A tire wear prediction method and system based on artificial intelligence, a storage medium and a software product

Abstract

The present application relates to the technical field of tire performance prediction, and more particularly to a tire wear prediction method and system based on artificial intelligence, a storage medium and a software product. The present application obtains radial difference timing, tangential slip and friction work distribution through multi-working condition rolling finite element simulation, extracts radial difference statistics and frequency domain features in the steady state window; constructs a slip direction angle field from two-direction slip, identifies the slip direction demarcation line and extracts the number, position and stability features, while extracting the total amount of friction work and unevenness features. A physical model from working condition to radial difference and residual learning fusion mapping are established and physical consistency constraints are applied, and then the wear prediction model is trained with radial difference features, demarcation line features and friction work features, and the wear prediction, threshold early warning and recommended working condition are output. The scheme improves the cross-working condition prediction accuracy and interpretability, reduces the test cost, and is suitable for tire design and operation.

Description

Technical Field

[0001] This invention relates to the field of tire performance prediction technology, and in particular to a tire wear prediction method, system, storage medium, and software product based on artificial intelligence. Background Technology

[0002] As the only contact component between a vehicle and the road surface, the wear condition of tires directly affects the vehicle's braking distance, handling stability, fuel / energy consumption, and total lifecycle cost. In passenger cars, commercial vehicles, and even intelligent connected vehicles, tire wear is also closely related to fleet maintenance (replacement cycle, inventory strategy, and downtime risk). Traditional tire wear assessment mainly relies on road mileage tests, indoor wear test machines, and bench rolling tests, using statistical analysis of indicators such as tread depth decay, mass loss, or wear volume to reflect wear resistance. However, these methods generally suffer from long cycles, high costs, limited coverage of operating conditions, and significant impacts on repeatability due to environmental fluctuations. Furthermore, wear is a typical multi-factor coupled phenomenon, with factors such as load, speed, tire pressure, road friction coefficient, longitudinal / lateral slippage, and temperature field interacting with each other, making it difficult to extrapolate test conclusions obtained under single or limited operating conditions to complex real-world roads and diverse user conditions.

[0003] With the development of computing power and tire finite element simulation technology, simulation-based wear analysis has gradually become an important tool in the tire design stage. By constructing a constitutive model of the tire structure and materials, and applying loads, speeds, tire pressures, and contact boundary conditions, key field quantities such as contact pressure, shear stress, slip vector field, and friction work density can be output. Furthermore, wear mechanism models (such as engineering models based on friction work, slip work, or Arcard-like wear concepts) can be combined to estimate the wear distribution. The advantage of simulation methods lies in their ability to iterate rapidly during the design stage, cover more operating condition combinations, and obtain intermediate variables that are difficult to directly measure through real-vehicle sensors. However, their shortcomings are also quite obvious: on the one hand, wear model parameters (wear coefficient, temperature-sensitive terms, etc.) usually rely on experimental calibration, and their transferability is limited under different rubber compounds / structures / road conditions; on the other hand, the simulation output has high dimensionality and high spatiotemporal resolution. How to transform multi-source field quantities into learnable, interpretable, and wear-sensitive features, and form a stable and reliable predictive model, remains a challenge in engineering practice.

[0004] In recent years, artificial intelligence and machine learning have been widely used for tire condition estimation and predictive maintenance. A typical approach involves correlating vehicle sensor data (wheel speed, GPS speed, brake pressure, steering angle, acceleration, etc.) with tire wear / life labels to create a model that outputs tread depth, remaining tire life, or health metrics. While these methods are easy to deploy and highly linear, they are often limited by the observability of available sensor data regarding contact mechanics mechanisms. Many variables strongly correlated with wear (such as changes in local slip direction within the contact area, spatial concentration of friction work, and local temperature rise) are difficult to reconstruct accurately from conventional sensors, resulting in insufficient model generalization ability to changes in operating conditions. Furthermore, several key geometric / uniformity indicators in tire manufacturing and quality control scenarios (such as radially related geometric deviations and features related to rolling radius) have complex coupling relationships with wear. Relying solely on vehicle-side signals or single field quantities can easily lead to problems where predictions are possible but interpretations are difficult, and fitting is possible but transferability is challenging.

[0005] Chinese patent CN118369220A discloses a digital tire health model / digital twin framework. It generates data through model aggregation and iteratively generates multiple tire health models. Finite element analysis can be used to establish a mapping from input conditions (such as load, speed, pressure, and air temperature) to strain and temperature at different tire locations. The results of sub-models such as wear / fatigue are integrated into a health metric, thereby achieving tire health estimation and predictive maintenance. This scheme emphasizes enhancing tire health monitoring capabilities through simulation and data-driven methods and proposes an iterative model update approach. However, this prior art focuses more on the comprehensive modeling and online updating of health metrics. It does not establish a systematic feature extraction system for wear based on the common characteristics related to diameter difference / rolling radius in the manufacturing and design stages; nor does it provide a specific technical path for unifying and integrating working condition characteristics, diameter difference time-series characteristics, slip direction boundary structures, and friction work characteristics for wear prediction and optimization. Therefore, it still has shortcomings in wear prediction driven by multi-dimensional coupled features of diameter difference, slip, and friction work.

[0006] Chinese patent CN112334335A discloses a hybrid machine learning system and method for tire wear estimation. It determines the dynamic rolling radius of the tire from vehicle sensor data using a geometric model and outputs a tread wear estimate based on this dynamic rolling radius. Simultaneously, a machine learning model utilizes sensor data and the correlation between the geometric model output and historical data instances to further output a wear estimate. A switching mechanism allows for the activation selection of the geometric model, machine learning model, or a combination thereof. The advantage of this approach is that it is applicable to in-use vehicle scenarios and can achieve wear estimation without relying on additional expensive sensors. However, it primarily derives the dynamic rolling radius based on vehicle-side signals and geometric relationships, using this as a crucial intermediate variable for wear estimation. It struggles to fully incorporate intermediate variables such as the contact area slip vector field and friction work density field obtainable from simulation. Furthermore, this approach does not provide a complete system for intelligent extraction of multi-scale (static statistics, frequency domain dominant frequency, rate of change, autocorrelation, etc.) features related to diameter difference, nor does it address feature engineering methods for identifying slip direction boundaries from the slip vector field and using them to improve wear prediction accuracy. Therefore, there is still room for improvement in complex operating conditions, multi-scheme comparison during the design phase, and interpretability optimization in manufacturing quality control.

[0007] Chinese patent CN119459715A discloses a method and device for reducing tire wear. The idea is to establish a tire model based on simulation, and perform inflation, loading, and rolling analysis under given working conditions such as air pressure, load, and speed. The friction work data is extracted and post-processed. The tire is divided into multiple cross-sections, and the friction work of multiple cross-section nodes in contact with the ground is fitted and merged. The wear profile is iteratively updated through wear post-processing, and the contact friction behavior is repeatedly calculated to obtain the friction work distribution and form a friction work model. It also proposes to obtain a wear index based on the integral of friction work by torque input to measure the degree of wear resistance. This scheme emphasizes the engineering expression and iterative calculation process of friction work and wear index, which is suitable for wear resistance assessment and optimization in the design stage. However, its focus is on the construction of friction work model and the calculation of wear index. It has not established a machine learning prediction framework for large-scale multi-condition data, nor has it incorporated factors that are strongly related to wear but difficult to quantify manually, such as the characteristics of the diameter difference and the boundary structure of the slip direction, into a unified data fusion model. Furthermore, it lacks a serial prediction from the condition to the diameter difference and then to the wear, as well as a mechanism for generating condition optimization / usage suggestions within the constraint domain. Therefore, it is difficult to meet the actual needs of multi-source data fusion, cross-condition generalization, and prediction-optimization integration.

