figure 8 It can be seen that the predicted value of the PSO-BP neural network model is close to the experimental value. The trained PSO-BP neural network model has a high prediction accuracy and can better realize the prediction of the wet clutch displacement torque.

[0035] The belt torque value after point A can be obtained from the PSO-BP neural network prediction model. By combining the mechanism model with the PSO-BP neural network model, the hybrid model belt torque prediction result can be obtained, such as Fig. 9 As shown. Fig. 9 It can be seen that the corrected belt torque prediction results are more consistent with the test results, and the average relative error is reduced from 53.04% to 14.45%, indicating that the accuracy of predicting the wet clutch belt torque is effectively improved by improving and correcting the model through data-driven methods.

[0036] 4. Model performance analysis This section compares the prediction performance of the model under different input parameter conditions, the prediction results of different prediction methods and the prediction error of the belt torque at the highest speed (5000r / min), and comprehensively evaluates the stability and accuracy of the model.

[0037] 4.1 Model Evaluation Indicators MAE represents the average value of the absolute value of the error between the predicted value and the true value, which can directly reflect the actual situation of the predicted value error; RMSE represents the average deviation between the predicted value and the true value, which can be used to measure the deviation between the predicted value and the true value; R2 represents the similarity of variable changes, which is used to measure the strength and direction of the linear relationship between variables. The smaller the MAE and RMSE, the higher the prediction accuracy of the model. 2 The larger it is, the better the model fit is.

[0038] The mean absolute error (MAE), root mean square error (RMSE) and correlation coefficient (R) between the predicted belt torque and the actual belt torque are calculated. 2 ) is used as an evaluation index to measure the performance of each neural network model and more comprehensively measure the prediction performance of the model. The calculation formulas are: (13) (14) (15) Where Y i and Y i ’ are the experimental value and predicted value in the i-th sample respectively; is the average of the predicted values ​​in the sample; n is the number of samples.

[0039] 4.2 Analysis of model prediction performance under different input parameters In order to simplify the PSO-BP neural network model, we retain the operating parameters that affect the wet clutch torque, namely, friction plate speed, friction pair clearance, oil flow rate, and oil temperature. Therefore, we constructed a PSO-BP neural network prediction model with 4 inputs and 1 output, and compared its prediction results with those of the PSO-BP neural network model with 8 inputs and 1 output. Fig.10 As shown. Fig.10 It can be seen that the prediction accuracy of the PSO-BP neural network model with 8 inputs and 1 output is higher than that of the PSO-BP neural network model with 4 inputs and 1 output, and the prediction results are closer to the experimental results.

[0040] After calculation, the MAE, RMSE and R 2 The results are as follows Fig.11 As shown. Fig.11 It can be seen that the MAE, RMSE, and R 2 The MAE, RMSE, and R 2 The MAE and RMSE of the 8-input parameter model decreased by 30.45% and 28.55% respectively compared with the 4-input parameter model. 2 The increase from 0.92 to 0.96 further illustrates that the 8-input and 1-output model proposed in this embodiment has a higher fit to the test data.

[0041] 4.3 Analysis of model prediction results under different prediction methods In order to verify the effectiveness of the method, this section compares the PSO-BP neural network with the BP neural network and the GA-BP neural network. Fig.12 This is a comparison chart of the prediction results of each neural network test set. Fig.12 It can be seen that the BP neural network has a larger deviation from the experimental value compared with the GA-BP neural network and PSO-BP neural network models. This is because the BP network is a local search optimization method, which is easy to fall into the local minimum rather than the global optimal solution during the training process. In general, the performance of the PSO-BP neural network is better than that of the GA-BP neural network and the BP neural network, and it can achieve accurate prediction of the belt torque.

[0042] In order to more intuitively compare the performance differences between PSO-BP neural network, BP neural network and GA-BP neural network, Fig.13 This is a comparison chart of evaluation indicators of different prediction methods. Fig.13 It can be seen that the MAE and RMSE of the PSO-BP neural network are smaller than those of the BP neural network and the GA-BP neural network. 2 Both are greater than the BP neural network and GA-BP neural network, which further illustrates that the performance of the PSO-BP neural network model is better, and the hybrid model based on the mechanism model and the PSO-BP neural network has a certain degree of accuracy.

[0043] 4.4. Analysis of belt torque prediction error at maximum speed The belt torque will not only increase the wear of the friction pair and reduce the safety and reliability of the vehicle, but also cause a sharp increase in the heat of the clutch, resulting in thermal damage to the friction plate and making the clutch unable to work properly. Accurately predicting the belt torque of the wet clutch will help improve the efficiency and safety of the vehicle transmission system and reduce the occurrence of thermal failure of the friction pair. Therefore, it is very necessary to accurately predict the belt torque of the wet clutch.

