A method and system for identifying internal fatigue damage in tires and its application in improving tire fatigue durability.

Abstract

This invention relates to the field of tire performance testing technology, and more particularly to a method and system for identifying internal fatigue damage in tires based on intelligent tire sensors. The method involves acquiring triaxial acceleration data, filtering and normalizing it, and extracting temporal features by locating grounding regions based on extreme values. An adaptive window segmentation based on time intervals is introduced to construct a dual-mode anomaly framework of "extreme value detection + abrupt change detection." This framework, combined with saliency weighting, multiple thresholds, and time-series aggregation constraints, achieves accurate damage identification and temporal localization. PCA is used to screen core features for dimensionality reduction and speed improvement, significantly enhancing robustness and accuracy under complex operating conditions, making it suitable for online monitoring and early warning.

Description

Technical Field

[0001] This invention relates to the field of tire performance testing technology, and in particular to a method and system for identifying internal fatigue damage in tires and its application in improving tire fatigue durability. Background Technology

[0002] With the booming development of the global automotive industry and the rapid increase in urban road traffic density, tires, as the only contact medium between vehicles and the road surface, directly determine the level of driving safety through their structural integrity and performance. Under complex operating conditions, tires inevitably suffer from various forms of progressive damage, such as wear, fatigue cracks, and tread bulges. If these damages are not warned and intervened in a timely manner, they can easily induce sudden tire blowouts, posing a significant threat to traffic safety. Therefore, researching highly reliable intelligent tire damage identification and early warning methods has important theoretical value and engineering application significance for ensuring road traffic safety.

[0003] Traditional tire inspection techniques primarily rely on manual visual inspection and periodic maintenance, which suffers from low efficiency, insufficient real-time performance, and missed early damage. With the deep integration and rapid iteration of sensor technology, information processing technology, and the Internet of Things (IoT), the need for intelligent transportation systems is increasingly urgent. Intelligent tires, as a core technology carrier for vehicle-road interaction perception, have become a research hotspot in academia and industry. Intelligent tire systems integrate sensors within the tire to achieve real-time monitoring of tire status, providing a high-precision and highly reliable data foundation for applications such as vehicle dynamics control and driving safety warnings. As the core sensing unit of the intelligent tire system, the performance of sensors directly determines the system's monitoring accuracy and operational stability. Based on installation methods, intelligent tire sensors can be divided into two main categories: contact and non-contact. Contact sensors are directly attached to the inner wall of the tire, enabling real-time sensing of changes in the tire's dynamic characteristics. Among these, triaxial accelerometers, due to their compact structure, strong impact resistance, and high measurement accuracy, are highly sensitive to vibration signals generated by vehicle dynamics excitation and have become the most widely used contact sensing device.

[0004] As a key component of a vehicle in contact with the road surface, tires inevitably suffer various forms of damage during use. Based on the observability of the damage, tire damage can be divided into two main categories: surface damage and structural damage. Surface damage refers to various defects occurring on the outer surface of the tire that can be directly observed and identified. These mainly include uneven tread wear, physical cuts, punctures caused by foreign objects, sidewall bulges and deformations, and material aging and cracking. Existing identification technologies mainly include manual visual inspection, traditional machine vision technology, and deep learning-based object detection algorithms.

[0005] In the area of ​​surface and near-surface defect detection, the applicant has proposed using tactile perception for the precise detection of tire appearance and near-surface defects. For example, Chinese invention patent CN119935876B discloses a method and system for detecting fine cracks on the tire surface and predicting crack propagation direction based on tactile perception. This method uses a three-dimensional displacement stage combined with a triaxial force sensor to scan line by line. In image processing, Canny, morphological, Sobel, and Hough transforms are combined to extract crack regions, force change directions, and main directions. Based on this, crack propagation trends and danger zone warnings are predicted, enabling high-precision detection and propagation trend assessment of micro-cracks. Another example is Chinese invention patent CN119936322B, which proposes a tactile perception-based method for detecting internal tire air bubble defects. A multi-axis motion mechanism collects X / Y / Z triaxial forces under different normal forces to construct a "multi-force level mechanical fingerprint map," which is then compared with a standard fingerprint database to identify low-stiffness / abnormal deformation areas and estimate air bubble size and severity. This is used for offline screening and assessment of "internal air bubble" type defects. The aforementioned tactile routes can achieve high-resolution detection in static or semi-automated workstations, but they generally rely on external scanning platforms and point-by-point / line-by-line scanning. In engineering applications, they are mostly used for offline detection / rework determination, and their online timing positioning capabilities during actual vehicle operation are limited.

