A method for estimating the degree of tire wear based on in-utero strain

Abstract

The application discloses a tire wear degree estimation method based on in-utero strain, modeling is carried out through a finite element analysis software, three-dimensional tire models of different wear degrees are obtained, strain data of the three-dimensional tire models under different working conditions (air pressure, load and speed) are extracted and a curve graph is drawn to extract features, correlation analysis is carried out on the extracted features and tire wear, features with relatively large correlation with tire wear are selected as regression analysis features, air pressure, load, speed and the regression analysis features are taken as input quantities, and tire wear degree is taken as output quantity to train and construct a tire wear degree estimation model. The established model is used for tire wear degree estimation, and the effectiveness of the model is verified by using actual test data. The application makes the tire wear degree estimation result more accurate and can better adapt to tire wear degree estimation under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of tire wear estimation technology, and more specifically to a method for estimating tire wear based on in-tire strain. Background Technology

[0002] As the only part of a car in contact with the road, tires' performance directly impacts driving safety, stability, and comfort. Tire wear is a crucial indicator of tire performance, altering tread shape, stiffness, and contact characteristics. This leads to reduced water drainage and grip, affecting driving safety. Estimating tire wear levels can promptly remind the driver to replace tires when they reach a certain wear threshold, preventing excessive wear, improving vehicle grip, and ultimately enhancing driving safety and stability.

[0003] With the rapid development of sensor technology, more and more scholars at home and abroad are using sensors to acquire data related to tire wear and combining them with corresponding algorithms to estimate the degree of tire wear. Based on different measurement principles, these sensors mainly include acceleration sensors, strain sensors, and optical sensors.

[0004] In the invention patent application "A method and system for monitoring tire wear" with application number 202110174140.0, triaxial acceleration sensors and tire pressure sensors attached to the inner wall of the tire collect triaxial acceleration signals and tire pressure signals during the tire's operation under different wear levels. Feature parameters are analyzed and extracted, and a wear prediction model is constructed with wear level as the target and feature parameters as independent variables. Furthermore, by real-time acquisition of triaxial acceleration and tire pressure signals during tire operation, corresponding feature parameters are extracted, and the wear prediction model is used to estimate the real-time tire wear level. However, the number of extracted features is relatively small, consisting mainly of the valley areas in the characteristic waveforms of tire pressure, speed, and z-axis acceleration, resulting in reduced accuracy in tire wear level estimation.

[0005] In the invention patent application No. 202080073649.X, entitled "Tire Wear Estimation System, Tire Wear Estimation Program, and Tire Wear Estimation Method," a strain sensor installed inside the pneumatic tire is used to acquire the strain signal generated during tire operation. Based on the strain signal output at the moment the contact area of ​​the tread corresponding to the strain sensor is pushed off the road surface and the reference value of the strain signal, the wear state of the pneumatic tire's tread is estimated. However, the extracted strain signal only has one type of time-series waveform, and the feature type is relatively simple, resulting in insufficient adaptability to estimate tire wear under complex working conditions.

[0006] It is evident that current methods for estimating tire wear based on sensor-collected data suffer from limitations such as a limited number and type of extracted features, making it difficult to acquire a large amount of data related to tire wear. Consequently, these methods result in low accuracy in estimating tire wear and poor adaptability to complex tire operating conditions. Summary of the Invention

[0007] The present invention aims to address the problems of low accuracy and poor adaptability of current methods for estimating tire wear based on sensor-collected data, and provides a method for estimating tire wear based on in-tire strain.

[0008] To solve the above problems, the present invention is achieved through the following technical solution:

[0009] A method for estimating tire wear based on in-tire strain includes the following steps:

[0010] Step 1: Use finite element analysis software to model the tire and obtain three-dimensional tire models with different wear levels;

[0011] Step 2: Using the established three-dimensional tire models with different wear levels, simulate different working conditions with different air pressures, loads, and speeds.

[0012] Step 3: For each working condition, extract the strain data in the axial, circumferential and radial directions within a circle of the centerline of the inner liner of the three-dimensional tire model, and plot the corresponding strain curves.

[0013] Step 4: For each working condition, select the circumferential strain curve and the radial strain curve from the axial strain curve, circumferential strain curve, and radial strain curve for feature extraction.

[0014] The features extracted from the circumferential strain curve include the small peak height, small peak width, small peak area, large peak height, large peak width, and large peak area of ​​circumferential strain.

[0015] The features extracted from the radial strain curve include the height of the small radial strain valley, the width of the small radial strain valley, the area of ​​the small radial strain valley, the height of the large radial strain valley, the width of the large radial strain valley, and the area of ​​the large radial strain valley.

[0016] Step 5: First, calculate the correlation coefficient between each extracted feature and tire wear under each working condition. Then, calculate the average value of the correlation coefficient between the same feature and tire wear under all working conditions. Finally, select the feature with the larger absolute value of the average correlation coefficient as the regression analysis feature.