[0008] In summary, while existing technologies have been researched and applied from the perspectives of digital twin health modeling, vehicle-side dynamic rolling radius fusion estimation with machine learning, and friction work-based simulation-based iterative wear assessment, they still have common shortcomings in several key issues of tire wear prediction: First, regarding the coupling relationship between diameter difference / rolling radius related features and wear, existing solutions either remain at the level of geometric derivation or macroscopic estimation, or lack a multi-scale feature system for wear, resulting in insufficient sensitivity to manufacturing quality fluctuations and structural differences; Second, the unified fusion analysis between multi-source mechanistic information (slip vector field, slip direction structural features, friction work spatial distribution features, etc.) and operating condition features is still insufficient, making it difficult to obtain a high-precision, highly generalized prediction model while ensuring interpretability; Third, existing solutions generally lack a composable modeling approach starting from the operating condition-diameter difference-wear link and closed-loop outputs for engineering applications (such as early warning for high-wear-risk operating conditions, optimization suggestions for load / speed / tire pressure combinations, and a controllable mechanism for incremental model updates with new data). Therefore, there is an urgent need for an intelligent tire wear prediction method and system that can simultaneously address both design and application aspects, based on the fusion of multi-dimensional data such as diameter difference, slippage, and friction work, in order to improve the accuracy, interpretability, and engineering feasibility of wear prediction. Summary of the Invention

[0009] The technical objective of this invention is to provide an artificial intelligence-based tire wear prediction method and system. By acquiring multi-source data such as diameter difference, slip direction boundary line, and friction work under multi-condition finite element simulation and performing feature fusion modeling, a series model of working condition, diameter difference mapping, and diameter difference and wear prediction is established. This enables high-precision prediction and risk warning of tire wear under different loads, speeds, tire pressures, and road conditions, and outputs working condition optimization and usage and maintenance suggestions accordingly. This addresses the problems of insufficient characterization of the coupling mechanism between diameter difference and wear, insufficient data fusion, and insufficient intelligent and generalized feature extraction capabilities in existing technologies.

[0010] Firstly, in order to achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0011] An artificial intelligence-based method for predicting tire wear, comprising the following steps:

[0012] S1, set the working condition parameter vector and perform rolling finite element simulation to obtain the radial difference time sequence, the first tangential slip, the second tangential slip in the contact area, and the friction work distribution; the radial difference time sequence is recorded according to the time step, the first tangential slip, the second tangential slip, and the friction work distribution are recorded according to the contact surface coordinates, and the wear amount corresponding to each working condition is obtained as the supervised learning label of the wear prediction model;

[0013] S2, extract the radial difference feature during the steady-state rolling stage, construct the slip direction angle field from the first tangential slip amount and the second tangential slip amount, and then perform threshold segmentation based on the spatial gradient of the slip direction angle field to obtain the slip direction boundary line set and extract the boundary line feature. At the same time, extract the friction work feature from the friction work distribution.

[0014] S3, train the working condition to diameter difference fusion model, and apply physical consistency constraints to the working condition to diameter difference fusion model so that the average diameter difference characteristic remains monotonically unchanged as the load increases and remains monotonically unchanged as the tire pressure increases.

[0015] S4 uses a fusion feature vector formed by the diameter difference feature, the boundary line feature, and the friction work feature to train the wear prediction model and output the predicted wear.

[0016] S5, the working condition to diameter difference fusion model and the wear prediction model are connected in series to obtain the working condition to wear prediction function. When the predicted wear reaches or exceeds the wear threshold, an early warning is output, and the recommended working condition that minimizes the predicted wear is output within the feasible working condition domain.

[0017] As a further improvement, in step S1, a rolling finite element simulation is performed on the target tire under a preset set of working conditions; a working condition parameter vector is established for any working condition, which includes tire vertical load, vehicle speed, tire pressure, road friction coefficient, longitudinal slip ratio, and sideslip angle; during the steady-state rolling stage, the radial difference time sequence, slip vector field, and friction work density field are output, wherein the slip vector field consists of a first tangential slip and a second tangential slip; and the wear amount is obtained as a supervised learning label; wherein the radial difference is the difference between the actual rolling radius and the reference rolling radius;

[0018] The set of operating conditions is generated by full factorial design or response surface methodology, and the number of operating condition combinations is not less than 100, which is the total number of operating condition samples generated.

[0019] And / or, in step S1, the load is 50% to 150% of the rated load, the speed is 20 to 120 km / h, the tire pressure is 80% to 120% of the standard tire pressure, and each variable has at least 4 discrete levels.

[0020] As a further improvement, in step S2, a path difference feature vector and an auxiliary feature vector are extracted within the steady-state window, wherein the steady-state window is determined by the steady-state start time step and the steady-state end time step; the path difference feature vector includes at least the average path difference, the path difference fluctuation range, the path difference standard deviation, and the path difference dominant frequency; and a slip direction angle field is constructed based on the slip vector field, wherein the slip direction angle field is calculated by the second tangential slip amount and the first tangential slip amount through the four-quadrant arctangent function; by thresholding the spatial gradient magnitude of the slip direction angle field and performing skeletonization on the connected domain, a set of slip direction boundary lines is obtained, wherein the spatial gradient is used to characterize the degree of change of the slip direction angle field on the contact surface; boundary line features are formed accordingly, wherein the boundary line features include the number of boundary lines, the boundary line position index, and the boundary line stability index; and friction work features are formed from the friction work density field, wherein the friction work features include the steady-state total friction work and the friction work distribution non-uniformity;

[0021] Wherein, the steady-state total frictional work is the value obtained by accumulating the frictional work density in the contact area within the steady-state window with respect to area and time; the frictional work distribution non-uniformity is the ratio between the standard deviation and the mean of the frictional work density in the contact area and within the steady-state window.

[0022] As a further improvement, in step S2, the steady-state window is determined by the path difference convergence criterion: when the absolute value of the path difference change rate is not greater than the convergence threshold and a preset number of time steps are continuously satisfied, it is determined that the steady state has been entered. Here, the absolute value of the path difference change rate is the absolute amount of how fast the path difference changes with time, the convergence threshold is the threshold for determining that the path difference has entered a steady state, and the preset number is the number of time steps that continuously satisfy the convergence condition.

[0023] And / or, in step S2, the dominant frequency of the path difference is determined by the frequency corresponding to the maximum power spectral density obtained by performing a fast Fourier transform on the path difference time series;

[0024] And / or, in step S2, the set of sliding direction boundary lines is obtained by the following process: after smoothing the sliding direction angular field, the spatial gradient magnitude is calculated. When the spatial gradient magnitude is not less than the boundary threshold, it is marked as a boundary candidate region, and the set of sliding direction boundary lines is obtained through skeletonization. The boundary line stability index is the standard deviation of the boundary line position index within the steady-state window.