[0006] Indirect methods based on vehicle signals offer advantages such as low cost and simple deployment, but they are significantly affected by load, road conditions, and driving conditions, making it difficult to stably characterize the internal fatigue evolution process and its occurrence time in complex scenarios. For example, the "Intelligent Vehicle Tire Safety Detection System" comprehensively judges by collecting multiple parameters such as temperature, tire pressure, wheel acceleration, and ambient temperature, focusing on comprehensive risk warning rather than time-series localization of weak time-varying characteristics of internal fatigue damage (Chinese Patent CN107379898A). In contrast, in-tire acceleration sensing can directly reflect the interaction between the tire and the road surface and the structural response. Several patents have proposed using in-tire acceleration for abnormal wear / internal fault identification. For example, Chinese Patent CN103068597A calculates the band value of energy in a specified frequency band in the radial acceleration spectrum and compares it with that of a normal tire to determine internal faults. However, parameter selection and threshold settings largely depend on fixed frequency bands / fixed rules, limiting its adaptability to changes in operating conditions. For example, Chinese patent CN105793687A constructs an abnormal wear index by comparing the vibration level of a specific frequency band of in-tire acceleration with a reference frequency band. Its judgment mechanism also relies on preset frequency bands and thresholds, limiting its robustness to multi-source noise and non-stationary operating conditions. Furthermore, some solutions focus on wear estimation and labeling management or statistical quantities such as the ground contact time / amplitude ratio to characterize the tread wear process, but these tend to focus more on the wear state rather than the early online identification and timing of internal fatigue damage (e.g., Chinese patents CN112789182A, CN113506254B, CN112373248B).

[0007] In summary, existing technologies for tire safety monitoring exhibit two prominent limitations: 1. The gap between offline precision detection and online identification: Tactile / visual technologies can achieve high-resolution detection and trend assessment of minute defects under stationary conditions, but relying on scanning mechanisms and contact sensing makes it difficult to meet the real-time online monitoring and temporal positioning requirements during vehicle operation. 2. Limitations of fixed features / single criteria: In-tire acceleration methods often use fixed-length windows and single-band / threshold criteria, which are prone to missed detections and false detections when faced with disturbances in operating conditions such as speed, load, and road surface, and lack a systematic mechanism for quantifying the significance of anomalies and fusing multiple features. Summary of the Invention

[0008] To address the aforementioned technical problems, the present invention aims to provide an intelligent tire internal damage identification method that integrates adaptive time window segmentation, dual-mode anomaly detection, and weighted significance assessment. This method constructs a preprocessing framework for triaxial acceleration signals, employs an adaptive algorithm for dynamic window segmentation, designs a dual-mode anomaly detection mechanism, and establishes a multi-feature-dimensional weighted statistical fusion system to achieve high-precision identification and temporal localization of tire damage. Simultaneously, it utilizes Principal Component Analysis (PCA) to optimize the feature selection strategy, thereby improving the efficiency and accuracy of the identification algorithm.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A method for identifying internal fatigue damage in tires based on intelligent tire sensors, the method comprising the following steps:

[0011] S1. Obtain the output of the sensor installed in the tire liner. The sensor collects information on tire temperature, tire pressure and triaxial acceleration changes throughout the tire's journey from brand new to damaged and failed.

[0012] S2. Based on the collected change information, determine the boundary of the tire ground contact area at the extreme point in each cycle according to the tire rotation cycle, and calculate the multi-dimensional time domain feature vector in each ground contact area.

[0013] S3. Under the assumption of approximate normal distribution, outlier identification is performed on each feature using the margin factor threshold, and abnormal periodic samples are removed using the multi-feature consistency rule.

[0014] S4. Based on the time interval between adjacent periods, and with the dual constraints of the maximum time interval threshold and the minimum sample size threshold, the feature sequence is dynamically divided into windows to obtain several time window sequences.

[0015] S5. Simultaneously execute the following for the window sequence:

[0016] (i) The first anomaly degree is obtained by extreme point detection based on neighborhood comparison;

[0017] (ii) The second anomaly degree is obtained by detecting mutation points based on the statistical differences between adjacent windows;

[0018] Differential weights are applied to the two to form a weighted anomaly score, and a decay function based on saliency ranking is used to weight and fuse each feature dimension to obtain a window-level comprehensive anomaly score.

[0019] S6. Apply the absolute threshold criterion, relative threshold criterion, and minimum significance criterion to the comprehensive anomaly score, and introduce time clustering constraints to suppress occasional misjudgments, and finally output the time period and location result of fatigue damage.

[0020] Preferably, the time-domain characteristics include at least 4-11 of the following 11 items: mean, standard deviation, root mean square, peak value, peak-to-peak value, kurtosis, skewness, peak factor, impulse factor, shape factor, and margin factor, and are calculated periodically according to the grounding region.

[0021] As a preferred approach, a feature selection step is introduced after S2: principal component analysis is used to weight the importance of each feature by feature load × principal component eigenvalue, and the K core features with the highest contribution are selected as model inputs, where K is 4–8.

[0022] Preferably, the boundary of the grounding area is determined by the peak and valley extreme values ​​of the x-axis acceleration during each tire rotation cycle, and the transition segment entering and leaving the grounding area is distinguished from the stable grounding area.