[0017] Step 6: Using the regression analysis features of the three-dimensional tire model and the tire's operating conditions (i.e., air pressure, load, and speed) as inputs and the wear level of the three-dimensional tire model as output, perform machine learning on the support vector regression model to obtain the tire wear level estimation model.

[0018] Step 7: Use the strain sensor installed inside the tire to collect strain data of the tire rotating one revolution and plot the corresponding strain curve to extract the regression analysis characteristics of the tire. At the same time, use the air pressure sensor and load sensor installed inside the tire, as well as the speed sensor installed in the vehicle, to collect the air pressure, load, and speed of the tire. Then, input the regression analysis characteristics, air pressure, load, and speed of the tire into the tire wear estimation model to estimate the wear degree of the tire.

[0019] The wear level in step 1 above includes zero wear level, the air pressure condition in step 2 above needs to include rated air pressure condition, and the load condition needs to include rated load condition.

[0020] The speed in step 2 above is the speed at which the tire reaches a steady-state free rolling state.

[0021] In step 4 above, the specific process of feature extraction for the circumferential strain curve with sampling points on the horizontal axis and circumferential strain on the vertical axis is as follows:

[0022] Calculate the mean circumferential strain at all sampling points on the circumferential strain curve, and plot a mean circumferential strain line on the circumferential strain curve;

[0023] The difference in the ordinate of the mean circumferential strain line and the maximum peak point of the circumferential strain curve is used as the characteristic of the small peak height of circumferential strain.

[0024] The difference in the abscissa of the two intersection points of the mean circumferential strain line and the circumferential strain curve below the maximum peak point is used as the characteristic of small peak width in circumferential strain.

[0025] The peak area of ​​the circumferential strain curve between the two intersection points of the mean circumferential strain line and the circumferential strain curve below the maximum peak point is used as the small peak area characteristic of circumferential strain.

[0026] The difference in the ordinate between the maximum peak point and the minimum valley point of the circumferential strain curve is used as the characteristic of the large peak height of the circumferential strain.

[0027] The difference in the horizontal coordinate between the two small valley points of the circumferential strain curve below the mean circumferential strain line is taken as the characteristic of large peak width of circumferential strain.

[0028] The peak area of ​​the circumferential strain curve between two small valley points below the mean circumferential strain curve is taken as the characteristic of the large peak area of ​​circumferential strain.

[0029] In step 4 above, the specific process of feature extraction for the radial strain curve with sampling points on the horizontal axis and radial strain on the vertical axis is as follows:

[0030] Calculate the mean radial strain at all sampling points on the radial strain curve, and plot a mean radial strain line on the radial strain curve;

[0031] The difference in the ordinate of the minimum valley point between the mean radial strain line and the radial strain curve is used as the characteristic of the small valley height of radial strain.

[0032] The difference in the abscissa of the two intersection points of the radial strain mean line and the radial strain curve above the minimum valley point is used as the characteristic of the small valley width of radial strain.

[0033] The valley area of ​​the radial strain curve between the two intersection points above the minimum valley point of the mean radial strain line and the radial strain curve is used as the characteristic of the small valley area of ​​radial strain.

[0034] The difference in the ordinate between the maximum peak point and the minimum valley point of the radial strain curve is used as the characteristic of the radial strain valley height.

[0035] The difference in the horizontal coordinate between the two large peak points of the radial strain curve above the mean radial strain line is used as the characteristic of the large valley width of radial strain.

[0036] The valley area of ​​the radial strain curve between the two large peak points above the mean radial strain line is used as the characteristic of the large valley area of ​​radial strain.

[0037] In step 5 above, the correlation coefficient is the Pearson correlation coefficient.

[0038] In step 6 above, the dataset consisting of the regression analysis features of the three-dimensional tire model, the tire's working conditions (i.e., air pressure, load and speed), and the wear degree of the three-dimensional tire model needs to be expanded to increase the amount of data before machine learning is performed on the support vector regression model to improve the accuracy of the regression prediction results and reduce the error.

[0039] Compared with existing technologies, this invention first uses finite element method (FEM) software to model tires under various working conditions, thereby extracting the necessary strain data and significantly reducing the time and financial costs of actual experiments. Then, based on the strain data obtained from the FEM model, multiple features on the circumferential and radial strain curves are extracted, and these features are selected based on correlation analysis. Finally, a tire wear estimation model is constructed using machine learning methods to describe the complex nonlinear relationship between different features and tire wear. During prediction and estimation, the tire pressure, load, speed, and features highly correlated with tire wear are used as inputs. The established model accurately estimates tire wear, solving the problems of low accuracy in tire wear estimation and inability to better adapt to complex tire working conditions caused by a limited number of input features. Attached Figure Description

[0040] Figure 1 This is a flowchart of a method for estimating tire wear based on in-tire strain.

