Intelligent tire cornering characteristic estimation method based on tire model embedded in neural network

Abstract

The present invention belongs to the application field of tire dynamic characteristics, and specifically relates to an intelligent tire side slip characteristic estimation method based on tire model embedded neural network. The method arranges PVDF strain sensors on the inner wall of the tire, conducts tire mechanical characteristic tests under pure side slip conditions, obtains tire mechanical characteristic test data and PVDF strain sensor test data; uses tire mechanical characteristic test data to identify the UniTire steady-state tire model; establishes a neural network model, embeds the identified UniTire tire model into the neural network loss function, and obtains a neural network model embedded with tire model constraints; extracts PVDF strain sensor test data and intelligent tire side slip characteristic test data in the tire contact patch, and trains the neural network model; uses the trained neural network model to estimate the tire vertical force, lateral force and side slip angle. The method plays a very important role in improving the calculation efficiency of the neural network model and the estimation accuracy of the intelligent tire side slip characteristics.

Description

Technical Field

[0001] The invention belongs to the field of application of tire dynamic characteristics, and in particular relates to an intelligent tire side slip characteristic estimation method based on embedding a tire model into a neural network. Background Art

[0002] The development of automobile electrification and intelligence has put forward higher requirements on the state perception of vehicle chassis. As tires are components that are in contact with the road surface and continuously act during vehicle driving, accurate estimation of their mechanical properties is of great significance to improving the accuracy of vehicle chassis state perception and improving vehicle active safety.

[0003] At present, typical steady-state tire models include the UniTire model and the Magic Formula model. The MagicFormula model is an empirical tire model. To ensure the accuracy of the model, a large amount of tire test data is required to fit the model parameters. The UniTire model is a semi-empirical tire model based on theory, which has good tire mechanical property expression accuracy and prediction ability. Using the PVDF strain sensor data in the tire contact patch, the side slip characteristics of the smart tire are estimated based on the data-driven neural network model, which has problems such as low computational efficiency and insufficient accuracy. In order to improve the computational efficiency of the neural network model and the estimation accuracy of the side slip characteristics of the smart tire, a method for estimating the side slip characteristics of the smart tire based on the tire model embedded in the neural network was developed. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides an intelligent tire cornering characteristic estimation method based on tire model embedded in neural network, which is characterized in that the UniTire steady-state tire model is embedded in the neural network loss function, and the tire cornering characteristics output by the neural network model training are physically constrained, thereby improving the computational efficiency of the neural network model and the estimation accuracy of the intelligent tire cornering characteristics.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] Step 1: Arrange a PVDF (polyvinylidene difluoride) strain sensor on the inner wall of the tire, install the tire on a tire six-component force testing machine, set test conditions according to test requirements, conduct a tire mechanical property test under a pure cornering condition, and obtain tire mechanical property test data and PVDF strain sensor test data under the pure cornering condition;

[0007] Step 2: Using the test data of tire vertical load, footprint update speed, slip angle and lateral force under pure cornering condition, the UniTire steady-state tire model under pure cornering condition is identified and the tire lateral force is modeled;

[0008] Step 3: Establish a neural network model. The UniTire tire model calculates the tire lateral force according to the tire vertical force and sideslip angle output by the neural network model. The error between the tire lateral force calculated by the UniTire tire model and the test value is integrated into the neural network loss function. The tire sideslip characteristics output by the neural network training are constrained to obtain a neural network model embedded with tire model constraints.

[0009] Step 4, extracting the test data of the PVDF strain sensor in the tire contact patch, taking the test data of the PVDF strain sensor in the tire contact patch as input and taking the intelligent tire cornering characteristics as output, training a neural network model embedded with tire model constraints;

[0010] Step 5: Use the trained neural network model to estimate the cornering characteristics of the smart tire under pure cornering conditions based on the PVDF strain sensor data in the tire rolling contact patch.

[0011] Preferably, in step 1, the six-component tire test machine uses an MTS tire test bench.