[0023] Preferably, the abnormal feature cleaning uses a margin factor threshold to perform parallel consistency determination on the results of single feature detection: if any feature goes out of bounds within a certain period, the entire set of features for that period is invalidated.

[0024] And / or, the adaptive time window segmentation satisfies the following: when the time interval between adjacent periods exceeds a preset maximum threshold τ, window segmentation is triggered, and the sample size of each segmented window is not less than a preset minimum threshold M; otherwise, the samples are merged into adjacent windows.

[0025] As a preferred option

[0026] a) Extreme point detection generates extreme value significance based on the local neighborhood comparison operator of the window sequence;

[0027] b) Mutation point detection generates mutation significance based on statistical differences in mean / variance between adjacent windows;

[0028] c) The two are linearly combined with weights α and β, where α > β;

[0029] d) The significance-weighted fusion uses a monotonically decaying function based on feature ordering to weight and sum the scores of each feature.

[0030] As a preferred option, multiple determinations include:

[0031] i) Overall anomaly score ≥ absolute threshold T abs ;or

[0032] ii) The proportion threshold ρ where the overall anomaly score is greater than or equal to the global maximum value; and

[0033] iii) The significance level is not lower than the minimum significance threshold T. min ;

[0034] Provided that either i) or ii) and iii) are satisfied, the time clustering constraint must still be satisfied: the number of consecutive anomaly windows ≥ the threshold N for the number of consecutive time clustering windows or the cumulative duration of anomalies ≥ the threshold Δt for the cumulative duration of anomalies. th The event was determined to be a fatigue injury event.

[0035] Furthermore, the present invention also provides the application of the method in tires designed to improve tire fatigue durability.

[0036] Furthermore, the present invention also provides a tire internal fatigue damage identification system, comprising:

[0037] A. Tire temperature, tire pressure and triaxial acceleration sensors installed in the tire liner, or a composite sensing unit integrating tire temperature and tire pressure sensors and a triaxial accelerometer.

[0038] B. Data acquisition and communication unit;

[0039] C. A processing and storage unit on which a program runs to cause the system to execute the method described.

[0040] D. Display / Alarm Unit, used to output the time period of damage occurrence and location results.

[0041] Furthermore, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0042] Furthermore, the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the method.

[0043] This invention, by employing the aforementioned technical solutions, significantly improves the early identification capability and timing accuracy of tire internal fatigue damage: First, the adaptive window dynamically reconstructs the analysis unit according to changes in vehicle speed, load, and road conditions, avoiding aliasing across operating conditions within a fixed window. Combined with a 200 Hz low-pass filter and anomaly feature cleaning, this significantly improves the signal-to-noise ratio and statistical consistency of triaxial acceleration. Second, extreme value detection captures gradual response drift, while abrupt change detection focuses on rapid changes. Significance weighting with a weight α>β simultaneously covers multiple types of damage characterization and suppresses bias from a single indicator. Third, the comprehensive anomaly score simultaneously satisfies the absolute threshold T. abs Relative threshold ρ and lowest significance T min Then, time-based aggregation constraints are applied (number of consecutive windows ≥ N or cumulative duration ≥ Δt). th The system effectively filters out instantaneous sensor disturbances and occasional road surface excitations, significantly reducing false alarms and missed alarms. Fourth, based on PCA feature selection, 11 time-domain features are compressed into 6 core features (the dimensionality is reduced by about 45%). While reducing the computational load and meeting the real-time requirements of vehicle-mounted systems, experiments show that the overall recognition accuracy reaches 96.71%, the detection rate of damaged samples reaches 100%, and the specificity of normal samples is about 93%. It can also provide the abnormal duration range and evolution trend. Compared with the existing fixed frequency band / single threshold method, it achieves a comprehensive leap forward in robustness under complex working conditions, temporal localization, and engineering usability. Attached Figure Description

[0044] Figure 1 Schematic diagram of smart sensor installation.

[0045] Figure 2 Photo of a high-speed durability performance testing machine for heavy-duty tires.

[0046] Figure 3 Data preprocessing flowchart.

[0047] Figure 4 Before and after three-axis acceleration filtering.

[0048] Figure 5 Schematic diagram of the tire contact area.

[0049] Figure 6 Damage recognition model framework diagram.

[0050] Figure 7 Damage identification confusion matrix diagram.

[0051] Figure 8 Image showing the results of time series anomaly detection.

[0052] Figure 9 Image showing the results of sample anomaly detection. Detailed Implementation

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0054] I. Intelligent Tire System

[0055] This application employs an intelligent tire system and a high-speed durability performance testing machine for heavy-duty tires to construct an experimental platform. The intelligent tire system mainly consists of tire temperature and pressure sensors, a triaxial acceleration sensor, an intelligent on-board host, and a display terminal, such as... Figure 1 As shown, the intelligent tire sensor (a composite sensing unit integrating a tire temperature and pressure sensor and a triaxial accelerometer) is installed at the center of the tire liner to measure the longitudinal, lateral, and vertical acceleration components in the x, y, and z directions in the sensor coordinate system in real time.