[0025] And / or, in step S2, the friction work density is determined by the contact tangential friction shear stress and the relative sliding velocity, specifically the product of the absolute values ​​of the contact tangential friction shear stress and the relative sliding velocity. The contact tangential friction shear stress is the value along the tangential direction of the contact surface. For further improvement, in step S3, a physical prediction function is constructed to output the predicted diameter difference feature, and the training data drives the residual model to output the residual, resulting in a fusion mapping. The fusion mapping serves as the working condition-to-diameter difference mapping model, with the working condition features as input. During training, a loss function consisting of a mean square error term and a physical consistency penalty term is used. The mean square error term measures the difference between the true diameter difference feature vector and the output of the working condition-to-diameter difference mapping model. The physical consistency penalty term is weighted by physical constraint weights and calculated based on the gradient or sensitivity of the working condition variables. The model is designed to at least satisfy the condition that the average diameter difference feature remains monotonically constant with increasing vertical load and monotonically constant with increasing tire pressure.

[0026] The data-driven residual model is either a gradient boosting decision tree model or a neural network model, and the physical consistency penalty term includes either a monotonicity violation metric or a boundary violation metric.

[0027] As a further improvement, in step S4, the wear prediction model is trained with the fusion feature vector of the sample as input, and the wear prediction value is output. The fusion feature vector is formed by splicing the diameter difference feature, boundary line feature and friction work feature of the sample. The wear amount is the wear amount or wear index obtained by accumulating the product of the absolute value of the relative sliding velocity in the contact area within the steady-state window and the friction work density on the area and time, and then multiplying it by the wear coefficient.

[0028] The wear prediction model uses quantile regression to output confidence intervals, where the first quantile and the second quantile are both greater than 0 and less than 1, and the first quantile is less than the second quantile; the lower limit and the upper limit of the confidence interval are the wear prediction values ​​for the corresponding quantiles.

[0029] As a further improvement, in step S5, for the working condition to be evaluated, the predicted diameter difference features are obtained from the working condition to diameter difference fusion model, and the predicted boundary line features and predicted friction work features are obtained from the auxiliary predictor. The working condition to be evaluated is the working condition parameter vector to be input into the model for wear prediction, and the auxiliary predictor is a regression model with the working condition features as input. The predicted diameter difference features, predicted boundary line features and predicted friction work features are concatenated into vectors and input into the wear prediction model to calculate the predicted wear of the working condition to be evaluated.

[0030] The optimal working condition parameter vector that minimizes predicted wear is solved within the set of working condition constraints, where the set of working condition constraints is the feasible region of the working condition. When the predicted wear of the working condition to be evaluated reaches or exceeds the wear threshold, an early warning is output. The optimal working condition parameter vector is solved using either Bayesian optimization or gradient search, and the maximum number of iterations is the upper limit of the iteration.

[0031] Secondly, the present invention also provides an artificial intelligence-based tire wear prediction system, which is used to implement the method, including:

[0032] The simulation data management module is used to generate a set of working conditions and schedule finite element simulations, storing the diameter difference, slip vector field, friction work density and wear amount;

[0033] The feature extraction module is used to generate diameter difference features, boundary line features, and friction work features;

[0034] The model building module is used to train the working condition to diameter difference mapping model and the diameter difference to wear model;

[0035] The prediction application module is used to calculate predicted wear, solve for the optimal working condition parameter vector, and output early warning based on the wear threshold;

[0036] The model management module is used for model updates and performance monitoring.

[0037] Thirdly, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the method.

[0038] Fourthly, the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the method.

[0039] This invention utilizes a two-level modeling chain encompassing working conditions, radial difference, and wear. It integrates working condition characteristics such as load, speed, tire pressure, and road friction coefficient with statistical / frequency domain / rate-of-change characteristics of radial difference time series, characteristics of the slip direction boundary line (quantity, location, stability) obtained from slip vector field identification, and key mechanistic features such as the total frictional work and distribution unevenness in the contact area. Furthermore, it introduces the fusion of physical models and machine learning residual models, along with physical consistency constraints such as gradient / monotonicity. This results in a prediction model that significantly improves the accuracy and generalization stability of cross-working-condition wear prediction, while reducing the impact on... The reliance on a single sensor value or a single wear index can be overcome by outputting the contribution and sensitivity results of influencing factors, enhancing interpretability and traceability. At the same time, based on the comprehensive prediction model, the load, speed and tire pressure combination with the minimum wear can be quickly searched within the constrained operating condition domain to generate actionable usage and maintenance recommendations. It can also provide early warning when the predicted wear exceeds the threshold, thereby reducing the cost of long-cycle road testing and repeated calibration in tire design iteration, manufacturing quality control and fleet operation and maintenance scenarios, shortening the R&D cycle, reducing maintenance downtime risks and improving the engineering implementation capability of tire life management. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of an AI-based intelligent tire wear prediction system.

[0041] Figure 2 This is a schematic diagram of a tire wear prediction method based on artificial intelligence.

[0042] Figure 3 A schematic diagram illustrating the alignment of the steady-state window with the time series.

[0043] Figure 4 This is a schematic diagram for extracting the angular field of the slip direction and the boundary line of the slip direction.

[0044] Figure 5 This is a schematic diagram of integrated prediction and optimization of working conditions and wear.

[0045] Figure 6 This diagram illustrates the process of model training, validation, incremental updates, and version management.

[0046] Figure 7 This is a data distribution diagram of the first tangential slip CSLIP1 within the contact area of ​​Case 2.

[0047] Figure 8 This is a data distribution diagram of the second tangential slip CSLIP2 within the contact area of ​​Case 2.

[0048] Figure 9 This is a cloud map showing the distribution of frictional work within the contact area in Case 2.

[0049] Figure 10A scatter plot comparing the wear predictions and measured values ​​from different models.

[0050] Figure 11 A comparison chart of wear prediction errors for different models. Detailed Implementation

[0051] The preferred embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be understood that the following embodiments are used to illustrate the technical solution, implementation path, and achievable technical effects of the present invention, and do not constitute a limitation on the scope of protection. Equivalent substitutions made by those skilled in the art in terms of structural composition, data processing order, model form, or parameter range without departing from the concept of the present invention should all fall within the scope of protection of the present invention.

[0052] I. Terminology Explanation

[0053] Operating condition parameter vector: denoted as ,in For the vertical load of the tire, For speed, Tire pressure, The coefficient of friction of the road surface. For longitudinal slip ratio, It is the sideslip angle.

[0054] Radius difference: The difference between the actual rolling radius of the tire and the reference radius, denoted as . .

[0055] CSLIP1 / CSLIP2: The cumulative relative slip (or equivalent tangential slip displacement component) in the first / second tangential direction of the contact output. The symbol is used to characterize the slip direction.

[0056] Slip direction angle field: The slip direction angle constructed by CSLIP1 / CSLIP2, denoted as... .

[0057] Slip direction boundary: A linear set of lines where the angular field of the slip direction undergoes abrupt changes or rapid transitions, denoted as... .

[0058] Frictional work (moca-ABS): an absolute scalar quantity (or an equivalent scalar quantity of frictional work density) of the energy dissipated in the contact area, used to characterize the intensity of energy dissipation and hot spot areas.

