Rail soft wheel multi-sensor health detection method and system

Abstract

This invention discloses a multi-sensor health detection method and system for soft rail wheels, belonging to the field of wheel inspection technology. It includes the following steps: S1, real-time acquisition of wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information, followed by preprocessing; S2, extraction of penetration attenuation characteristics reflecting the internal propagation properties of the material, and formation of a robust attenuation characterization; S3, analysis of ultrasonic echo reflection characteristics and spectral structure stability, comprehensive evaluation of the structural properties of internal anomalies, and determination of the anomaly response type and its spatial distribution; S4, analysis of surface degradation trends and internal evolution behavior for lifespan constraint assessment, outputting maintenance warnings and safety decision guidance. This invention solves the problems of traditional ultrasonic testing's difficulty in distinguishing between structural cracks and non-structural internal defects, and the instability of early crack evolution criteria, leading to misjudgments in health assessments and inaccurate maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of wheel inspection technology, specifically to a multi-sensor health detection method and system for soft rail wheels. Background Technology

[0002] In the field of rail transit equipment operation and maintenance, laser inspection and ultrasonic inspection have been gradually applied to wheel condition monitoring. Laser measurement technology can acquire information on wheel diameter changes and surface morphology non-contactly, making it suitable for online detection of wear and geometric conditions. Ultrasonic inspection technology can be used to obtain internal echo information of materials, showing a certain sensitivity to internal damage and structural anomalies. With the enrichment of inspection methods, how to coordinate various inspection information during operation to achieve unified perception and comprehensive assessment of wheel surface degradation and internal evolution has become a crucial fundamental issue in wheel health monitoring and life management research.

[0003] For example, invention patent CN118464389A discloses a testing device and method for testing the external lighting brightness of rail vehicles, including: a lighting test plate; the lighting test plate is connected to the upper part of a support via an adjusting slide and a cylinder; the cylinder is used to raise and / or lower the lighting test plate along the adjusting slide; a single-point solid-state lidar is installed at the center point of the track at the lower part of the support, and the single-point solid-state lidar is used to measure the distance between the lighting test plate and the vehicle. The rail vehicle external lighting brightness testing device of this invention has flexible track movement capability, can adapt to testing needs at different locations, and is suitable for testing various vehicle models. By uniformly distributing a light sensor array and precisely amplifying the data, the accuracy of the test data is ensured.

[0004] For example, invention patent CN109752170B discloses a car headlight detector that uses wheel location for alignment measurement. This invention primarily addresses the numerous defects and shortcomings in existing headlight detectors during the headlight location and alignment process, leading to misjudgments in test results. The device includes a base that moves along a headlight sensor guide rail; a laser rangefinder mounted on the side wall of the base for measuring the distance from the base to the wheel; a column mounted on the base; a head unit fitted onto the column and movable up and down along it for image acquisition of the car headlights; a control box mounted on the top of the column; and a display screen fitted into the control box for displaying test information. Through this solution, the invention achieves accurate detection and possesses high practical and promotional value.

[0005] However, soft-walled wheels may contain internal defects such as material bubbles, localized density inhomogeneities, or early microcracks during manufacturing and use. These defects can cause echo amplitude attenuation or waveform distortion during ultrasonic testing. Traditional ultrasonic testing methods struggle to distinguish between structural cracks and non-structural defects, easily misjudging bubbles with minimal safety impact as cracks, or lacking reliable criteria for early crack evolution stages, thus affecting the accuracy of health assessments and the rationality of maintenance decisions.

[0006] Therefore, in order to address the above problems, there is an urgent need for a multi-sensor health detection method and system for soft rail wheels. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a multi-sensor health detection method and system for soft rail wheels, which solves the problems of traditional ultrasonic testing's inability to distinguish between structural cracks and non-structural internal defects, and the instability of early crack evolution criteria, leading to misjudgments in health assessments and inaccurate maintenance decisions.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: a multi-sensor health detection method for soft rail wheels, comprising the following steps: S1, real-time acquisition of wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information; preprocessing of multi-source data to construct a wheel health feature dataset indexed by the turning angle; S2, based on dual-time-window ultrasonic echo energy comparison, extraction of penetration attenuation characteristics reflecting the internal propagation characteristics of the material, and formation of robust attenuation characterization through time stabilization processing; S3, based on ultrasonic echo analysis of ultrasonic echo reflection characteristics and spectral structure stability, comprehensive evaluation of the structural properties of internal anomalies at the turning angle, and determination of the anomaly response type and its spatial distribution state based on the structural properties of the internal anomalies; S4, for anomaly locations, analysis of surface degradation trends and internal evolution behavior using the wheel health feature dataset for life constraint assessment, and output of maintenance warnings and safety decision guidance based on the life constraint assessment.

[0011] Furthermore, the specific process of real-time acquisition of wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information, and preprocessing of multi-source data to construct a wheel health feature dataset indexed by rotation angle is as follows: A line-scan laser camera is fixedly installed on the side of the track to continuously scan the wheel-track contact area, acquiring wheel diameter distance and surface morphology height; an ultrasonic probe is installed on one side of the wheel's running path, emitting ultrasonic pulses into the wheel's interior according to an emission cycle synchronized with the wheel's rotational speed, acquiring the corresponding ultrasonic A-scan echo time-domain waveform; simultaneously, the wheel angular velocity is acquired in real time, and the ultrasonic propagation velocity in the wheel material is obtained; using the wheel angular velocity as a reference, the wheel diameter distance data and surface morphology height data are mapped from time to rotation angle, and the data at different sampling times are uniformly converted into a sequence indexed by the wheel rotation angle, and then... The sampling time of the ultrasonic A-scan echo time-domain waveform is synchronously corrected according to the wheel angular velocity; median filtering is performed on the wheel diameter distance data and surface topography height data to suppress local anomalies caused by track vibration and optical measurement noise; bandpass filtering is performed on the ultrasonic A-scan echo time-domain waveform to remove low-frequency mechanical disturbances and high-frequency noise components, and envelope extraction is performed on the ultrasonic A-scan echo time-domain waveform after filtering to obtain the ultrasonic envelope signal; time window smoothing is performed on the wheel angular velocity; the preprocessed wheel diameter distance, surface topography height, ultrasonic envelope signal, wheel angular velocity, and ultrasonic propagation speed are combined to construct wheel health feature data, and normalization processing is performed on the wheel health feature data respectively to establish a wheel health detection database. The original and preprocessed wheel health feature data are written into the wheel health detection database.

[0012] Furthermore, based on the comparison of ultrasonic echo energy in dual time windows, the specific process of extracting penetration attenuation characteristics reflecting the internal propagation properties of materials and forming robust attenuation characterization through time stabilization is as follows: For the ultrasonic envelope signal corresponding to each corner position, an incident interface time window and a deep response time window are constructed. The incident interface time window corresponds to the echo interval formed by the ultrasonic wave from the probe to the contact interface on the wheel surface, and the deep response time window corresponds to the echo interval formed by the ultrasonic wave after propagation inside the wheel and in the hub direction. The ultrasonic envelope signals in the incident interface time window and the deep response time window are integrated to calculate the ultrasonic energy of the incident interface time window and the ultrasonic energy of the deep response time window. The ratio of the ultrasonic energy of the incident interface time window to the ultrasonic energy of the deep response time window is calculated, and the natural logarithm of the ratio is taken to obtain the normalized penetration attenuation value. Based on the sliding time window, the median of the normalized penetration attenuation value at the same corner position is taken to obtain the stable normalized penetration attenuation characteristic value. The stable normalized penetration attenuation characteristic value is associated with the wheel health characteristic data according to the corner index and written into the wheel health detection database.

[0013] Furthermore, the specific process of analyzing the ultrasonic echo reflection characteristics and spectral structure stability based on ultrasonic echo is as follows: Within the deep response time window, the ultrasonic envelope signal at the corresponding corner position is extracted, and the median of the ultrasonic envelope signal amplitude is taken as the noise baseline value; simultaneously, a local maximum search is performed on the ultrasonic envelope signal within the deep response time window, and the echo peak with the largest amplitude is selected from the echo peaks whose ultrasonic envelope signal amplitude is higher than the noise baseline value, as the significant reflection response value of the deep echo at the current corner position; the ratio of the significant reflection response value to the noise baseline value is calculated to obtain the significant reflection response value of the newly added echo peak. The time-domain waveform of the ultrasonic A-scan echo at the same turning angle is subjected to a short-time Fourier transform to obtain the time spectrum. Within the corresponding deep response time window, the energy of each frequency component in the time spectrum is integrated to obtain the contribution energy value of each frequency component. The contribution energy values ​​of all frequency components are summed to obtain the total contribution energy value. The ratio of the contribution energy value of each frequency component to the total contribution energy value is calculated to obtain the spectral energy probability of each frequency component. The spectral energy probability of each frequency component is multiplied by its own logarithm, and the product of all frequency components is summed and the negative number is taken to obtain the spectral entropy.