[0056] like Figure 2 As shown, a tire pre-equipped with a smart tire sensor was mounted on a high-speed durability performance testing machine for heavy-duty tires. Under constant speed and load conditions, acceleration signals were collected at a sampling frequency of 1600Hz. The sensor data was transmitted in real-time to the intelligent vehicle host via Bluetooth module, allowing operators to monitor the tire's operating status via a display terminal. A 12R22.5 HF262 heavy-duty tire was selected as the test object. Tire fatigue durability damage tests were conducted under different vertical load conditions. Specific parameters are shown in Table 1. Tire temperature, tire pressure, and triaxial acceleration changes were collected throughout the entire process from the tire's new condition to damage and failure.

[0057] Table 1 Tire fatigue damage test conditions

[0058]

[0059] II. Research Methods

[0060] 2.1 Data Preprocessing

[0061] Data preprocessing is a crucial step in converting raw sensor data into standardized feature vectors suitable for machine learning algorithms, directly impacting the effectiveness of subsequent model training and recognition accuracy. The core objectives of data preprocessing include dimensionality reduction optimization, relational smoothing, data standardization, noise suppression, and feature extraction, such as... Figure 3 As shown, this application will elaborate on the main preprocessing process in light of the characteristics of triaxial acceleration signals in tire damage identification.

[0062] 2.1.1 Filtering and Noise Reduction

[0063] For the raw triaxial acceleration signals acquired by sensors in the intelligent tire system, this application sets the sampling frequency to 1600Hz. This sampling frequency is much higher than the tire's fundamental rotation frequency and its main harmonic components, which can fully capture the dynamic characteristic changes caused by internal tire damage and meet the accuracy requirements for tire damage identification. Considering that the spectral characteristics of tire vibration signals are mainly concentrated in the low-frequency band, while the high-frequency part often contains sensor electronic noise, random environmental disturbances, and interference signals introduced by the data acquisition system, this application uses a fourth-order Butterworth low-pass filter with a cutoff frequency of 200Hz to filter and reduce noise in the triaxial acceleration signals. The processing effect is as follows: Figure 4 As shown, after filtering and noise reduction, the signal-to-noise ratio of the triaxial acceleration signal is significantly improved, laying a good data foundation for subsequent feature extraction and damage identification.

[0064] Analysis of the characteristics of the triaxial acceleration signal revealed that the x-axis acceleration signal is distributed along the tire rolling direction, which can more directly reflect the dynamic changes in the tire-road contact process and has a stronger ability to distinguish internal structural damage. Therefore, this application selects the x-axis acceleration signal as the main data source for damage identification.

[0065] 2.1.2 Determine the grounding area

[0066] As the triaxial accelerometer moves into and out of the contact patch with the tire, distinct peak and trough characteristics are observed in the x-axis acceleration signal. This phenomenon originates from the dynamic deformation process of the tire structure. Specifically, as the sensor approaches the contact patch, the tire body undergoes pre-stretch deformation, which then transitions to compression deformation upon contact with the road surface. After entering the contact patch, the tire's deformation tends to stabilize under a constant load. When the sensor leaves the contact patch, the tire body undergoes a reverse deformation process from compression to stretch, forming a vibration response mode symmetrical to that upon entry.

[0067] like Figure 5 As shown, the BD rolling section of the tire corresponds to the grounding area of ​​the sensor, and the AE rolling section also includes the transition phase where the sensor is about to enter and leave the grounding area. Based on the above grounding characteristics, this application determines the boundary of the sensor's grounding area by identifying the extreme points (peak and valley values) of the x-axis acceleration signal in each rotation cycle in real time, providing a reliable basis for area division for subsequent damage feature extraction and analysis.

[0068] 2.1.3 Data Normalization

[0069] This application employs the min-max normalization method to standardize the acceleration data. Min-max normalization is a classic linear transformation method that maps the original data to the interval [0, 1] while accurately preserving the relative relationships and distribution characteristics between data points. It effectively eliminates the influence of different units and numerical ranges on subsequent machine learning algorithms. Its mathematical expression is as follows:

[0070] (1),

[0071] in, It is the measured acceleration value. and These are the minimum and maximum values ​​in the dataset, respectively. This is the normalized value.

[0072] 2.1.4 Temporal Feature Extraction

[0073] This application first accurately identifies the grounding area boundary of the sensor in each rotation cycle. Then, it calculates 11 time-domain feature parameters for the x-axis acceleration signal within each grounding area. Finally, the extracted feature vectors are input into a machine learning model to achieve the identification and prediction of internal tire damage. Time-domain features can intuitively reflect key information such as the amplitude distribution, fluctuation degree, and morphological changes of the signal, exhibiting good sensitivity and interpretability for tire internal structural damage. The 11 extracted time-domain feature parameters and their mathematical definitions are shown in Table 2.