[0041] Figure 2 Two-dimensional tire cross-section models with four different wear levels: (a) 0 mm wear, (b) 2 mm wear, (c) 4 mm wear, and (d) 6 mm wear.

[0042] Figure 3 To create a circumferential path along the centerline of the tire inner liner.

[0043] Figure 4 Features extracted from the circumferential strain curve.

[0044] Figure 5 These are features extracted from the radial strain curve.

[0045] Figure 6 The results of the training set for the SVR algorithm to estimate tire wear.

[0046] Figure 7 Results of the test set for estimating tire wear using the SVR algorithm.

[0047] Figure 8 This is the result of estimating tire wear using experimental test data. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0049] A method for estimating tire wear based on in-tire strain, such as... Figure 1 As shown, the specific steps include the following:

[0050] Step 1: Use finite element analysis software to model the tire, obtaining n three-dimensional tire models with different wear levels. Here, n is a set value, n≥4.

[0051] Finite Element Analysis (FEA) uses mathematical approximations to simulate real physical systems. By using simple, interacting elements, or cells, a finite number of unknowns can approximate an infinite number of unknowns in a real system. FEA solves a complex problem by substituting a simpler one. It considers the solution domain as composed of many small, interconnected subdomains called finite elements. For each element, a suitable approximate solution is assumed, and then the overall conditions satisfying the solution domain are derived to obtain the solution to the problem. This solution is not an exact solution, but an approximate one, because the real problem has been replaced by a simpler one. Since most real-world problems are difficult to solve exactly, and finite element analysis offers high computational accuracy and can adapt to various complex shapes, it has become an effective engineering analysis tool.

[0052] Step 1.1: Use finite element software to create a two-dimensional tire cross-section model with 0mm wear (no wear), and then translate the tread portion inward by the same or different distances each time to obtain n two-dimensional tire cross-section models with different wear levels.

[0053] In this embodiment, a two-dimensional tire cross-sectional model with 0mm wear was established using Abaqus finite element software. The tread portion was modified by tracing the tread edge line, translating the edge line inward to obtain a new tread edge contour, and removing excess tread material. This method ensures a smooth and natural tread surface after wear. In this embodiment, n=4. The tread portion was translated inward by 2mm each time, and this translation was repeated three times, ultimately resulting in four two-dimensional tire cross-sectional models with different wear levels: 0mm, 2mm, 4mm, and 6mm.

[0054] Step 1.2: Mesh the two-dimensional tire cross-section models with n different wear levels and check the mesh quality, then submit the analysis job.

[0055] In this embodiment, meshing is performed on two-dimensional tire cross-section models with four different wear levels. The number of meshes significantly impacts the computation time and results of tire analysis. More meshes improve accuracy but require more computation time; conversely, fewer meshes save computation time but may reduce accuracy. Therefore, an appropriate number of meshes is needed to balance accuracy and computation time. After meshing, the mesh quality is checked to ensure no erroneous elements are displayed. The job is then submitted for analysis, and the results are viewed through the visualization module after completion. The results of the four two-dimensional tire cross-section models with different wear levels are shown below. Figure 2 As shown.

[0056] Step 1.3: Generate n different three-dimensional tire models with varying degrees of wear by rotating the two-dimensional tire cross-section model around its central axis, and generate the road surface that contacts the three-dimensional tire model.

[0057] In this embodiment, four two-dimensional tire cross-section models with different wear levels are rotated around the central axis to obtain four three-dimensional tire models with different wear levels, and the road surface type in contact with the three-dimensional tire model is set as a rigid plane.

[0058] Step 2: Using the established 3D tire models with n different wear levels, simulate the conditions under m different air pressures, k different loads, and t different speeds. Finally, obtain tire simulation data for the n wear levels under m×k×t conditions. Here, m, k, and t are set values, and m, k, and t ≥ 4.

[0059] Step 2.1: Set m different air pressures for n different wear levels of three-dimensional tire models.

[0060] The tire pressure is simulated by applying a pressure load to the inner surface of the tire. In this embodiment, m = 4. Based on the actual range of car tire pressure, the tire pressure is set to four levels: 0.211 MPa, 0.241 MPa, 0.271 MPa, and 0.301 MPa, with the rated tire pressure being 0.241 MPa.

[0061] Step 2.2: Set k different loads for n different wear levels of the three-dimensional tire models.

[0062] When applying the load, the road surface is moved towards the tire to establish contact between them. Once contact is established, a vertical load is applied to the road surface, thus simulating the load. In this embodiment, k = 4. The load is set to four values: 1100N, 2200N, 3300N, and 4400N, with the rated load being 3300N.

[0063] Step 2.3: Set t different speeds for n different wear levels of the three-dimensional tire models.

[0064] The speed is the speed at which the tire reaches a steady-state free-rolling state, which is the state where the torque acting on the tire is zero, distinct from the tire's braking and driving states. In this embodiment, t = 4. Four speeds are set: 20 km / h, 40 km / h, 60 km / h, and 80 km / h.