[0014] Furthermore, the specific process for comprehensively evaluating the structural attributes of internal anomalies at corner positions is as follows: The historical spectral entropy sequence corresponding to the same corner position is read from the wheel health detection database, and the median is taken as the spectral entropy reference value; the logarithm of the peak significance value is taken to obtain the temporal reflection enhancement; the negative of the temporal reflection enhancement is used as the exponent for natural exponentiation; the result of the natural exponentiation is added by one and the reciprocal is taken to obtain the temporal structural response value; the current spectral entropy is subtracted from the spectral entropy reference value to obtain the frequency domain structural deviation; the negative of the frequency domain structural deviation is used as the exponent for natural exponentiation; the result of the natural exponentiation is added by one and the reciprocal is taken to obtain the frequency domain structural stability response value; the temporal structural response value and the frequency domain structural stability response value are multiplied to obtain the structural defect confidence value.

[0015] Furthermore, the specific process for determining the type of abnormal response and its spatial distribution based on the structural attributes of internal anomalies is as follows: Based on a sliding time window, the median of the confidence values ​​of structural defects at the same corner position is taken to obtain the confidence value of stable structural defects; along the wheel corner direction, the confidence values ​​of stable structural defects at the N adjacent corner positions are counted, and it is determined whether they are simultaneously greater than the discrimination threshold. If so, it is considered that there is a continuous structural defect response inside the wheel corresponding to the corner interval; when the number of corner positions where the confidence value of stable structural defects exceeds the discrimination threshold is less than the judgment number threshold, it is considered that there is a discontinuous structural defect response; when the confidence values ​​of stable structural defects corresponding to all corner positions do not exceed the discrimination threshold, it is considered that no structural defect response has been detected at present; the corresponding wheel health feature data and stable penetration normalized attenuation feature value are acquired simultaneously to perform a typological determination of the wheel abnormal response.

[0016] Furthermore, the specific process for categorizing wheel abnormal responses by simultaneously acquiring corresponding wheel health characteristic data and stable penetration normalized attenuation characteristic values ​​is as follows: When there is a continuous structural defect response, if the change in wheel diameter distance and the change in surface morphology height at adjacent sampling points at the corresponding corner position do not exceed the corresponding abnormal change threshold, the abnormality at the corner position is determined to be an internal structural defect-dominated abnormality; otherwise, it is a structural-surface composite abnormality. When there is a discontinuous structural defect response, if the difference between the stable penetration normalized attenuation characteristic value and the median of historical stable penetration normalized attenuation characteristic values ​​exceeds the deviation threshold, the abnormality at the corner position is determined to be a non-structural defect abnormality; otherwise, it is determined to be no abnormality. When no structural defect response is detected, if the corresponding wheel diameter distance change, surface morphology height change, and stable penetration normalized attenuation characteristic value are all within the corresponding normal distribution range, it is determined to be no abnormality; otherwise, it is a non-structural defect abnormality. The stable penetration normalized attenuation characteristic value and the corresponding abnormality determination result are written into the wheel health detection database.

[0017] Furthermore, for abnormal locations, the specific process of using wheel health feature datasets to analyze surface degradation trends and internal evolution behavior for life constraint assessment is as follows: For abnormal corner locations, the corresponding wheel diameter distance is obtained, and the rate of change of the wheel diameter distance is calculated using the adjacent sampling point difference method to obtain the surface wear degradation rate; simultaneously, within the incident interface time window at the corresponding corner location, a local maximum search is performed on the ultrasonic envelope signal, and the time corresponding to the echo peak with the largest amplitude is selected as the reference echo time; within the deep response time window, a local maximum search is performed on the ultrasonic envelope signal, and the time corresponding to the echo peak with the largest amplitude is selected as the deep echo time; the difference between the deep echo time and the reference echo time is calculated to obtain the ultrasonic propagation delay. The equivalent depth of the internal defect is obtained by multiplying the ultrasonic propagation delay by the ultrasonic propagation speed and then dividing by two. The rate of change of the equivalent depth of the internal defect is calculated using the difference method between adjacent sampling points, which is taken as the internal defect propagation rate. The remaining life of the surface wear is obtained by dividing the difference between the wheel diameter distance limit and the current wheel diameter distance by the sum of the surface wear degradation rate and the smallest positive number. The remaining life of the internal defect is obtained by dividing the difference between the equivalent depth of the internal defect and the current internal defect equivalent depth by the sum of the internal defect propagation rate and the smallest positive number. The reciprocals of the remaining life of the surface wear and the remaining life of the internal defect are taken respectively, and the two reciprocals are added together to obtain the comprehensive degradation rate. The reciprocal of the comprehensive degradation rate is taken and multiplied by a constant of two to obtain the fusion remaining life assessment value.

[0018] Further, the specific process of evaluating the output for maintenance warning and safety decision guidance according to the life constraint is as follows: associate the fused remaining life assessment value with the corresponding abnormal type according to the corner index and write it into the wheel health detection database; if there is more than one abnormal corner position existing in the wheel currently, select the corner position with the minimum fused remaining life assessment value for evaluation; the fused remaining life assessment value is compared with the multi-level life assessment thresholds R1 and R2: when ≥R2, it is determined that the corresponding corner position is in the normal operation state, and only routine data monitoring is maintained; when R1≤ <R2, it is determined that the corresponding corner position is in the warning maintenance state, a maintenance warning flag is generated, and a maintenance prompt is pushed; when <R1, it is determined that the corresponding corner position is in the safety restricted state, a deactivation suggestion flag is generated, and a wheel replacement prompt is pushed; write the fused remaining life assessment value, the corresponding abnormal position and the classification result into the wheel health detection database, and output the corresponding flag to the maintenance decision interface.

[0019] The second aspect of the present invention provides a multi-sensor health detection system for a track soft wheel, including: a feature data acquisition and processing module, which is used to collect the wheel diameter distance, surface topography, ultrasonic echo and operating condition information in real time, preprocess the multi-source data, and construct a wheel health feature data set indexed by the corner; an ultrasonic penetration attenuation feature extraction module, which is used to extract the penetration attenuation features reflecting the internal propagation characteristics of the material based on the comparison of ultrasonic echo energies in double time windows, and form a robust attenuation characterization through time stabilization processing; an internal defect structural anomaly discrimination module, which is used to comprehensively evaluate the structural attributes of internal anomalies at the corner position based on the analysis of ultrasonic echo reflection characteristics and spectral structure stability of ultrasonic echoes, and determine the abnormal response type and its spatial distribution state according to the structural attributes of internal anomalies; a remaining life prediction and maintenance module, which is used to analyze the surface degradation trend and internal evolution behavior for life constraint evaluation for the abnormal position by using the wheel health feature data set, and output maintenance warning and safety decision guidance according to the life constraint evaluation.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) In the present invention, through the multi-sensor fusion of laser measurement and ultrasonic detection, and constructing a health feature data set with the corner as the unified index, the wheel surface wear information and the internal propagation characteristics of the material are associated in the same spatial coordinate system, improving the overall perception ability of complex service states.

[0023] (2) This invention, through the comparison of ultrasonic echo energy in two time windows, the joint analysis of time-domain reflection characteristics and frequency-domain structural stability, and the introduction of time stabilization processing, effectively suppresses the influence of operating noise and transient disturbances, and improves the reliability and repeatability of identifying real internal structural anomalies.

[0024] (3) Based on the continuity analysis of the confidence level of structural defects in the angular direction, and combined with the surface change characteristics and penetration attenuation deviation characteristics, this invention can distinguish between continuous structural anomalies, discontinuous structural anomalies and non-structural anomalies, thereby avoiding maintenance risks caused by misjudgment of a single signal.

[0025] (4) This invention, by constraining and integrating the surface degradation trend and internal evolution behavior into a unified remaining life assessment result, and further combining multi-level life threshold output early warning and decommissioning decision, provides an executable and graded decision basis for wheel condition maintenance.

[0026] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0027] Figure 1 A flowchart of a multi-sensor health detection method for soft rail wheels;

[0028] Figure 2 A structural diagram of a multi-sensor health detection system for soft-rail wheels;

[0029] Figure 3 A schematic diagram of the installation for corner position detection;

[0030] Figure 4 A confidence value distribution diagram of structural defects at corner locations;

[0031] Figure 5 A flowchart for the overall process of multi-sensor health detection and life assessment of soft rail wheels. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figures 1-4 This invention provides a technical solution: a multi-sensor health detection method for soft-rail wheels, such as... Figure 1As shown, the process includes the following steps: S1, real-time acquisition of wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information; preprocessing of multi-source data to construct a wheel health feature dataset indexed by the turning angle; S2, extraction of penetration attenuation characteristics reflecting the internal propagation characteristics of the material based on dual-time-window ultrasonic echo energy comparison, and formation of robust attenuation characterization through time stabilization processing; S3, comprehensive evaluation of the structural properties of internal anomalies at the turning angle position based on ultrasonic echo analysis of ultrasonic echo reflection characteristics and spectral structure stability, and determination of the anomaly response type and its spatial distribution state based on the structural properties of the internal anomalies; S4, for the anomaly location, analysis of surface degradation trends and internal evolution behavior using the wheel health feature dataset to conduct a life constraint assessment, and outputting maintenance warnings and safety decision guidance based on the life constraint assessment.