[0074] Table 2 Time-domain characteristic variables

[0075]

[0076] 2.1.5 Abnormal Feature Cleaning

[0077] In actual tire operating environments, intelligent sensors are inevitably affected by a combination of factors, including inherent performance limitations such as temperature drift, nonlinear error, and quantization noise, as well as external environmental disturbances such as changes in road surface excitation, electromagnetic interference, and mechanical vibration coupling. These interference factors can cause outliers in the time-domain feature data collected by the sensor within the grounding area, severely affecting the statistical characteristics of the feature data. Directly inputting features containing outliers into the damage identification model will not only reduce the model's generalization ability and recognition accuracy but may also lead to misjudgments and missed detections, affecting the reliability of the tire safety monitoring system. Therefore, before model training, outlier detection and cleaning must be performed on the extracted time-domain damage features. Given that the time-domain features of tire vibration signals typically follow an approximately normal distribution under normal operating conditions, this application adopts the margin factor theory based on the normal distribution assumption to establish outlier identification and removal criteria, ensuring that the feature data input to the model has good statistical consistency. Its mathematical expression is as follows:

[0078] (2),

[0079] (3),

[0080] (4),

[0081] in, It is the first Characteristic values ​​of each grounding cycle N It is the total number of sensor grounding cycles. It is the mean of this feature. It is the mean squared error of this feature. It is the margin factor of this feature.

[0082] Tire internal damage, as a persistent structural defect, often causes changes in vibration characteristics that simultaneously affect multiple time-domain feature parameters. This necessitates the collaborative analysis and comprehensive judgment of these parameters. Abnormal fluctuations in a single feature may originate from momentary sensor malfunctions or incidental factors such as local road unevenness, rather than actual damage to the tire structure. This application designs a multi-feature joint screening strategy, requiring data points to meet the margin factor criterion across all extracted time-domain features to be considered valid data. Conversely, if any data point exhibits an anomaly in any feature dimension, the entire set of feature data within that rotation cycle is marked as an anomaly and discarded. This rigorous multi-feature consistency verification mechanism effectively improves the accuracy and robustness of outlier detection, ensuring the high reliability and representativeness of the feature data input to the model.

[0083] 2.2 Internal Fatigue Damage Identification Model

[0084] Traditional tire damage detection methods often rely on fixed analysis windows and single anomaly criteria, making them ill-suited for the complex, dynamically changing scenarios in real-world applications. This leads to both false positives and false negatives. Therefore, this application proposes a fusion recognition architecture integrating adaptive window segmentation and dual-mode anomaly detection, such as... Figure 5 As shown, this model dynamically adjusts the analysis window based on the inherent temporal characteristics of the data through adaptive time window segmentation technology. It employs a dual-mode anomaly identification algorithm combining extreme value detection and mutation detection to comprehensively capture different types of damage characteristics. Furthermore, it establishes multiple judgment criteria and a weighted evaluation mechanism to achieve accurate identification and risk quantification of tire damage.

[0085] 2.2.1 Adaptive Time Window Segmentation

[0086] Traditional tire damage detection methods generally use fixed-length time windows for data analysis. However, this method has significant limitations. Fixed windows cannot adapt to the complexity of changes in operating conditions during actual vehicle operation. This may lead to data features across different operating conditions being forcibly merged into the same analysis unit, thereby introducing additional noise interference and reducing the accuracy of damage feature identification. Therefore, this application proposes a dynamic window segmentation method based on time interval adaptation.

[0087] The core of this window segmentation method is to automatically identify natural breakpoints in the feature set based on the time interval characteristics during data acquisition and use these breakpoints as the basis for window segmentation. Let the time series be... The time interval between adjacent periods is defined as:

[0088] (5),

[0089] The algorithm employs a dual constraint mechanism. First, a maximum time interval threshold is set. Only when the time interval between two consecutive data feature points exceeds this threshold is it considered that there is a change in operating conditions or an interruption in data acquisition, triggering a window segmentation operation. The time window segmentation condition can be expressed as:

[0090] (6),

[0091] in, It is the first k A time window, and These are the start and end times of the window, respectively.

[0092] Secondly, a minimum sample size constraint is set to ensure that each segmentation window contains sufficient data samples to guarantee the reliability of the statistical analysis. Only windows that meet the minimum sample size requirement are considered valid analysis units. The window validity constraint conditions are as follows:

[0093] (7),

[0094] in, This represents the number of samples within that time window. It is the preset minimum sample size threshold.

[0095] This adaptive segmentation strategy can adapt to the data distribution characteristics under different working conditions, thereby improving the accuracy and robustness of damage identification.