[0059] Steady-state window: The time interval used for statistics and modeling after removing the initial transient state, denoted as .

[0060] Operating condition-path difference mapping model: A model that predicts path difference characteristics from operating condition characteristics, denoted as... .

[0061] The diameter difference-wear model: a model that predicts wear based on fused features, denoted as... .

[0062] Wear threshold: The wear threshold used for risk warning, denoted as... .

[0063] II. System Structure of the Invention

[0064] like Figure 1 As shown, the intelligent tire wear prediction system provided by this invention includes at least: a simulation data management module 100, a feature extraction module 200, a model building module 300, a prediction application module 400, and a model management module 500. Each module can interact with other modules through shared storage, a database, or a message queue. It can be centrally deployed on the same server or distributed across a computing cluster and business servers.

[0065] 1) Simulation Data Management Module 100

[0066] Used for generating set of operating conditions, scheduling finite element simulations, collecting field quantities and labels, and completing data storage and retrieval, preferably including:

[0067] Working condition design unit 110: generated using full factorial, orthogonal design, Latin hypercube, or response surface methodology. ;

[0068] Simulation task scheduling unit 120: schedules each set of working conditions Convert the input to the solver (boundary conditions such as load, tire pressure, speed, friction coefficient, slip ratio, sideslip angle, etc.) and perform batch rolling simulation;

[0069] Data storage and retrieval unit 130: Storage It collects multi-source data such as CSLIP1 / CSLIP2, friction work (moca-ABS), contact pressure, shear stress, and temperature, and establishes an index by tire version, working condition number, and time period.

[0070] 2) Feature extraction module 200

[0071] For converting high-dimensional spatiotemporal field quantities into learnable and interpretable structured features, preferably including:

[0072] Operating condition feature extraction unit 210: for Perform standardization and combination feature generation (e.g.) , wait);

[0073] Path difference feature extraction unit 220: for Extract statistical features, frequency domain features, rate of change, and autocorrelation features;

[0074] Boundary line identification unit 230: constructed from CSLIP1 / CSLIP2 And identify Output Structural features;

[0075] Friction work feature extraction unit 240: calculated from moca-ABS or friction work density field , , Features such as...

[0076] 3) Model building module 300

[0077] For training and validating two-level models and fixing the version, preferably including:

[0078] Working condition-path difference mapping model unit 310: Implementing the physical model Residual Model and integration ;

[0079] Circumference Difference-Wear Model Unit 320: Training ,in For the characteristics of diameter difference, For the structural features of the boundary line, This is a characteristic of frictional work;

[0080] Integrated Prediction Model Unit 330: will and Serialization, or training of end-to-end models, can be used to directly predict wear.

[0081] 4) Predictive Application Module 400

[0082] Used for wear prediction, confidence interval output, risk warning and optimization suggestion generation, preferably including:

[0083] Wear prediction unit 410: Output And risk level;

[0084] Report generation unit 420: Outputs the contribution of influencing factors, sensitivity conclusions, and visualizations;

[0085] Use suggestion generation unit 430: Search for the combination of operating conditions that minimizes wear within the constraint domain. ;

[0086] Early warning unit 440: When Warning triggered.

[0087] 5) Model Management Module 500

[0088] Used for training data versioning, incremental learning, error monitoring and rollback. Figure 6 ), preferably including:

[0089] Model Training and Validation Unit 510: Dataset Splitting, Cross-Validation, and Metric Evaluation;

[0090] Model update unit 520: Triggers incremental training / retraining by period or drift;

[0091] Performance monitoring unit 530: Monitors input distribution drift and prediction error drift;

[0092] Version Management Unit 540: Saves model weights, hyperparameters, standardized statistics, and training data fingerprints.

[0093] III. Overall Technical Route for Implementing the Method of the Invention

[0094] like Figure 2 As shown, steps S1–S5 of the method of the present invention are as follows:

[0095] Step S1, set the operating condition parameter vector Rolling finite element simulation was performed to obtain the path difference time sequence. First and second tangential slip in the contact area and and the distribution of frictional work ;in For the vertical load of the tire, For vehicle speed, For tire pressure, The coefficient of friction of the road surface. For longitudinal slip ratio, Side slip angle, For time steps, Let these be the coordinates of the contact surface;

[0096] Step S2: Extract the diameter difference feature during the steady-state rolling phase. And construct the slip direction angle field from the slip amount. Based on Thresholding of spatial gradients yields a set of sliding direction boundaries. And extract the boundary line features. Simultaneously, friction work characteristics are extracted from the friction work distribution. ;

[0097] Step S3: Train the working condition-path difference fusion model And apply physical consistency constraints to the fusion model to make the average path difference feature With load Monotonous and unchanging with tire pressure Monotonic and not increasing;

[0098] Step S4, to fuse feature vectors Training wear prediction model It also outputs predicted wear. ;

[0099] Step S5: Connect the working condition-path difference fusion model. Wear prediction model Obtain the prediction function from working condition to wear. ,when Output early warning and output within the feasible operating condition domain Minimum recommended operating conditions.

[0100] IV. Specific Implementation Methods of the Method Steps of the Invention

[0101] (I) Step S1: Generation and Acquisition of Multi-condition Simulation Data

[0102] 1. Working condition set design and coverage

[0103] In one implementation, the operating condition variables and their ranges are set as follows:

[0104] Rated load – Take 5–7 levels;

[0105] : 20–120 km / h, take 6–8 levels;

[0106] Standard tire pressure – Take 4–5 levels;

[0107] At least three typical levels should be selected (e.g., 0.5, 0.8, 0.9).

[0108] and Covers typical scenarios of straight-line movement, turning, and braking / traction (e.g.) ; (etc.). The number of working condition combinations satisfies This is to ensure data coverage and model generalization ability.

[0109] 2. Finite element rolling simulation and output fields

[0110] A rolling finite element model is established for each fixed tire design version (structure, material, and tread pattern remain unchanged). Apply boundary conditions and run a transient rolling analysis. The output data should include at least the path difference sequence. Contact areas CSLIP1, CSLIP2 (or equivalent tangential slip output); friction work moca-ABS (or friction work density equivalent field); optional auxiliary quantities: contact pressure, shear stress, temperature field, contact area, etc.

[0111] The diameter difference is defined as: ;in: For the difference in diameter; This is the actual rolling radius; For reference rolling radius (theoretical or nominal radius).

[0112] 3. Data storage and metadata

[0113] Each sample record includes: tire version ID, operating condition number, solver version, contact model parameters, and time step. The dataset index table, which includes grid version, output field list, steady-state window parameters (described later), etc., forms a traceable dataset index table, facilitating recalculation and retraining.

[0114] (II) Extraction of features of working condition S2 / diameter difference / slippage boundary / friction work

[0115] The goal of step S2 is to transform the multi-source high-dimensional data obtained in step S1 into a wear-sensitive, interpretable, and reproducible feature set, and to eliminate the interference of initial transients, sampling differences, and numerical noise on model training. This step can be divided into: steady-state window determination ( Figure 3 ), diameter difference feature extraction, slip direction boundary line feature extraction ( Figure 4 , Figure 7 , Figure 8 Friction work feature extraction () Figure 9 ), as well as feature splicing and standardization.