[0044] In order to verify the reliability of the hybrid model, the working conditions of friction pair clearance of 0.5mm, oil flow rate of 18 L / min, oil temperature of 40℃, number of grooves of 20, groove width of 12mm, groove depth of 0.5mm, and number of friction pairs of 10 were selected. The prediction error of the belt torque at the highest speed (5000r / min) of this working condition was used as an indicator to measure the prediction performance of each model. Fig.14 The belt torque prediction results of different prediction methods are shown in Figure 2. Fig.14 It can be seen that the prediction errors of the BP, GA-BP and PSO-BP neural networks for the belt torque are 17.549, 10.375 and 5.182 respectively, and the PSO-BP neural network has the smallest error in predicting the belt torque, which further verifies the accuracy and reliability of the model proposed in this embodiment.

[0045] The present invention proposes a hybrid model combining a mechanism model and a PSO-BP neural network to solve the problem that the traditional mechanism model has certain limitations in predicting the wet clutch torque. The accuracy and stability of the hybrid model are verified through experiments, and the following conclusions are obtained: (1) The proposed hybrid model combines the mechanism model and the data-driven model. Compared with the traditional mechanism model, the prediction results of the hybrid model are more consistent with the experimental results. It can quickly and accurately predict the recovery change of the belt torque in the high-speed stage. The average relative error is reduced from 53.04% to 14.45%, which improves the accuracy of predicting the belt torque of the wet clutch.

[0046] (2) The stability and accuracy of the hybrid model were comprehensively evaluated. The proposed hybrid model has significant improvements over the traditional neural network model. The MAE and RMSE of the 8-input 1-output model are lower than those of the 4-input 1-output model, and the prediction performance is better. In addition, the error of the model in predicting the belt torque is the smallest, indicating that the model has a certain reliability for studying the belt torque characteristics.

Claims

1. A method for predicting the belt torque of a wet dual clutch in an automobile under no-load condition, characterized in that: The steps include: S1. Low-speed stage: Prediction of viscous shear band torque using low-speed mechanism model The specific operation is as follows: the unified control equation of the gap rotating flow field of the wet clutch friction pair is solved by the finite volume method, the density and viscosity of the gap rotating flow field are obtained according to the mass flow conservation principle and the density-pressure and viscosity-pressure equations, and then the calculation expression of the wet clutch belt displacement torque is obtained by integration according to Newton's internal friction law; the belt displacement torque of the friction pair is obtained by using the calculation expression; S2, high-speed stage: PSO-BP neural network model is used to predict the recovery of belt torque in high-speed stage The specific operation is as follows: the belt torque of the wet clutch with different design parameters under different working conditions is obtained through experiments, the obtained belt torque data is used as the training set, and the appropriate number of neurons, training times, learning rate and error threshold model parameters are selected to construct a PSO-BP neural network model, and iterate and optimize continuously to achieve accurate prediction of the belt torque in the high-speed stage.

2. The method for predicting the belt torque of a wet dual clutch in an unloaded state of an automobile according to claim 1, characterized in that: The calculation expression of the belt torque in step S1 is: , where, Z is the number of friction pairs; μ l is the viscosity of the flow field in the non-slot area; μ g is the flow field viscosity in the slot area; T d is the friction pair belt row torque, the number of grooves is Ng; θ l is the angle corresponding to the platform area; the inner radius of the friction pair r i and outer radius r o ; w is the angular velocity difference between the friction plate and the steel plate; h0 is the friction pair clearance; θg is the angle corresponding to the groove area; the groove depth is h g .

3. The method for predicting the belt torque of a wet dual clutch in an unloaded state of an automobile according to claim 1, characterized in that: The specific operation of step S2 is as follows: (1) Data acquisition: Build a multi-plate wet clutch torque measurement test bench and design a torque measurement test process to obtain the operating data of wet clutches with different design parameters under different working conditions. Use the data as a training set to train the PSO-BP neural network model. (2) Neural network training: normalizing the sample data set, using the normalized sample data as input vectors, using the normalized sample labels as output vectors, training a pre-built PSO-BP neural network model on the normalized sample data set, and using the trained PSO-BP neural network model as a wet clutch displacement torque prediction model after reaching convergence conditions; (3) Prediction output: The main influencing factors of the wet clutch torque to be tested are taken as the input vector to be tested, and the input vector to be tested is input into the wet clutch torque prediction model after normalization to obtain an initial prediction output vector. The initial prediction output vector is denormalized to obtain a final prediction output vector, thereby realizing the wet clutch torque prediction; wherein, the initial prediction output vector is the normalized prediction value of the wet clutch torque, and the final prediction output vector is the denormalized prediction value of the wet clutch torque.

4. The method for predicting the belt torque of a wet dual clutch in an unloaded state of an automobile according to claim 3, characterized in that: The algorithm flow of the PSO-BP neural network model is: (1) Input the sample data collected by the wet clutch torque measurement test bench; (2) Establish the BP neural network structure and initialize it; (3) Initialize the position and velocity of the particle swarm, calculate the fitness value, and obtain the individual extreme value and global extreme value of the current population; (4) Iterative optimization allows each particle to continuously adjust its speed and position according to individual extreme values ​​and global extreme values, thereby obtaining the optimal values ​​of the BP neural network weights and thresholds; (5) The BP neural grid after PSO optimization iteratively calculates the output signal according to the input signal, weight and threshold until the minimum error accuracy or the maximum number of iterations is reached to obtain the final result.

Images