[0096] 2.2.2 Dual-mode anomaly detection algorithm

[0097] Considering the diverse manifestations of internal tire damage, a single anomaly detection mode is insufficient to comprehensively capture all types of damage features. This application designs a dual-mode anomaly recognition algorithm that integrates extreme value detection and abrupt change detection. The extreme value detection module specifically identifies local maxima and minima in the signal features; these feature points typically correspond to abnormal vibration responses at specific locations on the tire. The algorithm determines extreme values ​​based on a local comparison criterion by analyzing the numerical relationship between each data point and its neighboring points in the window sequence. Let the window statistics of the feature sequence be... The extreme point detection operator is defined as:

[0098] (8),

[0099] (9),

[0100] in, It is an indicator function. It is an extremum significance function.

[0101] The mutation point detection module identifies abrupt changes in signal characteristics based on the statistical characteristics of differences between adjacent windows. This module first calculates the difference values ​​of all adjacent windows and establishes a statistical model of the difference distribution, and then uses a statistical threshold to determine mutation points.

[0102] (10)

[0103] (11),

[0104] in , It is the statistical threshold of the model. It is the mathematical expectation function. Var It is a variance function.

[0105] Considering the different importance of different types of anomalies in damage detection, the algorithm adopts a differentiated weighting strategy. For extreme points that reflect gradual changes, lower weights are assigned to avoid normal operating condition changes being misjudged as damage; for abrupt change points that reflect sudden changes, higher weights are assigned to enhance the detection sensitivity of acute damage.

[0106] (12),

[0107] in and It is a differential weighting coefficient, and .

[0108] Traditional anomaly detection methods typically employ simple counting to determine the frequency of anomalies. This approach ignores differences in anomaly severity, potentially leading to an overestimation of the cumulative effect of multiple minor anomalies while underestimating the importance of a single severe anomaly. This application proposes a weighted statistical evaluation mechanism based on anomaly significance, achieving accurate tire damage risk assessment by quantifying anomaly severity.

[0109] (13)

[0110] in For the first j The saliency ranking function within column features, It is the attenuation parameter.

[0111] To improve the reliability of tire damage assessment, the algorithm employs a multi-criteria system: (1) Absolute threshold criterion: When the weighted anomaly score of a window exceeds a preset maximum absolute threshold, it is directly identified as a potential damage point; (2) Relative threshold criterion: When the weighted anomaly score of a window reaches a preset proportion of the highest global score, the window is identified as a potential damage point; (3) Maximum significance criterion: Only when the weighted anomaly score of a window exceeds a minimum significance threshold is it included in the candidate damage point set. Through the synergistic effect of these multiple criteria, the algorithm effectively balances detection sensitivity and false alarm rate, achieving reliable tire damage identification.

[0112] III. Results Analysis

[0113] 3.1 Feature Contribution Analysis

[0114] To deeply analyze the contribution of each feature variable to the performance of the tire damage identification model, this application uses principal component analysis (PCA) to quantify and rank the importance of 11 time-domain features in the input model. PCA maps the original feature space to a low-dimensional principal component space through linear transformation. The feature loading coefficients on the principal components can effectively reflect their contribution to data variability, thereby quantifying the importance of the features. The formula for calculating feature importance is as follows:

[0115] (14)

[0116] in, Features Importance score Features In the Loading coefficients on each principal component For the first The eigenvalues ​​of the principal components The number of principal components selected.

[0117] To ensure the reliability and stability of the analysis results, this application calculated the PCA importance score of each feature variable through multiple independent experiments and performed statistical analysis to eliminate the influence of random factors. The feature importance analysis results are shown in Table 3, which displays the influence weight of each feature on the damage identification model in descending order of contribution. The analysis found that the cumulative contribution of the six core features, namely mean, margin factor, impulse factor, peak factor, standard deviation, and shape factor, reached 65.81%, indicating that these features play a dominant role in the tire damage identification decision-making process and are the key information source for the model to accurately determine the damage state. This result provides an important theoretical basis for subsequent feature selection optimization and lightweight model design.

[0118] Table 3. Ranking of Feature Importance

[0119]

[0120] Although the aforementioned analysis involves numerous feature variables, the PCA importance ranking results show that the number of features that truly play a major role in identifying tire internal damage is limited, indicating significant feature redundancy. Excessive redundant features not only increase computational complexity but may also introduce noise interference, affecting the model's generalization performance. To verify the effectiveness of high-contribution features and optimize feature configuration, this application designed a comparative experiment based on combinations of the top 8 features by contribution.