[0116] 1) Determining the steady-state window and aligning it with the time series

[0117] In the initial stages of rolling simulations, transient vibrations and contact establishment often occur. Directly using full-range data can lead to characteristic bias. Therefore, it is preferable to use the convergence criterion of the rate of change of the diameter difference to determine the steady-state starting point. .

[0118] First, calculate the rate of change of the diameter difference:

[0119] ;

[0120] in: This represents the rate of change of the diameter difference. The sampling time interval; For time step indexing.

[0121] Set threshold With consecutive steps When the following conditions are met:

[0122] ;

[0123] The starting time step that satisfies the condition is defined as... ; and order This is the time step for the simulation to end or for the steady-state to end. Where: The convergence threshold (can be an empirical value on the order of the path difference, such as 1%–5% / s). To continuously satisfy the required number of steps (which can be 50–200 steps, depending on the situation) Adjustment).

[0124] To ensure statistical stability, the steady-state window length is constrained to satisfy... ,in Set the minimum number of steady-state sampling points (e.g., 1000 points). If this is not met, extend the simulation time or remove unstable condition samples.

[0125] Alignment strategy: Use the same sampling frequency and window length for all working condition samples; if the number of steps output by different solution tasks is inconsistent, then... Resampling (linear interpolation or spline interpolation) is performed within the same length. .

[0126] 2) Construction and combination features of working conditions

[0127] Basic operating condition characteristics are To enhance the model's representation of coupling patterns, combined features can be constructed, for example:

[0128] Load-speed ratio ;

[0129] Tire pressure-load ratio ;

[0130] Comprehensive operating conditions : in: For comprehensive operating conditions; These are the weighting coefficients (which can be equally weighted or obtained through regression fitting after standardization of the training set). The operating condition type (straight line / turning / braking, etc.) can be one-hot encoded to form the final operating condition feature vector. (Dimensions can be 10–20).

[0131] 3) Path difference feature extraction (statistical + dynamic, forming) )

[0132] Extract the following path difference features within the steady-state window (which may include at least the mean, standard deviation, fluctuation range, and dominant frequency):

[0133] (1) Average diameter difference:

[0134] ;

[0135] in: The average diameter difference; This represents the number of steady-state sampling points.

[0136] (2) Standard deviation of circumference:

[0137] ;

[0138] in: This represents the standard deviation of the diameter difference.

[0139] (3) Range of radii difference fluctuation:

[0140] ;

[0141] in: This represents the range of radial difference fluctuations.

[0142] (4) Path difference dominant frequency (FFT extraction)

[0143] The amplitude spectrum is obtained by performing a discrete Fourier transform on the steady-state sequence. The frequency corresponding to the largest amplitude value is taken as the main frequency:

[0144] ;

[0145] in: The dominant frequency is the difference in diameter. It is a set of discrete frequencies; This is the amplitude spectrum.

[0146] (5) Rate of change characteristics (optional):

[0147] ;

[0148] in: This represents the maximum rate of change of the diameter difference.

[0149] The above features are spliced ​​together to form Standardization is performed (Z-score: subtract the mean and divide the variance), and the standardized parameters (mean and standard deviation) are fixed along with the model version to ensure deployment consistency.

[0150] 4) Extraction of the glide direction angle field and boundary line

[0151] Using only the slip magnitude or average slip often fails to represent the directional partitioning and abrupt structural changes within the contact region. This invention utilizes CSLIP1 and CSLIP2 to construct the slip direction angular field and identify the boundary lines. Extracting structural features This enhances the sensitivity to wear hotspots and uneven wear.

[0152] (1) Calculation of the angle field of the slip direction

[0153] For the slip field at each time step within the steady-state window or after time aggregation, the orientation angle is calculated as follows:

[0154] ;

[0155] in: The sliding direction angle; It is the arctangent function in the four quadrants; Let these be the coordinates of the contact surface; For time step; CSLIP1 / CSLIP2 are the corresponding tangential slip amounts.

[0156] To reduce numerical noise, CSLIP1 / CSLIP2 or Apply a smoothing filter (Gaussian filter or median filter), and the filter kernel size can be 1–3 times the contact mesh size.

[0157] (2) Boundary candidate region identification

[0158] Calculate the angular field gradient magnitude:

[0159] ;

[0160] in: This represents the gradient magnitude. It is a two-dimensional gradient operator; It is the Euclidean norm.

[0161] Set threshold , will satisfy The region is marked as a candidate boundary region. Threshold An adaptive approach can be adopted:

[0162] ;

[0163] in: and They are respectively Mean and standard deviation in the contact area; This is the multiplier (e.g., 1–2).

[0164] (3) Morphological processing and skeletalization

[0165] Opening and closing operations are performed on candidate regions to remove discrete noise and holes. Then, connected component skeletonization is performed (refined to single-pixel or single-unit width lines) to obtain the set of boundary lines. To improve stability, time fusion can be performed within the steady-state window: the frequency at which each grid point is identified as a boundary line is counted, and if the frequency exceeds a threshold... (For example, 0.6–0.8) are retained as the final set of boundary lines. .

[0166] (4) Extraction of structural features of the boundary line

[0167] Number of dividing lines:

[0168] in: is the number of connected components; is the counting operator.

[0169] Position indicator (in scroll direction) (Taking the normalized center of mass as an example)

[0170] ;

[0171] ;

[0172] in: The centroid of the dividing line; The number of points on the dividing line; This is the boundary of the contact area; For location indicators (0–1).

[0173] Stability indicators (location fluctuates over time):

[0174] ;

[0175] in: For stability indicators; For a moment The location of the dividing line; Let it be its mean.

[0176] (5) Symbol partitioning bias characteristics

[0177] To achieve the goal of not requiring a complete extraction In cases where orientation bias can be quickly characterized, a sign-statistical bias index can be introduced. :

[0178] ;

[0179] in: For the first The number of sampling points with a directional slip of not less than 0; The number of sampling points less than 0; Directional offset (range of values) ).

[0180] Finally, the feature vector of the boundary line structure can be expressed as:

[0181] ;

[0182] And standardize the process.

[0183] 5) Extraction of frictional work features

[0184] Frictional work reflects the intensity of energy dissipation within the contact area, and its spatial non-uniformity and hot spots often correspond to localized accelerated wear. This invention extracts the total amount and distribution characteristics of the frictional work field within a steady-state window.

[0185] (1) If the solver directly outputs moca-ABS (absolute value of frictional work scalar), it can be directly applied to the contact area. The total frictional work is obtained by internal integration:

[0186] ;

[0187] in: The total frictional work in steady state; It is the frictional work density or equivalent frictional work scalar (which can be taken as moca-ABS or its discrete equivalent); This refers to the contact area.

[0188] (2) Distribution unevenness (coefficient of variation):

[0189] ;

[0190] in: Non-uniformity; for Standard deviation; for The mean.

[0191] (3) Percentage of hotspot area

[0192] Pick quantile threshold Define the proportion of high-friction work area:

[0193] ;

[0194] in: The percentage of hotspot area; For indicator functions; The 90th percentile threshold.

[0195] The final friction work eigenvector is:

[0196] ;

[0197] 6) Feature concatenation and dataset format

[0198] For each working condition sample form:

[0199] Step S3 Training Samples: Input Output ;

[0200] Step S4 Training Samples: Input Output wear labels (Given by experimental or wear mechanism models).