[0065] For the same linear velocity ν, the tire rotates at different angular velocities ω, corresponding to three motion conditions: braking, free rolling, and driving. Normally, the combination of the angular velocity ω0 and the linear velocity ν during free rolling is unknown beforehand. However, steady-state rolling analysis requires specifying both the free rolling angular velocity ω0 and the linear velocity ν. Therefore, an indirect method is needed to obtain the free rolling angular velocity ω0. The specific steps are as follows:

[0066] Step 2.3.1: First, perform tire inflation and static loading analysis to obtain the tire inflation radius R1 and static loading radius R2.

[0067] Step 2.3.2: Divide the linear velocity ν by the tire's inflation radius R1 to obtain the angular velocity ω1. At this time, the tire is in a braking state.

[0068] Step 2.3.3: Divide the linear velocity ν by the static loading radius R2 of the tire to obtain the angular velocity ω2. At this time, the tire is in driving state.

[0069] Step 2.3.4: Let the free rolling radius of the tire be R0, which is smaller than the inflation radius R1 but larger than the static loading radius R2. Therefore, the free rolling angular velocity ω0 is within the range determined by the following formula:

[0070]

[0071] By analyzing the angular velocities between ω1 and ω2, the relationship between torque and angular velocity is obtained. The angular velocity corresponding to zero torque is the free rolling angular velocity ω0.

[0072] Step 3: For each working condition, extract the strain data within one circumference (one cycle) on the center line of the inner liner of the three-dimensional tire model (extract strain data in the axial, circumferential and radial directions), and plot the corresponding strain curve.

[0073] The process for generating the strain curve for each working condition is as follows:

[0074] Step 3.1: Create a circular path on the 3D tire model along the centerline of the tire's inner liner, such as... Figure 3 As shown.

[0075] Since the trajectory of the circular path is the same as that of a strain sensor installed on the center line of the tire's inner liner as the tire moves through one revolution, extracting tire strain data along this path can be used to simulate the process of a strain sensor collecting tire strain data. The circular path is created as follows: select the point on the tire's inner liner center line that is farthest from the road surface (the vertex) as the starting point, and then sequentially select points on the tire's inner liner center line in a counter-clockwise direction. Finally, the selected endpoint coincides with the starting point, forming a complete circular path.

[0076] Step 3.2: Convert the rectangular coordinate system to a cylindrical coordinate system, and extract the axial, circumferential and radial strain data of the tire in the cylindrical coordinate system.

[0077] Because a tire is a cylinder, it's necessary to convert from a Cartesian coordinate system to a cylindrical coordinate system to extract the axial, circumferential, and radial strains of the tire. The axial, circumferential, and radial directions together constitute the three orthogonal directions of the cylindrical coordinate system: the axial direction is along the axis of rotation of the cylinder, the circumferential direction is along the circumference of the cylinder's cross-section, and the radial direction is along the radius of the cylinder's cross-section. The transformation relationship between the Cartesian and cylindrical coordinate systems is as follows:

[0078]

[0079] In the formula, r∈[0,+∞), z∈R. x, y, z are three coordinate variables in a rectangular coordinate system, r, ... z represents the three coordinate variables in a cylindrical coordinate system.

[0080] Step 3.3: Extract strain data of the three-dimensional tire model under different working conditions along the center line of the inner liner under the circumferential path created in step 3.1, and draw strain curves in the axial, circumferential and radial directions.

[0081] The tire carcass is simulated using shell elements in finite element analysis software. Since a tire is a typical nonlinear mechanical system, the output is the true strain LE under geometrically nonlinear conditions. According to the mapping relationship, the direction of shell element 1 is the z-direction in cylindrical coordinates, i.e., the axial direction, so the axial strain is LE11; the direction of shell element 2 is the z-direction in cylindrical coordinates... The direction, i.e., the circumferential direction, is the circumferential strain LE22; the direction of shell element 3 is the r-direction in cylindrical coordinates, i.e., the radial direction, so the radial strain is LE33.

[0082] Step 4: Compare and analyze the strain curves in the axial, circumferential, and radial directions, and select the strain curves in the circumferential and radial directions for feature extraction.

[0083] By analyzing the curves of axial strain, circumferential strain, and radial strain of tires under different working conditions, it was found that the axial strain curve did not show a good regularity as the working conditions changed, while the circumferential strain curve and radial strain curve showed a better regularity. Therefore, the circumferential strain curve and radial strain curve were selected to extract features.

[0084] Step 4.1: Extract features from the circumferential strain curve (the horizontal axis of the circumferential strain curve represents the sampling points, and the vertical axis represents the circumferential strain), such as... Figure 4 As shown, the specific process is as follows:

[0085] Calculate the mean circumferential strain at all sampling points on the circumferential strain curve, and plot a mean circumferential strain line on the circumferential strain curve;

[0086] The difference in the ordinate between the mean circumferential strain line and the maximum peak point of the circumferential strain curve is used as the characteristic of the small peak height of circumferential strain, denoted by LE22-h.