[0121] Table 4. Accuracy of different feature combinations

[0122]

[0123] By configuring different combinations of the top-ranked features, multiple feature subsets were constructed and damage recognition models were trained separately. The damage recognition performance of each combination scheme was systematically evaluated. The comparison results of the recognition accuracy of different feature combinations are shown in Table 4. Experimental results show that combination 1, which selects the top 6 features (X1~X6) in terms of contribution as model input, achieves a damage recognition accuracy of 96.71%, significantly better than other feature combination schemes. The recognition accuracy of combinations 2, 3, and 4 is lower than that of combination 1. The experimental results fully demonstrate that, while maintaining the same number of features, prioritizing the selection of features with higher contribution to tire internal damage recognition as model input can significantly improve the recognition accuracy, providing a scientific basis for feature dimensionality reduction and model optimization.

[0124] 3.2 Performance Analysis of the Tire Damage Recognition Model

[0125] The performance of the tire fatigue damage identification model on the validation set was quantitatively evaluated using a confusion matrix, such as... Figure 7 As shown, in the normal tire sample identification task, the model successfully classified 265 normal samples correctly, with only 20 normal samples misclassified, achieving a specificity of 93.0%. In abnormal tire sample detection, the model demonstrated excellent sensitivity, achieving 100% accurate identification of 323 damaged samples without any missed detections. The model's overall accuracy reached 96.71%, validating the effectiveness of the constructed damage identification algorithm.

[0126] Figure 8 This study demonstrates the performance of the tire damage identification model in time-series anomaly detection. The results show that the model can accurately capture abnormal time periods during tire operation and, through a scientifically sound threshold mechanism, identify areas of abnormal fluctuation in the data. Under normal operating conditions, the anomaly score remains consistently low, exhibiting good baseline characteristics, with values ​​generally remaining stable below 1.127. When internal tire damage occurs, the model can sensitively detect a significant jump in the anomaly score to a high level of 7-8, maintaining this high level throughout the abnormal period; the corresponding time period is marked as the anomaly area. This continuous monitoring capability has significant advantages over traditional discrete detection methods, enabling not only real-time status assessment but also capturing the evolution of damage, providing data support for early warning of tire damage.

[0127] Figure 9The single-sample anomaly detection results shown further reveal the model's refined discrimination capability at the sample level. Analysis revealed that anomaly scores for normal samples generally remained at low levels and were relatively concentrated, with most samples maintaining anomaly scores in the low range of 0-1.5. In contrast, anomalous samples exhibited significantly high anomaly scores, demonstrating a clear distinction between the two types of samples. Particularly noteworthy is that in the latter half of the sample sequence, starting from approximately the 5000th sample, the model detected a large number of high-anomaly-score samples exhibiting a clear clustering pattern. Anomaly scores were concentrated in the high-score range of 15-25. This gradual change in anomaly scores suggests that tire damage may involve a cumulative evolutionary process; early tire damage only manifests as slight changes in vibration characteristics, while as the damage intensifies, the anomaly features gradually become more pronounced.

[0128] It is worth noting that a small number of high-scoring anomalous samples were also observed in normal areas. This is a misjudgment by the model, which may be caused by factors such as transient sensor interference, changes in road conditions, or abnormal data acquisition. To effectively reduce the impact of such misjudgments on the overall damage identification performance, the model adopts a temporal clustering constraint strategy, setting a threshold for sample anomalous time. Only when the cumulative clustering time of anomalous samples exceeds this threshold will the model classify it as a true anomalous area. This method can effectively distinguish between real damage signals and occasional interference, suppress the negative impact of individual sample misjudgments on model accuracy, and significantly improve the robustness of the system.

[0129] This application proposes an intelligent tire damage identification method based on a triaxial accelerometer, focusing on the identification of internal tire damage. The method employs an adaptive time window segmentation and a dual-mode anomaly detection fusion algorithm to process tire vibration signals, and utilizes principal component analysis to explore the impact of different feature variables on the performance of the damage identification model. Specific conclusions are as follows:

[0130] (1) To address the problem that traditional fixed-length windows cannot adapt to the dynamic changes in vehicle operating conditions, this application proposes a dynamic window segmentation method based on time interval adaptation, which effectively overcomes the noise interference caused by feature merging across operating conditions. At the same time, a dual-mode anomaly recognition algorithm that integrates extreme point detection and mutation point detection is designed. Through a differentiated weighting strategy and a weighted statistical evaluation mechanism based on anomaly significance, a multi-judgment criterion system is established, which significantly improves the accuracy and reliability of tire damage detection.

[0131] (2) Using PCA importance analysis, the contribution of 11 time-domain features to the damage identification model was systematically evaluated, and the influence weight of each feature variable in the model decision-making process was quantified, providing a scientific basis for feature selection and model optimization. The analysis results show that the cumulative contribution of the six core features, including the mean, margin factor, and impulse factor, reached 65.81%, playing a dominant role in tire damage identification. Furthermore, the identification model built based on these six high-contribution features achieved an accuracy of 96.71%, fully verifying the effectiveness of feature selection based on PCA.