[0201] Dataset fields are recommended to include at least:

[0202] tire_version_id (tire version), case_id (condition number), Fz, v, P, mu, s, alpha;

[0203] Ts, Te, dt (steady-state and sampling parameters);

[0204] Y_DeltaR_* (diameter difference characteristic column), B_* (boundary line characteristic column), Q_* (friction work characteristic column);

[0205] Metadata such as W_label (wear label), source (simulation / experiment source), solver_version, etc.

[0206] Data partitioning principles: Duplicate samples with the same operating condition number should be grouped into the same subset to avoid data leakage; and it is recommended to stratify sampling by speed / load to ensure consistent distribution for training / validation / testing.

[0207] (III) Step S3: Establish the working condition-path difference mapping model

[0208] Step S3 aims to establish a mapping from operating conditions to radii difference characteristics. Furthermore, it enhances generalization ability and interpretability through physical modeling, residual learning, and physical consistency constraints.

[0209] 1) Physical Model

[0210] In one implementation, for key path difference features (such as...) Fitting in an interpretable form:

[0211] ;

[0212] in: The average diameter difference predicted by the physical model; The coefficients to be calibrated; Same as before. Coefficient calibration: Least squares or ridge regression are used; to enhance robustness, coefficients can be calibrated. Standardize first, then fit.

[0213] For other diameter difference characteristics (such as) (etc.) can be expressed using similar low-order polynomials or families of linear interpretable functions to form vector outputs. .

[0214] 2) Residual Model

[0215] Define residual:

[0216] ;

[0217] Training data-driven models study .

[0218] Preferred Model A: Gradient Boosting Decision Tree (GBDT)

[0219] Input dimensions: 10–20; Output dimensions: 5–10;

[0220] Number of trees 300–800; Maximum depth : 4–8; learning rate : 0.03–0.1; Subsampling rate : 0.6–0.9;

[0221] Early stopping: Stop when the RMSE on the validation set does not decrease for 20–50 consecutive rounds;

[0222] Optional monotonic constraints: for Apply a monotonically increasing force to Apply a monotonically decreasing constraint (if using an implementation that supports monotonic constraints).

[0223] Preferred Model B: Multilayer Perceptron (MLP)

[0224] Input layer: dimension Hidden layers: three layers (128, 64, 32); Output layer: dimension ;

[0225] Activation: ReLU or GELU;

[0226] Regular expression: Dropout 0.1–0.3, or L2 weight decay – ;

[0227] Optimizer: Adam; Learning Rate – Batch size 32–256;

[0228] Training rounds: 100–500, with early stops.

[0229] 3) Fusion and Loss Function

[0230] Fusion model:

[0231] ;

[0232] The training loss is the sum of the fitting error and the physical consistency penalty:

[0233] ;

[0234] in: Total loss; Mean square error; The physical constraint weights (can be 0.1–10 and mesh search is used); This is a penalty for physical consistency.

[0235] Preferred At least include the monotonicity constraint of the mean diameter difference:

[0236] ;

[0237] in: Predict the average path difference for the model; and Sensitivity to load and tire pressure; To calculate the expectation of the training samples.

[0238] Implementation method: Use finite difference to approximate the gradient, for example, for... or The output difference is calculated by perturbing by about 1%, thus enabling reproducible constrained training.

[0239] 4) Training / Verification / Testing and Output Consolidation

[0240] Data partitioning: 70% training, 15% validation, 15% testing; stratified sampling by speed / load; overall partitioning of duplicate samples under the same working conditions;

[0241] index: RMSE, MAE;

[0242] Output solidification: Model weights, hyperparameters, standardized parameters (mean / variance), feature column order, training data fingerprint (hash), and version number. .

[0243] (iv) Step S4: Establish the diameter difference-wear relationship model

[0244] Step S4 uses fusion features Input to train the wear prediction model This step demonstrates the invention's ability to model the coupling mechanism of directional structural abrupt change, energy hotspots, and wear acceleration.

[0245] 1) Wear label Acquisition (experimentation preferred; mechanistic model optional)

[0246] In one embodiment, the wear labels are preferably derived from indoor flat belt / drum wear tests (see application examples described later) and normalized to equivalent mileage. Wear is defined as:

[0247] ;

[0248] in: Wear amount (mm / 1000km); The average pattern depth (mm) before the test; The average pattern depth (mm) after the test; This represents the cumulative mileage (km).

[0249] When performing simulation verification only, the energy coupling wear index can also be used as a label:

[0250] ;

[0251] in: The wear coefficient; Relative slip velocity; This represents the frictional work density or equivalent scalar; the rest is the same as before. Experiments and simulations can be performed through calibration. Achieve dimensional consistency or use normalized comparison.

[0252] 2) Model Form and Hyperparameter Disclosure

[0253] Preferred Model: GBDT Regression

[0254] Input Dimensions : 15–30 (varying with the number of features); Output: ;

[0255] Number of trees: 500–1500; Depth: 4–10; Learning rate: 0.01–0.1; Minimum number of leaf samples: 10–100;

[0256] Evaluation: RMSE, MAE, MAPE, residual distribution;

[0257] Interpretability: Outputs feature importance or SHAP values ​​for report generation.

[0258] Optional model: Quantile regression output confidence interval

[0259] Based on quantiles Output The loss is quantile loss (which can be used to express uncertainty), which facilitates the formulation of engineering early warning strategies.

[0260] 3) Training strategies and overfitting prevention

[0261] Standardization: for Standardize using the mean / variance of the training set;

[0262] Cross-validation: K-fold (K=5 or 10);

[0263] Early stopping: Stop when the validation set error does not decrease for several consecutive rounds;

[0264] When there is class imbalance (few highly worn samples): highly worn samples can be weighted to improve risk identification capabilities.

[0265] 4) Output curing

[0266] Fixed model weights and version number Simultaneously, solidify the feature column order, standardized parameters, training logs, and error threshold strategies (e.g.) (Taking the 90th percentile of historical samples, etc.).

[0267] (V) Step S5: Comprehensive Prediction, Early Warning and Operating Condition Optimization

[0268] 1) Comprehensive forecast

[0269] Treatment of evaluation conditions First by Obtain the predicted path difference characteristics Simultaneously, it can be used by auxiliary predictors. Complete (Or, if the value is available through simulation / online sensing, it can be directly taken), ultimately:

[0270] ;

[0271] in: To predict wear; This involves concatenating vectors. A regression model that takes operating condition characteristics as input (optional).

[0272] 2) Early warning criteria

[0273] When the following conditions are met:

[0274] ;

[0275] It will then output a warning and suggest adjustments to the operating conditions. It can be set by the enterprise's maintenance strategy or determined by historical sample quantiles.

[0276] 3) Operating condition optimization and suggestion generation

[0277] In the feasible region The internal solution for the condition with minimum wear:

[0278] ;

[0279] in: This is the optimal operating condition; For the constraint domain (lower safe tire pressure, upper load, upper speed, etc.).

[0280] The solution can be obtained using Bayesian optimization or heuristic search, and will output actionable suggestions (such as recommended tire pressure ranges, speed limits, and load control strategies).