[0087] The difference in the abscissa of the two intersection points of the mean circumferential strain line and the circumferential strain curve below the maximum peak point is used as the characteristic of the small peak width of circumferential strain, denoted by LE22-w.

[0088] The peak area of ​​the circumferential strain curve between the two intersection points of the mean circumferential strain line and the circumferential strain curve below the maximum peak point is used as the characteristic of the small peak area of ​​circumferential strain, denoted by LE22-s.

[0089] The difference in the ordinate between the maximum peak point and the minimum valley point of the circumferential strain curve is used as the characteristic of the large peak height of circumferential strain, denoted by LE22-H.

[0090] The difference in the horizontal coordinate between the two small valley points of the circumferential strain curve below the mean circumferential strain line is taken as the characteristic of large peak width of circumferential strain, and is represented by LE22-W.

[0091] The peak area of ​​the circumferential strain curve between two small valley points below the mean circumferential strain curve is taken as the characteristic of the large peak area of ​​circumferential strain, and is denoted by LE22-S.

[0092] Step 4.2: Extract features from the radial strain curve (the horizontal axis of the radial strain curve represents the sampling points, and the vertical axis represents the radial strain), such as... Figure 5 As shown, the specific process is as follows:

[0093] Calculate the mean radial strain at all sampling points on the radial strain curve, and plot a mean radial strain line on the radial strain curve;

[0094] The difference in the ordinate of the minimum valley point between the mean radial strain line and the radial strain curve is used as the characteristic of the small valley height of radial strain, denoted by LE33-h;

[0095] The difference in the abscissa of the two intersection points of the radial strain mean line and the radial strain curve above the minimum valley point is used as the characteristic of the radial strain small valley width, denoted by LE33-w;

[0096] The valley area of ​​the radial strain curve between the two intersection points above the minimum valley point of the mean radial strain line and the radial strain curve is used as the characteristic of the small valley area of ​​radial strain, denoted by LE33-s.

[0097] The difference in the ordinate between the maximum peak point and the minimum valley point of the radial strain curve is used as the characteristic of the radial strain valley height, denoted by LE33-H.

[0098] The difference in the abscissa between the two large peak points of the radial strain curve above the mean radial strain line is taken as the characteristic of the large valley width of radial strain, and is represented by LE33-W.

[0099] The valley area of ​​the radial strain curve between the two large peak points above the mean radial strain line is used as the characteristic of the large valley area of ​​radial strain, denoted by LE33-S.

[0100] Step 5: First, calculate the correlation coefficient between each extracted feature and tire wear under each working condition. Then, calculate the average value of the correlation coefficient between the same feature and tire wear under all working conditions. Finally, select the feature with the larger absolute value of the average correlation coefficient as the regression analysis feature for subsequent regression analysis.

[0101] In statistics, the Pearson correlation coefficient is widely used to measure the degree of correlation between two variables, with a value between -1 and 1. A correlation coefficient greater than 0 indicates a positive correlation, meaning that the larger the value of one variable, the larger the value of the other. A correlation coefficient less than 0 indicates a negative correlation, meaning that the larger the value of one variable, the smaller the value of the other. The closer the absolute value of the correlation coefficient is to 1, the stronger the correlation between the two variables; the closer the absolute value is to 0, the weaker the correlation. Therefore, this invention uses Pearson correlation analysis. Let variable X be the feature value extracted in step 4 above, and variable Y be the tire wear degree. The Pearson correlation coefficient between the two variables can be calculated using the following formula:

[0102]

[0103] In the formula, N represents the number of variable values. In this embodiment, there are four types of tire wear, so N = 4.

[0104] First, under each operating condition, the correlation coefficient between each feature and tire wear is calculated, resulting in m×k×t sets of correlation coefficients for each feature and tire wear. Then, the correlation coefficients between the same feature and tire wear under all operating conditions are averaged; that is, the average of the m×k×t sets of correlation coefficients is calculated to determine the strength of the correlation between the feature and tire wear. Finally, after averaging the correlation coefficients, features whose absolute value of the average is greater than a preset value (e.g., 0.8) are selected for regression analysis.

[0105] In this embodiment, tire wear is used as a variable for correlation analysis with the extracted features. Tire wear levels under the same operating conditions (same tire pressure, load, and speed) are grouped together. Therefore, after correlation analysis between each feature and tire wear, 64 sets of correlation coefficients are obtained. The average of these 64 correlation coefficients is calculated, and the strength of the correlation between the feature and tire wear is determined based on the magnitude of the average. After averaging the correlation coefficients, the overall correlation coefficients between each feature and tire wear are obtained, as shown in the table below:

[0106] feature Correlation coefficient LE22-h 0.2579 LE22-w 0.8902 LE22-s 0.9377 LE22-H 0.4968 LE22-W 0.2053 LE22-S 0.6676 LE33-h -0.8213 LE33-w 0.3648 LE33-s -0.8992 LE33-H -0.8369 LE33-W 0.4355 LE33-S -0.9096

[0107] As can be seen from the table, the following six features have an absolute value of correlation coefficient greater than 0.8: LE22-w, LE22-s, LE33-h, LE33-s, LE33-H, and LE33-S. Therefore, these six features were selected for subsequent regression analysis.