[0132] (3) The tire damage identification model constructed in this application exhibits excellent performance on the validation set. The confusion matrix results show that the model correctly identified 265 normal samples with a misclassification rate of only 7.0%, while achieving 100% accurate identification of 323 abnormal samples, avoiding any missed detections. The time series anomaly detection and single sample anomaly analysis results further demonstrate that the model has the ability to accurately capture abnormal time periods during tire operation, providing a possibility for the transformation from traditional periodic inspections to intelligent real-time monitoring.

[0133] 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 in this application 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 in this application, but is to be accorded the widest scope consistent with the principles and novelty disclosed in this application.

[0134] 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.

[0135] 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.

[0136] 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.

[0137] 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.

[0138] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0139] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0140] 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 method for identifying internal fatigue damage in tires based on intelligent tire sensors, characterized in that, include: S1. Obtain the output of the sensor installed in the tire liner. The sensor collects information on tire temperature, tire pressure and triaxial acceleration changes throughout the tire's journey from brand new to damaged and failed. S2. Based on the collected change information, determine the boundary of the tire ground contact area at the extreme point in each cycle according to the tire rotation cycle, and calculate the multi-dimensional time domain feature vector in each ground contact area. S3. Under the assumption of approximate normal distribution, outlier identification is performed on each feature using the margin factor threshold, and abnormal periodic samples are removed using the multi-feature consistency rule. S4. Based on the time interval between adjacent periods, and with the dual constraints of the maximum time interval threshold and the minimum sample size threshold, the feature sequence is dynamically divided into windows to obtain several time window sequences. S5. Simultaneously execute the following for the window sequence: (i) The first anomaly degree is obtained by extreme point detection based on neighborhood comparison; (ii) The second anomaly degree is obtained by detecting mutation points based on the statistical differences between adjacent windows; Differential weights are applied to the two to form a weighted anomaly score, and a decay function based on saliency ranking is used to weight and fuse each feature dimension to obtain a window-level comprehensive anomaly score. S6. Apply the absolute threshold criterion, relative threshold criterion, and minimum significance criterion to the comprehensive anomaly score, and introduce time clustering constraints to suppress occasional misjudgments, and finally output the time period and location result of fatigue damage.

2. The method according to claim 1, characterized in that, The time-domain characteristics include at least 4 to 11 of the following 11 items: mean, standard deviation, root mean square, peak value, peak-to-peak value, kurtosis, skewness, peak factor, impulse factor, shape factor, and margin factor, and are calculated periodically according to the grounding region.

3. The method according to claim 1, characterized in that, Following S2, a feature selection step is introduced: Principal component analysis is used to weight the importance of each feature by feature loading × principal component eigenvalue, and the K core features with the highest contribution are selected as model inputs, where K is 4–8.

4. The method according to claim 1, characterized in that, The boundary of the grounding area is determined by the peak and valley extreme values ​​of the x-axis acceleration during each tire rotation cycle, and the transition section entering and leaving the grounding area is distinguished from the stable grounding area. And / or, abnormal feature cleaning uses a margin factor threshold to perform parallel consistency determination on the results of single feature detection: if any feature goes out of bounds in a certain period, the entire set of features in that period is invalidated. And / or, adaptive time window segmentation satisfies the following: when the time interval between adjacent periods exceeds the preset maximum threshold τ, window segmentation is triggered, and the sample size of each segmented window is not less than the preset minimum threshold M; otherwise, the samples are merged into adjacent windows.

5. The method according to claim 1, characterized in that, a) Extreme point detection generates extreme value significance based on the local neighborhood comparison operator of the window sequence; b) Mutation point detection generates mutation significance based on statistical differences in mean / variance between adjacent windows; c) The two are linearly combined with weights α and β, where α > β; d) The significance-weighted fusion uses a monotonically decaying function based on feature ordering to weight and sum the scores of each feature.

6. The method according to claim 1, characterized in that, Multiple decisions include: i) Overall anomaly score ≥ absolute threshold T abs ;or ii) The proportion threshold ρ where the overall anomaly score is greater than or equal to the global maximum value; and iii) The significance level is not lower than the minimum significance threshold T. min ; Provided that either i) or ii) and iii) are satisfied, the time clustering constraint must still be satisfied: the number of consecutive anomaly windows ≥ the threshold N for the number of consecutive time clustering windows or the cumulative duration of anomalies ≥ the threshold Δt for the cumulative duration of anomalies. th The event was determined to be a fatigue injury event.

7. The method of any one of claims 1-6 is used in tires designed to improve tire fatigue durability.

8. A tire internal fatigue damage identification system, characterized in that, include: A. Tire temperature, tire pressure and triaxial acceleration sensors installed in the tire liner, or a composite sensing unit integrating tire temperature and tire pressure sensors and a triaxial accelerometer; B. Data acquisition and communication unit; C. A processing and storage unit, on which a program runs causes the system to perform the method described in any one of claims 1-6; D. Display / Alarm Unit, used to output the time period of damage occurrence and location results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method of any one of claims 1–6.

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 method of any one of claims 1–6.

Images