[0281] V. Application Example 1: Verification of Indoor Wear Test Bench under Multiple Working Conditions

[0282] (I) Experimental Objective

[0283] The wear prediction method of the present invention, which integrates the diameter difference, slip direction boundary line, and friction work in multiple dimensions, was verified under different loads, speeds, tire pressures, friction coefficients, and slip / lateral deviation conditions. The method was compared with a comparative scheme that does not include boundary line features to demonstrate the technical effectiveness of the present invention.

[0284] (II) Test Subjects and Equipment

[0285] Test subjects: radial tires (205 / 55R16) of the same specification and batch. Each working condition was tested at least 3 times, and the average value was taken as the wear result of that working condition.

[0286] Testing equipment: Indoor flat belt / drum abrasion test bench, equipped with:

[0287] Vertical load closed-loop control (range covering 3–10kN).

[0288] Speed ​​closed-loop control (20–120 km / h);

[0289] Tire pressure control (200–300 kPa);

[0290] Longitudinal slip ratio Side slip angle Loading capability;

[0291] Tire force / torque acquisition system;

[0292] Tire tread temperature monitoring (infrared or embedded type);

[0293] Tire wear measurement (laser depth measurement or high-precision depth gauge).

[0294] Environmental conditions: temperature relative humidity Before the test, the tires should be preheated / broken in (e.g., at 60 km / h, rated load, for 20 minutes).

[0295] (III) Wear Index and Calculation Method

[0296] The tread depth was measured at multiple points before and after the test, and the average value was taken. The wear index was obtained by normalizing the values ​​based on the equivalent mileage. :

[0297] ;

[0298] in: Wear amount (mm / 1000km); The average pattern depth (mm) before the test; The average pattern depth (mm) after the test; The cumulative mileage (km) for this operating condition.

[0299] (iv) Data acquisition and feature extraction corresponding to the method of the present invention

[0300] For each machine's operating condition Establish the corresponding finite element rolling simulation boundary conditions within the steady-state window. Internal output and extraction:

[0301] path difference sequence and its characteristics (like , , wait);

[0302] Distribution of contact area CSLIP1 / CSLIP2 and structural characteristics of the boundary line between slip directions (like wait);

[0303] Distribution and characteristics of frictional work (moca-ABS) (like wait).

[0304] To facilitate rapid characterization of orientation bias without relying on a full set of image processing, a symbolic statistical bias index can also be used. :

[0305] ;

[0306] in: For the first The number of sampling points with a directional slip of not less than 0; The number of sampling points less than 0; Indicates the degree of directional offset.

[0307] (v) Setting up the comparison scheme (comparison ratio)

[0308] To highlight the technical effects of this invention, the following comparative model is set up:

[0309] Comparative Example 1 / Comparison Model A (Path Difference Only): Input consists only of path difference features Without introducing frictional work characteristics and boundary line structural characteristics;

[0310] Comparative Example 2 / Comparative Model B (Diameter Difference + Friction Work): Input is Without introducing boundary line structural features ;

[0311] Example / Model C of this invention (Diameter difference + Boundary line + Frictional work): Input is And adopt a physical model +Residual Model Training with fusion and physical consistency constraints.

[0312] (vi) Experimental Data and Results

[0313] Table 1 Comparison of measured and predicted wear under multiple working conditions (unit: mm / 1000km) (each working condition was repeated 3 times and the average value was taken)

[0314]

[0315] Table 2 Error Index Statistics (e.g.) Figure 11 )

[0316]

[0317] (vii) Conclusion on Technical Effectiveness (Application Example 1):

[0318] Compared with Comparison A, the RMSE of this invention C is reduced by approximately 90.55%; compared with Comparison B, the RMSE of this invention C is further reduced by approximately 75.02%. This indicates that using only the diameter difference or only the diameter difference and friction work are insufficient to stably characterize the wear difference caused by structural changes in the sliding direction, while introducing the boundary line structural feature can significantly improve the prediction accuracy and cross-condition generalization stability.

[0319] VI. Application Example 2: Combining Figure 7 – Figure 9 Verify the contribution of boundary / directional structural features

[0320] (I) Application Example 2: Description of Features in the Attached Figures

[0321] In application example 2, Figure 7 and Figure 8 The symbol partition statistics for CSLIP1 and CSLIP2 within the contact area are given:

[0322] CSLIP1: There are 2082 points. There are 2047 points;

[0323] CSLIP2: There are 889 points. There are 3240 points;

[0324] Figure 9 The frictional work (moca-ABS) is shown to exhibit distinct hot spots and uneven distribution within the contact patch (upper limit of color scale approximately). (Scale), indicating the presence of localized energy concentration and potential risk of accelerated wear in localized areas.

[0325] (II) Directional bias quantization and boundary characteristics

[0326] By bias index calculate:

[0327] For CSLIP1: This indicates that the entire region is approximately symmetrically partitioned in direction 1.

[0328] For CSLIP2: This indicates that there is a significant negative bias in both directions (strong directional dominance).

[0329] In this invention, a slip direction angle field is further constructed:

[0330] ;

[0331] in: The sliding direction angle; and based on Threshold segmentation and skeletonization to obtain the boundary set ,extract These structural features can explicitly quantify the directional bias and transition band, thereby explaining and predicting... Figure 9 The diagram shows the migration of frictional work hotspots and the trend of uneven wear.

[0332] (III) Correspondence verification with bench wear results

[0333] Taking the combined slip / lateral deviation condition in Table 1 (e.g., E9 type) as an example, the measured wear on the test bench is approximately mm / 1000km:

[0334] Compare with A (only) The predicted value was 1.26, which has a large deviation.

[0335] Compare B ( The predicted value for frictional work is 1.07, which still shows a significant deviation.

[0336] This invention C ( The predicted value (including the boundary line and frictional work) is 0.99, which is more consistent with the actual measurement.

[0337] (iv) Conclusion on technical effectiveness:

[0338] The structure is significantly offset in the slip direction ( Under conditions where the frictional work distribution is significantly negative and hot spots exist, using only the radial difference or the radial difference plus the total energy can easily lead to systematic deviations. By introducing the boundary line / directional structural features, the coupling relationship between abrupt changes in directional structure, energy concentration, and accelerated wear can be incorporated into the learning input, thereby significantly reducing errors and improving the reliability of early warning.

[0339] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.

[0340] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0341] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0342] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0343] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0344] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0345] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0346] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

Claims

1. A tire wear prediction method based on artificial intelligence, characterized in that, The method includes the following steps: S1, set the working condition parameter vector and perform rolling finite element simulation to obtain the radial difference time sequence, the first tangential slip, the second tangential slip in the contact area, and the friction work distribution; the radial difference time sequence is recorded according to the time step, the first tangential slip, the second tangential slip, and the friction work distribution are recorded according to the contact surface coordinates, and the wear amount corresponding to each working condition is obtained as the supervised learning label of the wear prediction model; S2, extract the radial difference feature during the steady-state rolling stage, construct the slip direction angle field from the first tangential slip amount and the second tangential slip amount, and then perform threshold segmentation based on the spatial gradient of the slip direction angle field to obtain the slip direction boundary line set and extract the boundary line feature. At the same time, extract the friction work feature from the friction work distribution. S3, train the working condition to diameter difference fusion model, and apply physical consistency constraints to the working condition to diameter difference fusion model so that the average diameter difference characteristic remains monotonically unchanged as the load increases and remains monotonically unchanged as the tire pressure increases. S4 uses a fusion feature vector formed by the diameter difference feature, the boundary line feature, and the friction work feature to train the wear prediction model and output the predicted wear. S5, connect the working condition to diameter difference fusion model and the wear prediction model to obtain the working condition to wear prediction function. When the predicted wear reaches or exceeds the wear threshold, output an early warning and output the recommended working condition that minimizes the predicted wear within the feasible working condition domain. In step S5, for the working condition to be evaluated, the predicted diameter difference features are obtained from the working condition to diameter difference fusion model, and the predicted boundary line features and predicted friction work features are obtained from the auxiliary predictor. The working condition to be evaluated is the working condition parameter vector to be input into the model for wear prediction, and the auxiliary predictor is a regression model with the working condition features as input. The predicted diameter difference features, predicted boundary line features and predicted friction work features are concatenated into vectors and input into the wear prediction model to calculate the predicted wear of the working condition to be evaluated.