[0034] Specifically, the process of real-time acquisition of wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information, and preprocessing of multi-source data to construct a wheel health feature dataset indexed by rotation angle is as follows: A line-scan laser camera is fixedly installed on the side of the track to continuously scan the wheel-track contact area, acquiring wheel diameter distance and surface morphology height. The scanning direction of the line-scan laser camera is consistent with the radial section of the wheel, and the scanning frequency is adaptively set according to the wheel's running speed to ensure coverage of all rotation angle positions within the complete wheel rotation cycle. The acquired wheel diameter distance and surface morphology height correspond to the same radial section position of the wheel. An ultrasonic probe is installed on one side of the wheel's running path, emitting ultrasonic pulses into the wheel's interior according to a transmission cycle synchronized with the wheel's rotation speed, and acquiring the corresponding ultrasonic A-scan echo time-domain waveform. The ultrasonic probe is installed with a fixed incident angle and fixed coupling method. The ultrasonic transmission and reception times are synchronously triggered by the wheel's angular velocity signal to ensure the spatial consistency of the ultrasonic A-scan echo time-domain waveforms acquired at different rotation angle positions. Simultaneously, the wheel angular velocity is acquired in real time, and the ultrasonic propagation velocity in the wheel material is obtained. The wheel angular velocity is acquired by an encoder, and the ultrasonic propagation velocity is a constant parameter determined by a material parameter table based on the wheel material type, used for subsequent ultrasonic propagation delay and depth conversion. Using the wheel angular velocity as a reference, time-to-angle mapping is performed on the wheel diameter distance data and surface topography height data. Data from different sampling times are uniformly converted into a sequence indexed by the wheel angle. The sampling time of the ultrasonic A-scan echo time-domain waveform is synchronously corrected based on the wheel angular velocity. The time-to-angle mapping is achieved by integrating the angular velocity and matching discrete indices. Specifically, the wheel angular velocity signal is integrated over time to obtain the instantaneous angle value, and this angle value is mapped to a preset set of discrete angle indices. When the sampling time and the target angle index are not completely consistent, the nearest neighbor mapping method is preferably used to align the wheel diameter distance, surface topography height, and ultrasonic echo characteristics to ensure the synchronization of multi-source data in the angle dimension. Through mapping and correction processes, a one-to-one correspondence is established between multi-source data from laser and ultrasonic detection in the rotation dimension, thereby eliminating the influence of differences in sampling frequency and sampling time of different sensors on subsequent analysis. For example, taking the rotation range of a wheel completing a full rotation as 0° to 360°, the wheel rotation process is discretized according to a fixed angular step size. When the step size is set to 1°, each rotation position corresponds to a unique angular index. During the wheel's operation, the wheel rotation angle corresponding to any sampling time can be obtained by integrating the angular velocity at the corresponding time and mapped to the nearest angular index position, so that the wheel diameter distance, surface morphology height, and ultrasonic A-scan echo data collected at different times are all uniformly associated with the same rotation position for analysis.Median filtering is applied to the wheel diameter distance data and surface topography height data to suppress local anomalies caused by track vibration and optical measurement noise. Median filtering is performed with adjacent corner positions as the window to weaken isolated spikes without altering the overall geometric trend. Bandpass filtering is applied to the ultrasonic A-scan echo time-domain waveform to remove low-frequency mechanical disturbances and high-frequency noise components. After filtering, the ultrasonic A-scan echo time-domain waveform is enveloped to obtain the ultrasonic envelope signal. The passband range of the bandpass filter is set according to the center frequency and bandwidth of the ultrasonic probe, preferably 0.6 to 1.4 times the probe's center frequency, to retain the main echo energy and suppress non-operating frequency noise. Envelope extraction preferably uses the Hilbert transform method, constructing an analytical signal of the ultrasonic signal and obtaining its amplitude to obtain the ultrasonic envelope, which characterizes the distribution characteristics of echo energy over time, facilitating subsequent energy integration and structural analysis. Wheel angular velocity is smoothed using a time window, preferably employing a moving average method. The window length is set based on the wheel speed fluctuation characteristics to reduce high-frequency jitter without introducing significant phase delay. Preprocessed wheel diameter distance, surface morphology height, ultrasonic envelope signal, wheel angular velocity, and ultrasonic propagation speed are combined to construct wheel health feature data. Min-maximum normalization is then applied to each wheel health feature data to eliminate differences in the dimensions of different physical quantities, ensuring comparability of features in subsequent fusion analysis. A wheel health detection database is established, incorporating both raw and preprocessed wheel health feature data to achieve traceable data storage and provide a unified data support foundation for subsequent anomaly detection and lifespan assessment.

[0035] like Figure 3The diagram shows a schematic of the installation for corner position detection. It illustrates the installation method for synchronous detection of the wheel's fixed corner position during operation. A line-scan laser camera is fixedly mounted on the side of the track. Its emitted laser line covers the wheel-track contact area radially, continuously scanning the wheel diameter and surface morphology at different corner positions as the wheel rotates, achieving high-precision acquisition of the wheel's external surface geometry. The laser scanning direction is consistent with the wheel's radial cross-section, ensuring the comparability of data collected at each corner position. An ultrasonic probe is installed on one side of the wheel's running path, facing inwards and stably coupled to the wheel material. The ultrasonic probe emits ultrasonic pulses into the wheel's interior according to a transmission cycle synchronized with the wheel's rotational speed and receives ultrasonic echo signals reflected from the incident interface and the wheel's interior, thereby acquiring the ultrasonic A-scan echo time-domain waveform at the corresponding corner position to characterize the wheel's internal structural state. By using the wheel's angular velocity as a unified synchronization reference, laser detection and ultrasonic detection achieve spatiotemporal alignment at the same corner position, ensuring a one-to-one correspondence between external geometric information and internal ultrasonic response in the corner dimension. This installation method ensures that synchronous and stable multi-source detection data can be obtained at each corner position during the complete rotation cycle of the wheel, providing a reliable data foundation for subsequent structural defect identification, anomaly type identification and life assessment.

[0036] In this implementation scheme, by using the wheel rotation angle as a unified index, the external geometric information acquired by laser and the internal structural response acquired by ultrasound are spatiotemporally aligned and collaboratively preprocessed, achieving synchronous perception of the wheel surface state and internal characteristics. Through adaptive scanning, synchronous triggering, filtering and denoising, and stabilization processing, the influence of operating vibration and measurement noise on the detection results is effectively suppressed, ensuring the consistency and comparability of multi-source data in the rotation angle dimension. On this basis, a standardized wheel health feature dataset and traceable database are constructed, providing a reliable, stable, and feasible data foundation for subsequent anomaly type determination, structural defect identification, and remaining life assessment.

[0037] Specifically, based on the comparison of ultrasonic echo energy in dual time windows, the penetration attenuation characteristics reflecting the internal propagation properties of the material are extracted, and a robust attenuation characterization is formed through time stabilization processing. The specific process is as follows: For the ultrasonic envelope signal corresponding to each corner position, an incident interface time window and a deep response time window are constructed. The incident interface time window corresponds to the echo interval formed by the ultrasonic wave from the probe to the contact interface on the wheel surface, and the deep response time window corresponds to the echo interval formed by the ultrasonic wave after propagation inside the wheel and in the direction of the wheel hub. The start and end positions of the incident interface time window and the deep response time window are predetermined according to the installation position of the ultrasonic probe, the incident angle, and the sound velocity of the wheel material, and are calibrated during the initialization stage to ensure the consistency and repeatability of the time window division at different corner positions. The start and end boundaries of the incident interface time window and the deep response time window are determined according to the following rules: During the initial calibration phase, a defect-free standard wheel is used as the inspection object, and multiple ultrasonic A-scan echo signals are collected. The echo signals are then time-aligned and superimposed for analysis. First, the echo peak whose amplitude first significantly exceeds the noise baseline is identified in the ultrasonic envelope signal as the reference arrival time of the incident interface echo. The preset time width is extended forward and backward with the reference arrival time as the center to form the incident interface time window. Subsequently, based on the known ultrasonic propagation speed and the probe incident geometry parameters, the theoretical arrival time interval of the ultrasonic wave propagating inside the wheel to the expected deep region and returning is calculated. Within the theoretical interval, a stable deep reference echo peak is identified. The preset time width is extended forward and backward with the reference arrival time of the deep reference echo peak as the center to form the deep response time window. The preset time width is preferably determined based on the ultrasonic pulse width and the system time resolution so that the corresponding time window can completely cover the target echo energy while avoiding overlap with adjacent echo intervals. The incident interface time window and the deep response time window, determined by the above method, remain fixed during subsequent detection, only being updated with the wheel rotation angle index. This ensures consistent time window division rules and repeatable boundary definitions at different rotation angles. Energy integration is performed on the ultrasonic envelope signals within the incident interface time window and the deep response time window to calculate the ultrasonic energy of the incident interface time window and the ultrasonic energy of the deep response time window. The energy integration, based on the integral of the ultrasonic envelope signal amplitude over time, characterizes the effective energy level of the ultrasonic wave within the corresponding propagation path, thus avoiding instability caused by occasional noise affecting a single peak value.The ratio of ultrasonic energy in the incident interface time window to that in the deep response time window is calculated, and the natural logarithm of the ratio is taken to obtain the normalized penetration attenuation value. The normalized penetration attenuation value characterizes the relative attenuation of ultrasonic energy with propagation depth by comparing the energy changes of the ultrasonic wave along the propagation path at the incident interface and inside the wheel. Using the energy ratio and its logarithmic transformation, the influence of differences in absolute amplitude and coupling conditions can be eliminated. The absolute amplitude difference of the ultrasonic signal at different corner positions is transformed into a dimensionless feature that is sensitive to internal propagation attenuation characteristics and insensitive to changes in coupling conditions. This makes the indicator more sensitive to changes in the internal propagation characteristics of the material, thus stably reflecting the abnormal evolution of the internal structural state. Based on the sliding time window, the median of the normalized penetration attenuation values ​​at the same corner position is taken to obtain a stable normalized penetration attenuation characteristic value. This stable normalized penetration attenuation characteristic value is correlated with wheel health characteristic data according to the corner index and written into the wheel health detection database. The sliding time window covers multiple adjacent sampling periods, preferably 3 to 9 adjacent sampling periods. The sampling period is based on the single wheel angle index update period, which is used to reduce the impact of instantaneous noise, occasional coupling fluctuations, or single abnormal echoes on the penetration normalized attenuation value. Median calculation is used to further suppress extreme value interference, thereby obtaining robust attenuation characterization results with time stability and engineering feasibility. Through associated storage with the angle index, the attenuation characteristics can directly participate in the subsequent structural defect determination and anomaly type differentiation process.