[0108] Step 6: Using the regression analysis features of the 3D tire model, as well as the tire's operating conditions (air pressure, load, and speed) as inputs, and the wear level of the 3D tire model as the output, perform machine learning on the support vector regression model to obtain the tire wear level estimation model.

[0109] Step 6.1: The present invention uses the data (air pressure, load, speed, regression analysis characteristics and wear degree) obtained by the above finite element simulation to form a dataset.

[0110] The dataset consists of tire wear, tire pressure, load, speed, and six features that are highly correlated with wear. The first column contains tire wear data, and the second to tenth columns contain tire pressure, load, speed, and the six features that are highly correlated with wear (regression analysis features), respectively.

[0111] Step 6.2: Increase the amount of data in the dataset by using data augmentation.

[0112] To improve the accuracy of regression predictions and reduce errors, data augmentation is used to increase the amount of data. Since the extracted data is a one-dimensional array, linear interpolation is employed. Linear interpolation estimates the value of a point in the one-dimensional data sequence based on its left and right neighboring data points. It does not calculate the average of these two data points, but rather assigns weights based on their distances to each point. Taking two adjacent points of a certain feature in the dataset as an example, the principle of linear interpolation is explained as follows:

[0113] Assuming that the coordinates of two adjacent points are (x0, y0) and (x1, y1), forming a straight line, and the coordinates of an unknown point on this line are (x, y), then the following formula can be obtained:

[0114]

[0115] In the formula, k is the slope of the line. If the value of x is known and the value of y is to be found, formula (4) is transformed to obtain the following formula for solving the value of y:

[0116]

[0117] Given the value of y, the method to find the value of x is the same as the method above, except that x and y are interchanged.

[0118] Without altering the trend of feature changes, uniform interpolation is performed using linear interpolation to expand the dataset to a multiple of u (u≥2). In this embodiment, u=4, meaning the original 256 rows of data are expanded to 1024 rows.

[0119] Step 6.3: Using the six features with high correlation to wear in the expanded dataset, along with tire pressure, load, and speed, as inputs and tire wear degree as output, perform machine learning on the support vector regression (SVR) model to construct a tire wear degree estimation model.

[0120] Support Vector Regression (SVR) is a machine learning algorithm derived from Support Vector Machines (SVM), and it represents a significant research direction within SVM. The core idea of ​​SVR is to find a hyperplane that minimizes the margin between samples, which is used to solve regression problems. Compared to traditional machine learning regression algorithms, SVR demonstrates significant advantages in preventing overlearning, computational speed, and result accuracy.

[0121] The Support Vector Regression (SVR) algorithm is used to estimate tire wear. Let the sample set A = {(x1,y1),(x2,y2),…,(x...}. p ,y p )}, where x iThe input vector, in this embodiment, consists of air pressure, load, speed, and six features highly correlated with wear (regression analysis features); y i The output vector represents the tire wear level in this embodiment; p is the number of samples. Let the regression model be:

[0122] f(x i ) = w T x i +b (6)

[0123] In the formula, w and b are the model parameters to be solved.

[0124] The optimization objective of SVR is:

[0125]

[0126] In the formula, ‖w‖ is the Euclidean norm, which represents the distance between two points (the magnitude of a vector).

[0127] Points located within the boundary satisfy the following condition:

[0128] |y i -(w T x i +b)|≤ε (8)

[0129] In the formula, ε is the allowable predicted value f(x) i ) and actual value y i The difference between the two values ​​indicates that the prediction is correct when the absolute value of the prediction error is less than or equal to ε, meaning the prediction error is within the interval [-ε, ε]. In other words, the sample data points must fall within the interval [f(x...]. i )-ε,f(x i Within the interval [ ) + ε]. In practice, it's often impossible to directly find a suitable ε that ensures most data falls within the interval. To allow more data to fall within the interval, a slack variable is introduced, allowing some data to be outside the interval, but this incurs a penalty term. The slack variable ξ is introduced... It is worth noting that for any sample data x i When it is inside or on the edge of the interval, ξ = 0; when it is above the interval, ξ > 0. When it is below the interval, ξ = 0. Introducing slack variables transforms the support vector regression prediction problem into a minimum optimization problem, yielding the following equation:

[0130]

[0131] In the formula, C is the penalty coefficient, which is set to 1000 in this embodiment.