[0012] Preferably, the following test conditions are set for the tire mechanical properties test under pure side slip conditions in step 1:

[0013] 1) The tire pressure is set to 250 kPa;

[0014] 2) The tire roll angle is set to 0°;

[0015] 3) The test speeds are set at 30km / h, 60km / h and 90km / h respectively;

[0016] 4) The tire vertical load is set to 2000N, 4000N and 6000N respectively;

[0017] 5) The tire slip angle adopts a quasi-steady-state sweeping method, sweeping from the starting point 0° to 10° at a rate of 3° / s, then sweeping to -10° at the same rate, and finally returning to the 0° slip angle.

[0018] Preferably, the tire model in step 2 adopts the UniTire steady-state tire model. The lateral force modeling of the UniTire steady-state tire model under pure side slip conditions is as follows:

[0019] Normalized vertical load F zn The expression is:

[0020]

[0021] In the formula, F z is the vertical load of the tire; F z0 is the nominal tire load.

[0022] Lateral slip rate S y The expression is:

[0023] S y = -tan(α) (2)

[0024] Where α is the tire slip angle.

[0025] Tire footprint update speed V r The expression is:

[0026] V r =ΩR e (3)

[0027] Where, Ω is the angular velocity of the tire; R e is the effective rolling radius of the tire.

[0028] Blot lateral sliding speed V sy The expression is:

[0029] V sy =-S y ·V r (4)

[0030] In the formula, S y is the tire lateral slip rate; V r Update speed for tire tracks.

[0031] Pure cornering stiffness of tire The expression is:

[0032]

[0033] In the formula, F z is the vertical load of the tire; F zn is the normalized vertical load of the tire; s 11 、s 12 、s 13 is the tire cornering stiffness parameter.

[0034] Tire lateral curvature factor E y The expression is:

[0035]

[0036] In the formula, F zn is the normalized vertical load of the tire; s 21 、s 22 is the tire curvature factor parameter.

[0037] Tire lateral static friction coefficient μ y0 The expression is:

[0038]

[0039] In the formula, F zn is the normalized vertical load of the tire; s 31 、s 32 、s 33 is the friction coefficient parameter of the tire model.

[0040] Tire lateral dynamic friction coefficient μ y The expression is:

[0041] dμ y =s 34 ·μ y0 (8)

[0042]

[0043] In the formula, μ y0 V is the lateral static friction coefficient of the tire; sy is the lateral sliding velocity of the imprint; s 34 、s 35 is the friction coefficient parameter of the tire model.

[0044] Tire lateral dimensionless slip φ y The expression is:

[0045]

[0046] In the formula, is the pure cornering stiffness of the tire; S y is the tire lateral slip rate; μ y0 F is the lateral static friction coefficient of the tire; z is the vertical load on the tire.

[0047] Tire dimensionless lateral force The expression is:

[0048]

[0049] In the formula, φ y is the dimensionless lateral slip rate of the tire; E y is the tire lateral curvature factor.

[0050] Tire lateral force F under pure cornering condition yα The expression is:

[0051]

[0052] In the formula, is the dimensionless lateral force of the tire; φ y is the dimensionless lateral slip rate of the tire; μ y F is the lateral dynamic friction coefficient of the tire; z is the vertical load on the tire.

[0053] Preferably, the neural network model in step three adopts a BP neural network model, including an input layer, two hidden layers and an output layer, and the loss function adopts a mean square error function.

[0054] Preferably, the intelligent tire slip characteristic data output by the neural network model training in step 4 includes tire vertical force, lateral force and slip angle.

[0055] Preferably, in step five, a neural network model embedded with tire model constraints estimates the cornering characteristics of the smart tire under pure cornering conditions, including tire vertical force, lateral force and sideslip angle.

[0056] Beneficial effects of the present invention:

[0057] 1. The present invention discloses a method for estimating the sideslip characteristics of an intelligent tire under pure sideslip conditions, which improves the estimation accuracy of the vertical force, lateral force and sideslip angle of the intelligent tire under pure sideslip conditions.

[0058] 2. The present invention embeds the UniTire steady-state tire model into the loss function of the neural network model, thereby improving the training efficiency of the neural network model.