[0038] The specific formula for the penetration normalized attenuation value is as follows:

[0039] ;

[0040] In the formula, The normalized penetration attenuation value represents the relative energy attenuation level of ultrasonic waves as they propagate from the surface of a wheel to its interior. The larger the value, the more significant the attenuation of the deep echo energy relative to the incident interface echo, which usually corresponds to enhanced internal scattering or a disturbed propagation path. The incident interface time window corresponds to the echo interval formed when the ultrasonic wave is incident from the probe to the contact interface on the wheel surface. It mainly reflects the coupling state and surface reflection conditions, and serves as an energy reference for internal propagation. This indicates the deep response time window, which corresponds to the echo range formed after the ultrasonic wave propagates inside the wheel material and in the direction of the wheel hub. It includes the comprehensive response information after material absorption, scattering, and the effect of internal defects. This represents the ultrasonic envelope signal, and the ultrasonic echo amplitude sequence after filtering and envelope extraction. It is used to stably describe the echo energy distribution and avoid the influence of phase interference on energy calculation.

[0041] In this implementation scheme, by comparing the energy of ultrasonic echoes in two time windows along the corner dimension, a penetration normalized attenuation characterization sensitive to the internal propagation characteristics of materials is constructed, effectively reducing the influence of coupling conditions, amplitude fluctuations, and transient noise on the detection results. By initializing the calibration and mapping to a fixed time window, the spatial consistency and repeatability of attenuation characteristics at different corner positions are ensured. Combined with a sliding time window and median stabilization processing, attenuation characteristics with good temporal stability and engineering feasibility are obtained, thus providing reliable and robust feature support for subsequent internal structural defect determination and anomaly type differentiation.

[0042] Specifically, the process of analyzing the ultrasonic echo reflection characteristics and spectral structure stability based on ultrasonic echo is as follows: Within the deep response time window, the ultrasonic envelope signal at the corresponding corner position is extracted, and the median of the ultrasonic envelope signal amplitude is taken as the noise baseline value. The noise baseline value is used to characterize the stable background response level formed by the intrinsic scattering of materials, system noise, and coupled background at the current corner position and deep propagation path. The median calculation is used to suppress the interference of occasional spikes and single abnormal echoes on the baseline estimation. Before determining the noise baseline value, the boundary intervals at both ends of the ultrasonic envelope signal within the deep response time window are removed. The removal ratio is preferably 5% to 10% of the total length of the time window to avoid the influence of echo tailing or time window segmentation error on noise estimation. Among the remaining sampling points, robust statistical screening is preferably performed based on the absolute deviation of the median to remove abnormal sampling points that deviate from the median by more than 3 times the absolute deviation of the median. Then, the median of the ultrasonic envelope signal amplitude after screening is taken as the noise baseline value, thereby improving the stability and robustness of noise baseline estimation under complex working conditions. Simultaneously, a local maximum search is performed on the ultrasonic envelope signal within the deep response time window. Among the echo peaks whose ultrasonic envelope signal amplitude is higher than the noise baseline, the echo peak with the largest amplitude is selected as the significant reflection response value of the deep echo at the current corner position. By filtering only the local maxima above the noise baseline, background noise or weak scattering is avoided from being misjudged as a valid reflection response, thereby improving the reliability of the discrimination of internal abnormal reflection characteristics. In particular, during the local maximum search process, a minimum time interval constraint is preferably set between adjacent echo peaks. The minimum time interval preferably corresponds to a time length of not less than one ultrasonic center frequency cycle to avoid the same physical reflection event being repeatedly identified as multiple echo peaks, thereby suppressing the interference of ringing or local high-frequency fluctuations on the determination of significant reflection response. The ratio of the significant reflection response value to the noise baseline value is calculated to obtain the significance value of the newly added echo peak. This value is used to quantify the enhancement of the newly added reflection response relative to the background level at the current corner position, enabling the significant reflection feature to be compared across corner positions in a dimensionless form. Simultaneously, a short-time Fourier transform (SFT) is performed on the time-domain waveform of the ultrasonic A-scan echo at the same corner position to obtain the time spectrum. The SFT, while ensuring time resolution, performs local analysis of the frequency composition of the ultrasonic echo to characterize the changes in the echo spectrum structure with propagation depth and material state. The time window length of the SFT is preferably 3 to 8 ultrasonic center frequency cycles, and the overlap rate between adjacent time windows is preferably 50% to 75% to balance frequency and time resolution and improve the smoothness and stability of the spectrum estimation. Furthermore, the frequency analysis bandwidth used in the SFT is consistent with the effective working bandwidth of the ultrasonic probe, limiting the frequency range involved in the spectrum analysis to within the effective working bandwidth of the ultrasonic probe, ensuring that the calculated spectral entropy is consistent with the physical characteristics of the sensor.Within the corresponding deep response time window, the energy of each frequency component in the time spectrum is integrated to obtain the contribution energy value of each frequency component. The contribution energy values ​​of all frequency components are summed to obtain the total contribution energy value. The ratio of the contribution energy value of each frequency component to the total contribution energy value is calculated to obtain the spectral energy probability of each frequency component. Through energy normalization, the relative contribution of different frequency components in the spectrum is uniformly characterized, reducing the influence of absolute amplitude changes on the frequency domain analysis results. The spectral energy probability of each frequency component is multiplied by its own logarithm, and the product of all frequency components is summed and the negative number is taken to obtain the spectral entropy. The spectral entropy is used to characterize the complexity and dispersion characteristics of the deep echo spectral structure. When there are structural defects inside the material, its spectral structure usually undergoes enhanced order or energy concentration changes, and the spectral entropy shows a distinguishable trend, thus providing a frequency domain stability basis for subsequent structural attribute determination.

[0043] In this implementation scheme, by jointly extracting significant reflection features and spectral structure stability features of ultrasonic echoes within the deep propagation path, a synergistic characterization of internal abnormal reflection intensity and frequency domain structure changes is achieved. Through noise baseline constraints, saliency normalization, and spectral energy probability modeling, the influence of background noise and occasional fluctuations on the judgment results is effectively suppressed, making the structural defect response have good comparability and stability under different corner positions and operating conditions, thereby providing reliable and implementable feature support for the structural attribute discrimination of internal anomalies.

[0044] Specifically, the process for comprehensively evaluating the structural attributes of internal anomalies at corner positions is as follows: Historical spectral entropy sequences corresponding to the same corner position are read from the wheel health detection database, and the median is taken as the spectral entropy reference value. The number of samples in the historical spectral entropy sequence is preferably 20 to 30 sampling periods. The median is used to characterize the typical stability level of the spectral structure at the corner position, thereby reducing the influence of individual abnormal samples on the reference value. The logarithm of the peak significance value is taken to obtain the time-domain reflection enhancement. A logarithmic transformation is used to compress the exponential change of the significance value into a linearly comparable interval, giving the reflection enhancement degree at different corner positions better numerical stability. The negative of the time-domain reflection enhancement is used as the exponent for natural exponential operation. The result of the natural exponential operation is added by one and the reciprocal is taken to obtain the time-domain structural response value. This transformation is used to map the reflection enhancement into a response index between 0 and 1, so that the more significant the reflection enhancement, the more significant the corresponding response. The larger the structural response value, the more monotonically consistent the judgment relationship is formed. The frequency domain structural deviation is obtained by subtracting the current frequency domain entropy from the reference value of the frequency domain entropy. This deviation is used to quantify the degree of change of the frequency domain structure relative to the historical stable state at the current turning position. The negative of the frequency domain structural deviation is taken as the exponent for natural exponentiation. The result of natural exponentiation is added by one and the reciprocal is taken to obtain the frequency domain structural stability response value. Through the above mapping method, the frequency domain structural stability response value is larger when the frequency domain structure is closer to the reference state, and the response value decreases accordingly when the frequency domain structure changes significantly. The structural defect confidence value is obtained by multiplying the time domain structural response value and the frequency domain structural stability response value. Through the multiplicative fusion of time domain reflection features and frequency domain structural features, the structural defect confidence value will increase significantly only when both types of features simultaneously exhibit structural anomalies. This avoids misjudgment caused by a single feature anomaly and improves the reliability and robustness of internal structural defect judgment.