[0132] Introducing the Lagrange multiplier λ≥0, μ≥0, The minimum optimization problem is transformed into solving the dual problem, resulting in the following equation:

[0133]

[0134] In the formula K(x) i ,x j ) is the kernel function. In this embodiment, the radial basis function is used, and its formula is:

[0135] K(x i ,x j )=exp(-γ||x i -x j || 2 (11)

[0136] In the formula, γ is the radial basis kernel function parameter, which is set to 1 in this embodiment.

[0137] Solving formula (10), we obtain the expression for the tire wear estimation model using support vector regression:

[0138]

[0139] This embodiment uses the root mean square error (RMSE) to evaluate prediction accuracy. The smaller the RMSE value, the more accurate the model's prediction. The formula is as follows:

[0140]

[0141] In the formula, y represents the estimated tire wear level in the i-th data set. i This represents the actual value of tire wear in the i-th data set.

[0142] The dataset is divided into two parts: 75% is randomly selected as the training set to train the model parameters; the remaining 25% is used as the test set to evaluate the model's performance. The results of the SVR algorithm on the training set for tire wear estimation are shown below. Figure 6 As shown, the results of the test set are as follows Figure 7 As shown.

[0143] Four types of physical tires with wear levels of 0mm, 2mm, 4mm, and 6mm were subjected to different air pressures, loads, and speeds. The air pressures were 0.21MPa, 0.24MPa, and 0.27MPa; the loads were 1100N, 2200N, and 3300N; and the speeds were 20km / h, 40km / h, and 60km / h. Data (of the same type as that extracted from finite element simulation) was collected using strain sensors installed inside the physical tires under these conditions. This data was then fed into a constructed tire wear estimation model to estimate tire wear and verify the model's effectiveness. The results of tire wear estimation using experimental test data are shown below. Figure 8 As shown.

[0144] Step 7: Use strain sensors installed inside the tire to collect strain data as the tire rotates one revolution, and plot the corresponding strain curves (circumferential strain curve and radial strain curve) to extract the regression analysis features of the tire. At the same time, use air pressure sensors and load sensors installed inside the tire to collect the air pressure and load of the tire, and use the speed sensor installed in the vehicle to obtain the speed signal. Then, input the regression analysis features, air pressure, load and speed of the tire into the tire wear estimation model to estimate the wear degree of the tire.

[0145] Strain sensors, air pressure sensors, and load sensors are simultaneously installed at the centerline of the inner liner of the tire under test, by attaching the sensors to the inner wall of the tire. It is important to ensure the sensors are securely attached to the inner wall to prevent them from falling off during tire movement. The strain sensor collects strain data in the axial, circumferential, and radial directions during one rotation of the tire (one cycle). Circumferential and radial strain data are selected, and corresponding strain curves are plotted. The regression analysis features determined in step 5 are extracted using the same method as in step 4. The air pressure sensor collects the air pressure of the tire under test during the current cycle. The load sensor collects the load of the tire under test during the current cycle. A speed sensor installed inside the vehicle collects the speed of the tire under test during the current cycle. The collected regression analysis features, air pressure, load, and speed of the tire under test during the current cycle are fed into the tire wear estimation model, which outputs its predicted tire wear level.

[0146] This invention discloses a method for estimating tire wear based on in-tire strain. Finite element analysis software is used to model three-dimensional tires with different wear levels, and different tire pressure, load, and speed conditions are set for each tire model with different wear levels. Axial, circumferential, and radial strain data are extracted from the centerline of the inner liner of the tire models under different conditions, and curves are plotted. After analyzing the curves, circumferential and radial strain curves are selected for feature extraction. Correlation analysis is performed between the extracted features and tire wear, and features with a high correlation to tire wear are selected for subsequent regression analysis. A tire wear estimation model is constructed using machine learning methods, with tire pressure, load, speed, and the selected features with a high correlation to tire wear as inputs, and tire wear as the output. The established model is used to estimate tire wear, and the effectiveness of the model is verified using actual test data. This invention uses multiple extracted features, tire pressure, load, and speed as model inputs, thereby making the tire wear estimation results more accurate and better adaptable to tire wear estimation under complex working conditions.

[0147] It should be noted that although the embodiments described above are illustrative, they are not intended to limit the invention. Therefore, the invention is not limited to the specific embodiments described above. Any other embodiments obtained by those skilled in the art under the guidance of this invention without departing from its principles are considered to be within the protection scope of this invention.