[0059] 3. The present invention embeds the UniTire steady-state tire model into the loss function of the neural network model, constrains the tire side slip characteristics output by the neural network, and improves the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0061] Figure 1 The flowchart of the intelligent tire side slip characteristic estimation method based on tire model embedded in neural network is shown.

[0062] Figure 2 An example of comparing the calculated tire lateral force using the UniTire tire model and the actual tire test value under pure cornering conditions is shown.

[0063] Figure 3 An example of the comparison between the tire vertical force estimate of the neural network embedded with the UniTire tire model constraints and the actual value of the tire test under pure cornering conditions is shown.

[0064] Figure 4An example of comparing the tire lateral force estimate using a neural network embedded with UniTire tire model constraints with the actual tire test value under pure cornering conditions is shown.

[0065] Figure 5 An example of comparing the tire slip angle estimation value and the actual value of the tire test under pure slip conditions using a neural network embedded with UniTire tire model constraints DETAILED DESCRIPTION

[0066] The present invention will be further described in detail below in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method is given, but the protection scope of the present invention is not limited to the following embodiments.

[0067] Taking a 235 / 45R18 radial tire for a passenger car as an example, a PVDF strain sensor is arranged longitudinally on the inner wall of the tire. The tire is installed on an MTS tire six-component force test bench to conduct a tire mechanical property test under pure cornering conditions. The tire mechanical property test data under pure cornering conditions and the PVDF strain sensor test data are obtained. The test conditions are set as follows:

[0068] 1) The tire pressure is set to 250 kPa;

[0069] 2) The tire roll angle is set to 0°;

[0070] 3) The test speeds are set at 30km / h, 60km / h and 90km / h respectively;

[0071] 4) The tire vertical load is set to 2000N, 4000N and 6000N respectively;

[0072] 5) The tire slip angle adopts a quasi-steady-state sweeping method, sweeping from the starting point 0° to 10° at a rate of 3° / s, then sweeping to -10° at the same rate, and finally returning to the 0° slip angle.

[0073] Using the test data of tire vertical load, footprint update speed, slip angle and lateral force under pure cornering conditions, the UniTire steady-state tire model under pure cornering conditions was identified and the tire lateral force was modeled. The identified UniTire steady-state tire model parameters are shown in the following table.

[0074] <![CDATA[s 11 ]]> 29.018 <![CDATA[s 12 ]]> 3.764 <![CDATA[s 13 ]]> 0.953 <![CDATA[s 21 ]]> 11.231 <![CDATA[s 22 ]]> 0.983 <![CDATA[s 31 ]]> 1.219 <![CDATA[s 32 ]]> -61.759 <![CDATA[s 33 ]]> 388.406 <![CDATA[s 34 ]]> 0.090 <![CDATA[s 35 ]]> 1.549

[0075] A BP neural network model was established, which included an input layer, two hidden layers, and an output layer. The input layer was the PVDF strain sensor data and test speed in the tire contact patch. The first hidden layer had 64 neurons, and the activation function was the sigmoid function. The second hidden layer had 32 neurons, and the activation function was the sigmoid function. The output layer was the tire vertical force, lateral force, and sideslip angle. The UniTire tire model calculated the tire lateral force based on the tire vertical force and sideslip angle output by the neural network model. The error between the tire lateral force calculated by the UniTire tire model and the test value was incorporated into the neural network loss function, and the tire sideslip characteristics output by the constrained neural network training were obtained to obtain a neural network model embedded with tire model constraints.

[0076] The PVDF strain sensor data in the tire contact patch is extracted, and the PVDF strain sensor test data in the tire contact patch is used as input, and the intelligent tire vertical force, lateral force and sideslip angle are used as output to train a neural network model embedded with tire model constraints.

[0077] The trained neural network model is used to estimate the cornering characteristics of the smart tire under pure cornering conditions, including tire vertical force, lateral force, and slip angle, based on the data from the PVDF strain sensor in the tire contact patch.