[0045] The specific formula for the confidence value of structural defects is as follows:

[0046] ;

[0047] In the formula, This represents the confidence value for structural defects, used to comprehensively characterize whether the internal anomaly at the current corner position has structural defect characteristics. The larger the value, the more likely the anomaly originates from actual structural damage. This represents the deep response time window, which corresponds to the time interval during which ultrasonic waves propagate inside the wheel material and return to form an echo. It is the main window for carrying internal defect reflections and material structure information. The significant reflection response value represents the significant echo peak with the largest amplitude within the deep response time window, used to reflect the concentrated reflection intensity caused by internal interfaces or defects. The noise baseline value represents the median amplitude of the ultrasonic envelope signal within the deep response time window, used to characterize background noise and weak scattering levels, and as a reference benchmark for determining reflection enhancement. The reference value for spectral entropy represents the median level of the spectral entropy of historical ultrasound echoes at the same corner position, used to describe the spectral structure complexity of that position under normal conditions. The spectral energy probability of a frequency component represents the proportion of energy occupied by each frequency component within the deep response time window, and is used to characterize the distribution of energy in the frequency domain. Spectral entropy is used to measure the dispersion and complexity of the current echo spectrum. Deviation of spectral entropy from the reference value usually reflects changes in material structure or propagation path.

[0048] In this embodiment, Table 1 is a data table of confidence values ​​for structural defects. The table details the significant reflection response value, noise baseline value, spectral entropy reference value, spectral entropy, and structural defect confidence value for five corner positions; specifically, the significant reflection response value for corner position 1 is 2.40, the noise baseline value is 0.80, the spectral entropy reference value is 1.22, the spectral entropy is 0.92, and the structural defect confidence value is 0.4308; the significant reflection response value for corner position 2 is 2.05, the noise baseline value is 0.85, the spectral entropy reference value is 1.25, the spectral entropy is 1.02, and the structural defect confidence value is 0.3939; the significant reflection response value for corner position 3 is... The significant reflection response value at corner position 4 is 1.65, the noise baseline value is 0.90, the spectral entropy reference value is 1.11, the spectral entropy is 1.12, and the structural defect confidence value is 0.3219; the significant reflection response value at corner position 5 is 1.10, the noise baseline value is 1.00, the spectral entropy reference value is 1.22, the spectral entropy is 1.23, and the structural defect confidence value is 0.2606.

[0049] Table 1 Confidence Values ​​for Structural Defects

[0050]

[0051] like Figure 4 The figure shows the distribution of confidence values ​​for structural defects at corner locations. The graph illustrates the bar chart of the confidence values ​​for structural defects at different corner locations, with the horizontal axis representing the corner location number and the vertical axis representing the confidence value of the structural defect; the dashed line represents the structural defect discrimination threshold. (Refer to Table 1 and...) Figure 3It can be seen that the confidence values ​​decrease overall along the corner direction. From corner positions 1 to 5, the confidence values ​​for structural defects gradually decrease, indicating that the internal structural anomaly response has obvious local concentration in space, rather than uniform distribution. The confidence values ​​at corner positions 1, 2, and 3 are significantly higher than the discrimination threshold, indicating that there are relatively significant and stable structural defect responses in the area, which are potential key risk areas. Corner positions 4 and 5 are lower than the discrimination threshold, indicating that the internal structural response in the area is generally stable and does not show significant structural defect characteristics. The changes in the significant reflection response intensity and spectral entropy deviation in the data table are consistent with the trend of confidence value changes, verifying the effectiveness of joint modeling of temporal reflection enhancement and frequency domain structural stability for structural defect identification.

[0052] In this implementation scheme, by introducing the median of historical spectral entropy as a stable reference and combining a dual nonlinear transformation mechanism of logarithmic compression and exponential mapping, the time-domain reflection characteristics and frequency-domain structural characteristics of ultrasonic echoes are uniformly mapped into a response index with a consistent scale and monotonic judgment relationship. Furthermore, through the multiplicative fusion of time-domain and frequency-domain responses, the confidence of structural defects is significantly improved only when both exhibit structural anomalies. This effectively suppresses misjudgments caused by local noise, sporadic echoes, or single feature fluctuations, achieving robust identification and reliable differentiation of internal structural defects in wheels, and significantly improving the accuracy of anomaly judgment and engineering feasibility.

[0053] Specifically, the process of determining the type of abnormal response and its spatial distribution based on the structural properties of internal anomalies is as follows: Based on a sliding time window, the median of the confidence values ​​of structural defects at the same corner position is taken to obtain a stable structural defect confidence value; wherein, the sliding time window covers multiple adjacent sampling periods along the time axis, preferably covering 3 to 9 sampling periods, to reduce the influence of instantaneous noise, occasional echo enhancement, or single abnormal disturbance on the confidence value of structural defects, and the median is used to suppress extreme value interference, thereby obtaining a corner position structural characterization result that is stable and repeatable in engineering applications. Along the wheel turning direction, the confidence values ​​of stable structural defects at the N adjacent turning positions are statistically analyzed, and it is determined whether they simultaneously exceed the discrimination threshold. If so, a continuous structural defect response is considered to exist inside the wheel corresponding to the turning position. Here, N represents the minimum continuous response length along the circumferential direction, preferably set according to the wheel geometry, ultrasonic spatial resolution, and turning position sampling density, taking 3 to 10 adjacent turning positions to cover the equivalent length of at least one ultrasonic spatial resolution unit in the circumferential direction. This avoids isolated anomalies at a single turning position being misjudged as continuous structural defects, thus ensuring the spatial rationality of the continuous structural defect determination. When the number of turning positions where the confidence value of stable structural defects exceeds the discrimination threshold is less than the determination number threshold, a discontinuous structural defect response is considered to exist. Here, the determination number threshold is used to distinguish between continuous and discrete responses, preferably set to a fixed number less than N, to represent the situation where the abnormal response does not form a stable continuous distribution along the turning direction. If the confidence values ​​of stable structural defects at all corner positions do not exceed the discrimination threshold, it is considered that no structural defect response has been detected. This determination is used to characterize the situation where, within the current detection cycle, the temporal reflection characteristics and frequency domain structural characteristics of the ultrasonic echo do not show significant deviations from the historical stable state, thereby avoiding triggering unnecessary abnormal judgments under normal operating conditions. Simultaneously, corresponding wheel health characteristic data and stable penetration normalized attenuation characteristic values ​​are acquired to categorize abnormal wheel responses. The wheel health characteristic data includes changes in wheel diameter and surface morphology height, representing surface state characteristics. The stable penetration normalized attenuation characteristic value characterizes the overall attenuation characteristics of ultrasonic energy propagating within the material. By jointly analyzing the spatial distribution state determination results with the above characteristics, a reliable distinction can be made between different types of abnormal responses—structural defect-dominated, non-structural disturbance, and no-abnormality states—providing a clear and actionable basis for subsequent life assessment and maintenance decisions.

[0054] In this implementation plan, by combining median stabilization in the time dimension with continuity constraints in the corner spatial dimension, the confidence values ​​of structural defects are robustly characterized and their spatial distribution is determined. This effectively suppresses the interference of transient noise, isolated anomalies, and occasional echoes on the results. At the same time, the continuity determination results are jointly analyzed with surface change characteristics and penetration attenuation characteristics to clearly distinguish between continuous structural defects, discontinuous structural defects, non-structural disturbances, and anomaly-free states. This improves the reliability of internal anomaly identification while ensuring engineering repeatability and spatial rationality, providing clear and actionable judgment criteria for subsequent life assessment and maintenance decisions.