2. The method as described in claim 1, characterized in that, In step S1, a rolling finite element simulation is performed on the target tire under a preset set of working conditions; a working condition parameter vector is established for any working condition, which includes tire vertical load, vehicle speed, tire pressure, road friction coefficient, longitudinal slip ratio, and sideslip angle; during the steady-state rolling stage, the radial difference time series, slip vector field, and friction work density field are output, wherein the slip vector field consists of a first tangential slip and a second tangential slip; and the wear amount is obtained as a supervised learning label; wherein the radial difference is the difference between the actual rolling radius and the reference rolling radius; The set of operating conditions is generated by full factorial design or response surface methodology, and the number of operating condition combinations is not less than 100, which is the total number of operating condition samples generated. And / or, in step S1, the load is 50% to 150% of the rated load, the speed is 20 to 120 km / h, the tire pressure is 80% to 120% of the standard tire pressure, and each variable has at least 4 discrete levels.

3. The method as described in claim 2, characterized in that, In step S2, a radial difference feature vector and an auxiliary feature vector are extracted within the steady-state window. The steady-state window is determined by the steady-state start time step and the steady-state end time step. The radial difference feature vector includes at least the average radial difference, the radial difference fluctuation range, the radial difference standard deviation, and the radial difference dominant frequency. A sliding direction angle field is constructed based on the sliding vector field. The sliding direction angle field is calculated by the second tangential sliding amount and the first tangential sliding amount using the four-quadrant arctangent function. By thresholding the spatial gradient magnitude of the sliding direction angle field and performing skeletonization on the connected components, a set of sliding direction boundary lines is obtained. The spatial gradient is used to characterize the degree of change of the sliding direction angle field on the contact surface. Boundary line features are formed accordingly, including the number of boundary lines, the boundary line position index, and the boundary line stability index. Friction work features are formed from the friction work density field, including the steady-state total friction work and the friction work distribution non-uniformity. Wherein, the steady-state total frictional work is the value obtained by accumulating the frictional work density in the contact area within the steady-state window with respect to area and time; the frictional work distribution non-uniformity is the ratio between the standard deviation and the mean of the frictional work density in the contact area and within the steady-state window.

4. The method as described in claim 3, characterized in that, In step S2, the steady-state window is determined by the path difference convergence criterion: when the absolute value of the path difference change rate is not greater than the convergence threshold and a preset number of time steps are continuously met, it is determined that the steady state has been entered. Here, the absolute value of the path difference change rate is the absolute amount of how fast the path difference changes with time, the convergence threshold is the threshold for determining that the path difference has entered a steady state, and the preset number is the number of time steps that continuously meet the convergence condition. And / or, in step S2, the dominant frequency of the path difference is determined by the frequency corresponding to the maximum power spectral density obtained by performing a fast Fourier transform on the path difference time series; And / or, in step S2, the set of sliding direction boundary lines is obtained by the following process: after smoothing the sliding direction angular field, the spatial gradient magnitude is calculated. When the spatial gradient magnitude is not less than the boundary threshold, it is marked as a boundary candidate region, and the set of sliding direction boundary lines is obtained through skeletonization. The boundary line stability index is the standard deviation of the boundary line position index within the steady-state window. And / or, in step S2, the frictional work density is determined by the contact tangential frictional shear stress and the relative sliding velocity, specifically the product of the absolute values ​​of the contact tangential frictional shear stress and the relative sliding velocity, wherein the contact tangential frictional shear stress is the frictional shear stress in the tangential direction of the contact surface, and the relative sliding velocity is the relative sliding velocity on the contact surface.

5. The method as described in claim 1, characterized in that, In step S3, a physical prediction function is constructed to output the predicted diameter difference feature, and the training data drives the residual model to output the residual, resulting in a fusion mapping. This fusion mapping serves as the working condition-to-diameter difference mapping model, with the working condition features as input. During training, a loss function consisting of a mean squared error term and a physical consistency penalty term is used. The mean squared error term measures the difference between the true diameter difference feature vector and the output of the working condition-to-diameter difference mapping model. The physical consistency penalty term is weighted by physical constraint weights and calculated based on the gradient or sensitivity of the working condition variables. The model is designed to at least satisfy the following conditions: the average diameter difference feature remains monotonically constant with increasing vertical load and monotonically constant with increasing tire pressure. The data-driven residual model is either a gradient boosting decision tree model or a neural network model, and the physical consistency penalty term includes either a monotonicity violation metric or a boundary violation metric.

6. The method as described in claim 1, characterized in that, In step S4, the wear prediction model is trained using the fusion feature vector of the sample as input, and the wear prediction value is output. The fusion feature vector is formed by splicing the diameter difference feature, boundary line feature and friction work feature of the sample. The wear amount is the wear amount or wear index obtained by accumulating the product of the absolute value of the relative sliding velocity in the contact area within the steady-state window and the friction work density on the area and time, and then multiplying it by the wear coefficient. The wear prediction model uses quantile regression to output confidence intervals, where the first quantile and the second quantile are both greater than 0 and less than 1, and the first quantile is less than the second quantile; the lower limit and the upper limit of the confidence interval are the wear prediction values ​​for the corresponding quantiles.

7. The method as described in claim 1, characterized in that, In step S5, the optimal working condition parameter vector that minimizes the predicted wear is solved within the set of working condition constraints, wherein the set of working condition constraints is the feasible region of the working condition; when the predicted wear of the working condition to be evaluated reaches or exceeds the wear threshold, an early warning is output; the optimal working condition parameter vector is solved using either Bayesian optimization or gradient search, and the maximum number of iterations is the upper limit of iteration.

8. A tire wear prediction system based on artificial intelligence, characterized in that, This system is used to implement the method described in any one of claims 1 to 7, comprising: a simulation data management module, a feature extraction module, a model building module, a prediction application module, and a model management module; wherein: The simulation data management module is used to generate a set of working conditions and schedule finite element simulations, storing the diameter difference, slip vector field, friction work density and wear amount; The feature extraction module is used to generate diameter difference features, boundary line features, and friction work features; The model building module is used to train the working condition to diameter difference mapping model and the diameter difference to wear model; The prediction application module is used to calculate predicted wear, solve for the optimal working condition parameter vector, and output early warning based on the wear threshold; The model management module is used for model updates and performance monitoring.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images