[0078] The test results of this embodiment are as follows:

[0079] Figure 2 An example of comparing the calculated tire lateral force using the UniTire tire model and the actual tire test value under pure cornering conditions is shown; Figure 3 An example of the comparison between the tire vertical force estimation value and the actual value of the tire test under pure cornering condition by the neural network embedded with UniTire tire model constraints is shown; Figure 4 An example of comparing the tire lateral force estimation value and the actual value of the tire test under pure cornering conditions using a neural network embedded with UniTire tire model constraints is shown; Figure 5 An example of comparison between the tire slip angle estimation value and the actual value of the tire test under pure slip conditions using a neural network embedded with UniTire tire model constraints is shown.

[0080] The above is a description of the embodiments of the present invention. Through the above description of the disclosed embodiments, professionals and technicians in the field can implement or use the present invention. Various modifications to the embodiments will be apparent to professionals and technicians in the field. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest range consistent with the principles and novelties disclosed herein.

Claims

1. An intelligent tire cornering characteristic estimation method based on tire model embedded in neural network, characterized in that: The following steps are involved: Step 1: Arrange a PVDF strain sensor on the inner wall of the tire, install the tire on a tire six-component force testing machine, set test conditions according to test requirements, conduct a tire mechanical property test under a pure cornering condition, and obtain tire mechanical property test data and PVDF strain sensor test data under the pure cornering condition; Step 2: Using the test data of tire vertical load, footprint update speed, slip angle and lateral force under pure cornering condition, the UniTire steady-state tire model under pure cornering condition is identified and the tire lateral force is modeled; Step 3: Establish a neural network model. The UniTire steady-state tire model calculates the tire lateral force according to the tire vertical force and sideslip angle output by the neural network model. The error between the tire lateral force calculated by the UniTire steady-state tire model and the test value is integrated into the neural network loss function. The tire sideslip characteristics output by the neural network training are constrained to obtain a neural network model embedded with tire model constraints. Step 4, extracting the test data of the PVDF strain sensor in the tire rolling contact patch, taking the test data of the PVDF strain sensor in the tire rolling contact patch as input and taking the intelligent tire cornering characteristics as output, training a neural network model embedded with tire model constraints; Step 5: Use the trained neural network model to estimate the cornering characteristics of the smart tire under pure cornering conditions based on the PVDF strain sensor data in the tire rolling contact patch.

2. The intelligent tire cornering characteristic estimation method based on tire model embedded in neural network according to claim 1 is characterized by: Step 1: Test conditions for tire mechanical properties under pure cornering conditions are as follows: 1) The tire pressure is set to 250 kPa; 2) The tire roll angle is set to 0°; 3) The test speeds are set at 30km / h, 60km / h and 90km / h respectively; 4) The tire vertical loads are set to 2000N, 4000N and 6000N respectively; 5) The tire slip angle adopts a quasi-steady-state sweeping method, sweeping from the starting point 0° to 10° at a rate of 3° / s, then sweeping to -10° at the same rate, and finally returning to the 0° slip angle.

3. The intelligent tire cornering characteristic estimation method based on tire model embedded in neural network according to claim 1, characterized in that: In step three, the neural network model adopts the BP neural network model, which includes an input layer, two hidden layers and an output layer, and the loss function adopts the mean square error function.

4. The intelligent tire cornering characteristic estimation method based on tire model embedded in neural network according to claim 1, characterized in that: Step 3: Incorporate the error between the tire lateral force calculated by the UniTire steady-state tire model and the test value into the neural network loss function to constrain the tire lateral characteristics output by the neural network model training.

5. The intelligent tire cornering characteristic estimation method based on tire model embedded in neural network according to claim 1, characterized in that: Step 4: The intelligent tire slip characteristic data output by the neural network model training includes tire vertical force, lateral force and slip angle.

6. The intelligent tire cornering characteristic estimation method based on tire model embedded in neural network according to claim 1, characterized in that: Step 5: The neural network model embedded with tire model constraints estimates the cornering characteristics of the smart tire under pure cornering conditions, including tire vertical force, lateral force and sideslip angle.

7. Application of the intelligent tire cornering characteristic estimation method based on tire model embedded in neural network according to any one of claims 1 to 6 in the estimation of vertical force, lateral force and sideslip angle of the intelligent tire under pure cornering conditions.

Images