[0055] Specifically, the process of synchronously acquiring corresponding wheel health characteristic data and stable penetration normalized attenuation characteristic values, and then categorizing wheel abnormal responses, is as follows: When there is a continuous structural defect response, if the changes in wheel diameter distance and surface morphology height at adjacent sampling points at the corresponding corner position do not exceed the corresponding abnormal change threshold, the anomaly at the corner position is determined to be an anomaly dominated by internal structural defects. Adjacent sampling points refer to two adjacent corner positions in the corner index sequence. The changes in wheel diameter distance and surface morphology height are obtained by the numerical difference between adjacent corner positions. The abnormal change threshold is determined based on the statistics of the corresponding values. The distribution is determined to distinguish between slow changes and significant anomalous changes in surface geometry, indicating that when the surface geometry remains stable, the anomalous response mainly originates from changes in the internal structure of the material. Conversely, a composite anomaly of structure and surface is indicated, representing a situation where significant changes in the wheel surface geometry occur simultaneously with internal structural anomalies, reflecting the anomalous state of coupled evolution between internal defects and surface degradation. When discontinuous structural defect responses exist, if the difference between the stable penetration normalized attenuation eigenvalue and the median of historical stable penetration normalized attenuation eigenvalues ​​exceeds a deviation threshold, the anomaly corresponding to the corner position is determined to be a non-structural defect anomaly. The deviation threshold is determined based on the statistical distribution of historical sample deviations and is used to characterize significant changes in the overall propagation characteristics of ultrasonic energy relative to a steady state. When the change does not form a spatially continuous distribution, it is considered more likely to originate from local non-structural defects in the material or disturbances in coupling conditions; otherwise, it is judged as no anomaly, indicating that this type of discontinuous response does not show a continuous or significant change in internal propagation characteristics and does not constitute an abnormal state in an engineering sense. When no structural defect response is detected, if the corresponding wheel diameter distance change, surface morphology height change, and stable penetration normalized attenuation characteristic value are all within the corresponding normal distribution range, it is judged as... No abnormalities were found. Specifically, the normal distribution range of wheel diameter variation and surface morphology height variation indicates that neither exceeded the corresponding abnormal change threshold. The normal distribution range of stable penetration normalized attenuation characteristic values ​​indicates that the difference between the stable penetration normalized attenuation characteristic value and the median of historical stable penetration normalized attenuation characteristic values ​​does not exceed the deviation threshold. Conversely, non-structural defect anomalies are indicated, suggesting abnormal disturbances or localized non-structural changes exceeding the normal fluctuation range even without a structural defect response. The above anomaly typification rules are based on the propagation and reflection mechanism of ultrasound within materials and the differences in the influence of surface geometric changes on the detection signal. Internal structural cracks or delamination defects typically form stable acoustic impedance abrupt change interfaces, easily causing continuous reflection enhancement of deep echoes and concentrated changes in spectral structure, exhibiting spatially continuous distribution characteristics in the angular direction. In the early stages of this type of anomaly, the wheel surface geometry remains stable, and the wheel diameter variation and surface morphology variation are not significantly abnormal; therefore, it can be determined as an anomaly dominated by internal structural defects.As defects evolve further and couple with surface wear or localized spalling processes, internal reflection anomalies and surface geometric changes occur simultaneously, forming a composite anomaly of the structure and surface. In contrast, non-structural defects such as material bubbles, localized density inhomogeneities, or coupling condition disturbances typically only cause discrete deviations in the overall propagation attenuation of ultrasonic energy, lacking stable reflection enhancement and spatial continuity characteristics. Therefore, the degree of deviation of the penetration normalized attenuation characteristic can be combined to determine whether they are non-structural defects or normal states. Through the above joint judgment rule, the interpretable distinction between different anomaly mechanisms is achieved, misjudgments are avoided, and the reliability of anomaly diagnosis and maintenance decisions is improved. The stable penetration normalized attenuation characteristic value and the corresponding anomaly judgment result are written into the wheel health detection database to achieve traceable storage of anomaly judgment results and provide a unified data foundation and judgment basis for subsequent life assessment, trend analysis, and maintenance decisions.

[0056] In this implementation plan, by combining the spatial continuity determination of structural defect confidence values ​​with the joint analysis of multi-source wheel health characteristics, a fine distinction is made between different types of abnormal wheel responses, such as those dominated by internal structural defects, those involving structural surfaces, and those caused by non-structural disturbances. While ensuring sensitivity to real internal structural defects, it effectively suppresses misjudgments caused by surface wear fluctuations, changes in coupling conditions, or occasional echo anomalies, thereby improving the reliability, stability, and engineering feasibility of the anomaly determination results and providing a clear and actionable basis for subsequent life assessment and maintenance decisions.

[0057] Specifically, for abnormal locations, the process of using wheel health feature datasets to analyze surface degradation trends and internal evolution behavior for life constraint assessment is as follows: For abnormal corner locations, the corresponding wheel diameter distance is obtained, and the rate of change of wheel diameter distance is calculated using the adjacent sampling point difference method to obtain the surface wear degradation rate; wherein, the adjacent sampling point difference method is performed based on the wheel diameter distance change under adjacent corner indices at the same corner location, which is used to characterize the actual wear evolution rate of the wheel surface material during operation, thereby avoiding the influence of single measurement errors on the degradation trend judgment. Simultaneously, within the incident interface time window at the corresponding corner position, a local maximum search is performed on the ultrasonic envelope signal. The time corresponding to the echo peak with the largest amplitude is selected as the reference echo time, which is used to characterize the stable reflection reference position of the ultrasonic wave reaching the contact interface on the wheel surface. Within the deep response time window, a local maximum search is performed on the ultrasonic envelope signal, and the time corresponding to the echo peak with the largest amplitude is selected as the deep echo time, which is used to characterize the position where the ultrasonic wave propagates inside the wheel and undergoes major reflection. The difference between the deep echo time and the reference echo time is calculated to obtain the ultrasonic propagation delay. The propagation delay reflects the change in the effective propagation path length of the ultrasonic wave inside the material and is an important time-domain characterization parameter for the spatial location and evolution degree of internal defects. Multiplying the ultrasonic propagation delay by the ultrasonic propagation velocity and dividing by two yields the equivalent depth value of the internal defect. The ultrasonic propagation velocity is a material parameter obtained from a table based on the wheel material type. Dividing by two is used to correct the ultrasonic round-trip propagation path so that the equivalent depth value truly reflects the one-way spatial position of the defect from the incident interface. The rate of change of the equivalent depth of internal defects is calculated using the adjacent sampling point difference method, which serves as the internal defect propagation rate. This rate characterizes the growth trend of internal defects during operation, thus distinguishing between stable defects and continuously expanding defect states. The remaining surface wear life is obtained by dividing the difference between the wheel diameter distance limit and the current wheel diameter distance by the sum of the surface wear degradation rate and a minimum positive number. The wheel diameter distance limit is obtained through statistical analysis of historical wheel diameter distance sequences, preferably using the lower quantile of the historical wheel diameter distance distribution, such as the 1st to 5th percentile. The minimum positive number is used to avoid numerical divergence when the degradation rate approaches zero, thereby ensuring the numerical stability of the life calculation. The remaining lifetime value of the internal defect is obtained by dividing the difference between the internal defect equivalent depth limit and the current internal defect equivalent depth value by the sum of the internal defect propagation rate and the smallest positive number. The internal defect equivalent depth limit is obtained by statistical analysis of the historical internal defect equivalent depth value sequence. The lower quantile value of the historical internal defect equivalent depth value distribution, such as the 1% to 5% quantile, is preferably selected as the internal defect equivalent depth limit value to constrain the safety boundary of the internal structure.The reciprocals of the remaining lifetime values ​​for surface wear and internal defects are taken separately, and the two reciprocals are added together to obtain the comprehensive degradation rate. The comprehensive degradation rate is used to fuse the two failure mechanisms of surface degradation and internal evolution, so that the acceleration of either degradation path can be reflected in the fusion result. The reciprocal of the comprehensive degradation rate is taken and multiplied by a constant two to obtain the fused remaining lifetime assessment value. Here, the constant two is the normalization coefficient for the fusion of dual degradation channels. It is used to scale the comprehensive degradation rate when the two equally weighted failure mechanisms of surface wear degradation and internal defect evolution participate in the evaluation at the same time. This makes the numerical range of the fused remaining lifetime assessment value consistent with that of a single degradation channel, avoiding the systematic underestimation of the lifetime assessment result due to channel superposition. This ensures that the fused remaining lifetime assessment value has stable engineering interpretability and can be directly compared with the threshold.

[0058] The specific formula for the integrated remaining useful life assessment value is as follows:

[0059] ;

[0060] In the formula, This represents the combined remaining service life assessment value, which indicates the comprehensive remaining service life under the dual constraints of surface wear and internal defects, and is used for subsequent comparison and maintenance decisions. It represents the wheel diameter distance limit, indicating the minimum safe wheel diameter distance or wear limit that a wheel can withstand, and is used to constrain the safety boundary of surface wear; This indicates the current wheel diameter distance, and the actual wheel diameter distance corresponding to the current turning position, used to characterize the current surface wear state; It represents the surface wear degradation rate, the rate of change of wheel diameter distance over time or mileage, and is used to reflect the evolution speed of surface geometric degradation; This represents the limit value of the equivalent depth of internal defects, which indicates the maximum equivalent depth of internal defects allowed under safety design and operating specifications, and is used to limit the safety boundary of the internal structure. The value represents the current equivalent depth of the internal defect, which is estimated based on the ultrasonic propagation time delay and is used to characterize the degree of internal damage. It represents the rate of internal defect propagation and the rate of change of the equivalent depth of internal defects over time or mileage, used to reflect the propagation trend of cracks or internal damage. This represents a very small positive number, used to prevent numerical divergence when the degradation or propagation rate approaches zero, ensuring the stability of lifetime assessment calculations. The preferred value range is [value range missing]. arrive .