Claims

1. A method for estimating tire wear based on in-tire strain, characterized in that, The steps include the following: Step 1: Use finite element analysis software to model the tire and obtain three-dimensional tire models with different wear levels; Step 2: Using the established three-dimensional tire models with different wear levels, simulate different working conditions with different air pressures, loads, and speeds. Step 3: For each working condition, extract the strain data in the axial, circumferential and radial directions within a circle of the centerline of the inner liner of the three-dimensional tire model, and plot the corresponding strain curves. Step 4: For each working condition, select the circumferential strain curve and the radial strain curve from the axial strain curve, circumferential strain curve, and radial strain curve for feature extraction. The features extracted from the circumferential strain curve include the small peak height, small peak width, small peak area, large peak height, large peak width, and large peak area of ​​circumferential strain. The features extracted from the radial strain curve include the height of the small radial strain valley, the width of the small radial strain valley, the area of ​​the small radial strain valley, the height of the large radial strain valley, the width of the large radial strain valley, and the area of ​​the large radial strain valley. Step 5: First, calculate the correlation coefficient between each extracted feature and tire wear under each working condition. Then, calculate the average value of the correlation coefficient between the same feature and tire wear under all working conditions. Finally, select the feature with the larger absolute value of the average correlation coefficient as the regression analysis feature. Step 6: Using the regression analysis features of the three-dimensional tire model and the tire's operating conditions (i.e., air pressure, load, and speed) as inputs and the wear level of the three-dimensional tire model as output, perform machine learning on the support vector regression model to obtain the tire wear level estimation model. Step 7: Use the strain sensor installed inside the tire to collect strain data of the tire rotating one revolution and plot the corresponding strain curve to extract the regression analysis characteristics of the tire; at the same time, use the air pressure sensor and load sensor installed inside the tire, as well as the speed sensor installed in the vehicle, to collect the air pressure, load and speed of the tire. The regression analysis characteristics, air pressure, load, and speed of the tire under test are then fed into the tire wear estimation model to estimate the wear degree of the tire under test.

2. The method for estimating tire wear based on in-tire strain according to claim 1, characterized in that, The wear level in step 1 includes zero wear level, the air pressure condition in step 2 needs to include the rated air pressure condition, and the load condition needs to include the rated load condition.

3. The method for estimating tire wear based on in-tire strain according to claim 1, characterized in that, The speed in step 2 is the speed at which the tire reaches a steady-state free rolling state.

4. The method for estimating tire wear based on in-tire strain according to claim 1, characterized in that, In step 4, the specific process of feature extraction for the circumferential strain curve with sampling points on the horizontal axis and circumferential strain on the vertical axis is as follows: Calculate the mean circumferential strain at all sampling points on the circumferential strain curve, and plot a mean circumferential strain line on the circumferential strain curve; The difference in the ordinate of the mean circumferential strain line and the maximum peak point of the circumferential strain curve is used as the characteristic of the small peak height of circumferential strain. The difference in the abscissa of the two intersection points of the mean circumferential strain line and the circumferential strain curve below the maximum peak point is used as the characteristic of small peak width in circumferential strain. The peak area of ​​the circumferential strain curve between the two intersection points of the mean circumferential strain line and the circumferential strain curve below the maximum peak point is used as the small peak area characteristic of circumferential strain. The difference in the ordinate between the maximum peak point and the minimum valley point of the circumferential strain curve is used as the characteristic of the large peak height of the circumferential strain. The difference in the horizontal coordinate between the two small valley points of the circumferential strain curve below the mean circumferential strain line is taken as the characteristic of large peak width of circumferential strain. The peak area of ​​the circumferential strain curve between two small valley points below the mean circumferential strain curve is taken as the characteristic of the large peak area of ​​circumferential strain.

5. The method for estimating tire wear based on in-tire strain according to claim 1, characterized in that, In step 4, the specific process of feature extraction for the radial strain curve with sampling points on the horizontal axis and radial strain on the vertical axis is as follows: Calculate the mean radial strain at all sampling points on the radial strain curve, and plot a mean radial strain line on the radial strain curve; The difference in the ordinate of the minimum valley point between the mean radial strain line and the radial strain curve is used as the characteristic of the small valley height of radial strain. The difference in the abscissa of the two intersection points of the radial strain mean line and the radial strain curve above the minimum valley point is used as the characteristic of the small valley width of radial strain. The valley area of ​​the radial strain curve between the two intersection points above the minimum valley point of the mean radial strain line and the radial strain curve is used as the characteristic of the small valley area of ​​radial strain. The difference in the ordinate between the maximum peak point and the minimum valley point of the radial strain curve is used as the characteristic of the radial strain valley height. The difference in the horizontal coordinate between the two large peak points of the radial strain curve above the mean radial strain line is used as the characteristic of the large valley width of radial strain. The valley area of ​​the radial strain curve between the two large peak points above the mean radial strain line is used as the characteristic of the large valley area of ​​radial strain.

6. The method for estimating tire wear based on in-tire strain according to claim 1, characterized in that, In step 5, the correlation coefficient is the Pearson correlation coefficient.

7. The method for estimating tire wear based on in-tire strain according to claim 1, characterized in that, In step 6, the dataset consisting of the regression analysis features of the 3D tire model, the tire's working conditions (i.e., air pressure, load and speed), and the wear degree of the 3D tire model needs to be expanded to increase the amount of data before machine learning is applied to the support vector regression model.

Images