[0061] In this implementation scheme, by integrating the wheel diameter distance change and the ultrasonic propagation time delay evolution information in the corner position dimension, the wear degradation trend of the wheel surface and the internal defect expansion behavior are respectively characterized. On this basis, statistically robust limit constraints and numerical stabilization processing are introduced to achieve the unified life constraint evaluation of the two types of failure mechanisms on the surface and inside. This method can suppress the influence of single measurement errors and occasional disturbances while accurately reflecting the actual evolution level of the degradation rate, making the fusion remaining life evaluation results not only have engineering interpretability but also stability and repeatability, thus providing a reliable and executable quantitative basis for subsequent maintenance warning and safety decision-making.

[0062] Specifically, the specific process of outputting maintenance warning and safety decision-making guidelines according to the life constraint evaluation is as follows: The fusion remaining life evaluation value is associated with the corresponding abnormal type according to the corner index and written into the wheel health detection database; among them, the corner index is used to uniquely identify the spatial position of the abnormality in the circumferential direction of the wheel, and the abnormal type is used to distinguish the internal structure defect-dominated abnormality, the structural surface composite abnormality, and the non-structural defect abnormality, so that the life evaluation result can be consistently associated with the cause of the abnormality, thereby avoiding the confusion of the life evaluation results between different abnormal types. If there is more than one abnormal corner position in the current wheel, the corner position with the smallest fusion remaining life evaluation value is selected for evaluation; by selecting the position with the smallest fusion remaining life evaluation value, it is ensured that the maintenance decision is based on the most unfavorable working conditions, so that the safety redundancy requirements can still be met when multiple abnormalities coexist, and the underestimation of the overall risk caused by the neglect of local abnormalities can be avoided. The fusion remaining life evaluation value is compared with the multi-level life evaluation thresholds R1 and R2: When ≥R2, it is determined that the corresponding corner position is in the normal operation state, and only the conventional data monitoring is maintained; among them, the conventional data monitoring refers to maintaining the established sampling period and analysis process without triggering additional maintenance and intervention operations, which is used to characterize the situation where the current degradation level of the wheel has no substantial impact on the operation safety. When R1≤ <R2, it is determined that the corresponding corner position is in the warning maintenance state, a maintenance warning flag is generated, and a maintenance prompt is pushed; the warning maintenance state is used to characterize the operation stage where the wheel has shown a perceivable degradation trend but has not reached the safety limit, and the maintenance prompt is used to remind the operation and maintenance personnel to conduct key inspections and take preventive maintenance measures on the corresponding corner position within the planned maintenance window. When When <R1, it is determined that the corresponding corner position is in a safety-restricted state, a deactivation recommendation flag is generated, and a wheel replacement prompt is pushed; the safety-restricted state is used to indicate that the integrated remaining life is insufficient to support the safety operation requirements, and the deactivation recommendation flag is used to trigger a mandatory safety intervention process to prevent the defect from continuing to evolve and causing a failure accident. The integrated remaining life assessment value, the corresponding abnormal position, and the grading result are written into the wheel health detection database, and the corresponding flag is output to the maintenance decision interface. Through the method of database storage and maintenance decision interface output, the effective connection of the life assessment result between the detection system and the operation and maintenance system is realized, ensuring that the assessment result has traceability, callability, and engineering implementation, so as to support the long-term operation status management and maintenance strategy optimization.

[0063] As Figure 5 shown, it is the overall flow chart of multi-sensor health detection and life assessment for track soft wheels. In the figure, starting from multi-sensor data acquisition, the wheel diameter distance and surface topography are obtained through line-scanning laser, the A-scan echo signal is collected by an ultrasonic probe, and the wheel speed and operating conditions are synchronously obtained. After time synchronization, corner mapping, and preprocessing, a wheel health feature dataset indexed by the corner is constructed; on this basis, on the one hand, the stable penetration normalized attenuation feature is extracted by comparing the ultrasonic energy of double time windows, and on the other hand, the confidence value of the structural defect is calculated through the time-domain and frequency-domain structure analysis of the ultrasonic echo, and the continuous structural defect, non-continuous structural defect, or no structural defect response is distinguished by combining the continuity determination in the corner direction; and according to different responses, the abnormal type is divided by combining the wheel health feature dataset and the stable penetration normalized attenuation feature. Subsequently, the life constraint assessment is performed for the determined abnormal type, the surface wear degradation rate and the internal defect expansion rate are calculated respectively, the integrated remaining life assessment value is obtained by fusion, and it is compared with multiple levels of life thresholds. Finally, maintenance decision guidelines such as normal monitoring, maintenance warning, or safety restriction are output, realizing a closed-loop health management process from detection, determination to prediction and decision-making.

[0064] In this implementation plan, by associating the integrated remaining life assessment result with the corner position and abnormal type, a multi-level life determination and maintenance decision mechanism oriented to the most unfavorable working conditions is established, realizing the refined quantification and hierarchical control of the wheel degradation state; through the comparison of multiple levels of thresholds, the normal operation, warning maintenance, and safety-restricted states are clearly distinguished, making the maintenance decision change from passive response to active warning based on life constraints; at the same time, combined with database storage and maintenance decision interface output, ensuring that the assessment result has traceability and engineering executability, thereby improving the running safety of the wheel, avoiding the risks of over-maintenance or missed inspection, and enhancing the overall operation and maintenance efficiency and decision reliability.

[0065] Refer to Figure 2As shown, the second aspect of this invention provides a multi-sensor health detection system for soft rail wheels, applied to the aforementioned multi-sensor health detection method for soft rail wheels. The system includes: a feature data acquisition and processing module, used to acquire wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information in real time; preprocess the multi-source data to construct a wheel health feature dataset indexed by the turning angle; an ultrasonic penetration attenuation feature extraction module, used to extract penetration attenuation features reflecting the internal propagation characteristics of the material based on the comparison of ultrasonic echo energy in a dual-time window, and form a robust attenuation characterization through time stabilization processing; an internal defect structural anomaly discrimination module, used to comprehensively evaluate the structural attributes of internal anomalies at turning angles based on ultrasonic echo analysis of ultrasonic echo reflection characteristics and spectral structural stability, and determine the anomaly response type and its spatial distribution state according to the structural attributes of the internal anomalies; and a remaining life prediction and maintenance module, used to perform life constraint assessment for anomaly locations by analyzing surface degradation trends and internal evolution behavior using the wheel health feature dataset, and output maintenance warnings and safety decision guidance based on the life constraint assessment.

[0066] In this implementation plan, multi-sensor collaborative acquisition and unified corner index modeling enable synchronous perception and fusion analysis of wheel surface condition and internal structural information. Ultrasonic penetration attenuation feature extraction and structural anomaly discrimination effectively distinguish between internal structural defects and non-structural disturbances, avoiding misjudgments. Furthermore, a life constraint assessment and graded maintenance decision-making mechanism are introduced to directly transform detection results into actionable early warnings and safety guidelines, thereby improving the accuracy of wheel health assessment, operational safety, and the scientific and practical nature of maintenance decisions.

[0067] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0068] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-sensor health detection method for soft-rail wheels, characterized in that, Includes the following steps: S1 collects wheel diameter distance, surface morphology, ultrasonic echo and operating condition information in real time, preprocesses multi-source data, and constructs a wheel health feature dataset indexed by rotation angle. S2, based on the comparison of ultrasonic echo energy in two time windows, extracts the penetration attenuation characteristics that reflect the propagation characteristics inside the material, and forms a robust attenuation characterization through time stabilization processing; S3, based on ultrasonic echo analysis of ultrasonic echo reflection characteristics and spectral structure stability, comprehensively evaluate the structural properties of internal anomalies at corner positions, and determine the anomaly response type and its spatial distribution state according to the structural properties of internal anomalies; The specific process for comprehensively evaluating the structural properties of internal anomalies at corner locations is as follows: The historical spectral entropy sequence corresponding to the same turning position is read from the wheel health detection database, and the median is taken as the spectral entropy reference value. The logarithm of the peak significance value is taken to obtain the time-domain reflection enhancement. The negative of the time-domain reflection enhancement is used as the exponent for natural exponentiation. The result of natural exponentiation is added by one and the reciprocal is taken to obtain the time-domain structural response value. The frequency domain structure deviation is obtained by subtracting the current frequency domain entropy from the reference value of the frequency domain entropy. The negative of the frequency domain structure deviation is used as the exponent for natural exponentiation. The result of natural exponentiation is added by one and the reciprocal is taken to obtain the frequency domain structure stability response value. The time domain structure response value is multiplied by the frequency domain structure stability response value to obtain the structural defect confidence value. S4, for abnormal locations, uses wheel health feature datasets to analyze surface degradation trends and internal evolution behavior to conduct life constraint assessments, and outputs maintenance warnings and safety decision guidance based on the life constraint assessments.

2. The multi-sensor health detection method for soft rail wheels according to claim 1, characterized in that, The specific process of real-time acquisition of wheel diameter distance, surface morphology, ultrasonic echo, and operating condition information, preprocessing of multi-source data, and constructing a wheel health feature dataset indexed by steering angle is as follows: A line-scan laser camera is fixedly installed on the side of the track to continuously scan the wheel-track contact area and collect the wheel diameter distance and surface morphology height. An ultrasonic probe is installed on one side of the wheel running path to emit ultrasonic pulses into the wheel according to the emission period synchronized with the wheel rotation speed and collect the corresponding ultrasonic A-scan echo time-domain waveform. At the same time, the wheel angular velocity is collected in real time and the ultrasonic propagation speed in the wheel material is obtained. Using wheel angular velocity as a reference, time-to-rotation angle mapping is performed on wheel diameter distance data and surface topography height data. Data from different sampling times are uniformly converted into a sequence indexed by wheel rotation angle. The sampling time of ultrasonic A-scan echo time domain waveform is synchronously corrected according to wheel angular velocity. Median filtering is performed on wheel diameter distance data and surface topography height data to suppress local anomalies caused by track vibration and optical measurement noise. Bandpass filtering was performed on the time-domain waveform of the ultrasonic A-scan echo to remove low-frequency mechanical disturbances and high-frequency noise components. After filtering, the envelope of the ultrasonic A-scan echo time-domain waveform was extracted to obtain the ultrasonic envelope signal. The wheel angular velocity was smoothed by time window processing. The preprocessed wheel diameter distance, surface morphology height, ultrasonic envelope signal, wheel angular velocity, and ultrasonic propagation speed were combined to construct wheel health feature data. Normalization processing was performed on the wheel health feature data to establish a wheel health detection database. The original and preprocessed wheel health feature data were written into the wheel health detection database.

3. The multi-sensor health detection method for soft rail wheels according to claim 1, characterized in that, The specific process of extracting penetration attenuation characteristics reflecting the internal propagation properties of materials based on dual-time-window ultrasonic echo energy comparison, and forming a robust attenuation characterization through time stabilization processing is as follows: For the ultrasonic envelope signal corresponding to each corner position, an incident interface time window and a deep response time window are constructed. The incident interface time window corresponds to the echo interval formed by the ultrasonic wave from the probe to the contact interface on the wheel surface, and the deep response time window corresponds to the echo interval formed by the ultrasonic wave after it propagates inside the wheel and in the direction of the wheel hub. The ultrasonic envelope signals within the incident interface time window and the deep response time window are integrated separately to calculate the ultrasonic energy of the incident interface time window and the ultrasonic energy of the deep response time window. Calculate the ratio of ultrasonic energy in the incident interface time window to ultrasonic energy in the deep response time window, and take the natural logarithm of the ratio to obtain the penetration normalized attenuation value. Based on the sliding time window, the median of the penetration normalized attenuation value at the same corner position is taken to obtain the stable penetration normalized attenuation feature value. The stable penetration normalized attenuation feature value is associated with the wheel health feature data according to the corner index and written into the wheel health detection database.

4. The multi-sensor health detection method for soft rail wheels according to claim 1, characterized in that, The specific process of analyzing the ultrasonic echo reflection characteristics and spectral structure stability based on ultrasonic echo is as follows: Within the deep response time window, the ultrasonic envelope signal at the corresponding corner position is extracted, and the median of the ultrasonic envelope signal amplitude is taken as the noise baseline value. Simultaneously, a local maximum search is performed on the ultrasonic envelope signal within the deep response time window. Among the echo peaks where the ultrasonic envelope signal amplitude is higher than the noise baseline value, the echo peak with the largest amplitude is selected as the significant reflection response value of the deep echo at the current corner position. The ratio of the significant reflection response value to the noise baseline value is calculated to obtain the significance value of the newly added echo peak. Simultaneously, a short-time Fourier transform is performed on the time-domain waveform of the ultrasonic A-scan echo at the same turning position to obtain the time spectrum; within the corresponding deep response time window, the energy of each frequency component in the time spectrum is integrated to obtain the contribution energy value of each frequency component, and the contribution energy values ​​of all frequency components are summed to obtain the total contribution energy value. The ratio of the contribution energy value of each frequency component to the total contribution energy value is calculated to obtain the spectral energy probability of each frequency component. The spectral entropy is obtained by multiplying the spectral energy probability of each frequency component by its own logarithm, summing the products of all frequency components, and taking the opposite number.

5. The multi-sensor health detection method for soft rail wheels according to claim 1, characterized in that, The specific process for determining the anomaly response type and its spatial distribution state based on the structural properties of internal anomalies is as follows: Based on the sliding time window, the median of the confidence values ​​of structural defects at the same corner position is taken to obtain the confidence value of stable structural defects; Along the wheel turning direction, the confidence values ​​of stable structural defects at the N adjacent turning positions are statistically analyzed, and it is determined whether they are all greater than the discrimination threshold. If so, it is considered that there is a continuous structural defect response inside the wheel corresponding to the turning interval. When the number of corner positions where the confidence value of a stable structural defect exceeds the discrimination threshold is less than the number of decision thresholds, a discontinuous structural defect response is considered to exist. If the confidence values ​​of stable structural defects at all corner positions do not exceed the discrimination threshold, it is considered that no structural defect response has been detected. Simultaneously acquire the corresponding wheel health characteristic data and stable penetration normalized attenuation characteristic value, and perform typological determination of abnormal wheel response.

6. The multi-sensor health detection method for soft rail wheels according to claim 5, characterized in that, The specific process of synchronously acquiring the corresponding wheel health feature data and stable penetration normalized attenuation feature value, and classifying the abnormal wheel response, is as follows: When there is a continuous structural defect response, if the change in wheel diameter distance and the change in surface morphology height of adjacent sampling points at the corresponding corner position do not exceed the corresponding abnormal change threshold, the anomaly at the corner position is determined to be an internal structural defect-dominated anomaly; otherwise, it is a structural surface composite anomaly. When there is a discontinuous structural defect response, if the difference between the stable penetration normalized attenuation characteristic value and the median of the historical stable penetration normalized attenuation characteristic values ​​exceeds the deviation threshold, the anomaly corresponding to the corner position is determined to be a non-structural defect anomaly; otherwise, it is determined to be no anomaly. When no structural defect response is detected, if the corresponding wheel diameter distance change, surface morphology height change, and stable penetration normalized attenuation characteristic value are all within the corresponding normal distribution range, it is judged as no abnormality; otherwise, it is a non-structural defect abnormality. The stable penetration normalized attenuation characteristic value and the corresponding anomaly judgment result are written into the wheel health detection database.

7. The multi-sensor health detection method for soft rail wheels according to claim 1, characterized in that, The specific process for assessing life constraints by analyzing surface degradation trends and internal evolution behavior using wheel health feature datasets for abnormal locations is as follows: For abnormal corner positions, the corresponding wheel diameter distance is obtained, and the rate of change of wheel diameter distance is calculated by the difference method of adjacent sampling points to obtain the surface wear degradation rate; Simultaneously, within the incident interface time window at the corresponding corner position, a local maximum search is performed on the ultrasonic envelope signal, and the time corresponding to the echo peak with the largest amplitude is selected as the reference echo time. Within the deep response time window, a local maximum search is performed on the ultrasonic envelope signal, and the time corresponding to the echo peak with the largest amplitude is selected as the deep echo time. The difference between the deep echo time and the reference echo time is calculated to obtain the ultrasonic propagation delay. The ultrasonic propagation delay is multiplied by the ultrasonic propagation speed and divided by two to obtain the equivalent depth value of the internal defect. The rate of change of the equivalent depth value of the internal defect is calculated using the adjacent sampling point difference method as the internal defect propagation rate. The remaining life of surface wear is obtained by dividing the difference between the wheel diameter distance limit and the current wheel diameter distance by the sum of the surface wear degradation rate and the smallest positive number. The remaining life of internal defects is obtained by dividing the difference between the internal defect equivalent depth limit and the current internal defect equivalent depth by the sum of the internal defect propagation rate and the smallest positive number. The reciprocals of the remaining life of surface wear and the remaining life of internal defects are taken respectively, and the two reciprocals are added together to obtain the comprehensive degradation rate. The reciprocal of the comprehensive degradation rate is taken and multiplied by a constant two to obtain the fusion remaining life assessment value.

8. The multi-sensor health detection method for soft rail wheels according to claim 1, characterized in that, The specific process of outputting maintenance warnings and safety decision guidelines based on life constraint assessment is as follows: The remaining life assessment value is associated with the corresponding anomaly type according to the corner index and written into the wheel health detection database; if there is more than one abnormal corner position of the wheel, the corner position with the smallest remaining life assessment value is selected for assessment. Integrate remaining lifetime assessment values Compare with the multi-level lifetime assessment thresholds R1 and R2: when When R2 is greater than or equal to 2, the corresponding corner position is determined to be in normal operating condition, and only routine data monitoring is maintained; When R1 ≤ < R2, it is determined that the corresponding corner position is in the early warning maintenance state, a maintenance warning identifier is generated, and a maintenance prompt is pushed. When <When R1, it is determined that the corresponding corner position is in a safety-restricted state, a deactivation recommendation flag is generated, and a wheel replacement prompt is pushed; The remaining life assessment value, the corresponding abnormal location, and the classification results are written into the wheel health detection database, and the corresponding identifiers are output to the maintenance decision interface.

9. A multi-sensor health detection system for soft-rail wheels, employing the multi-sensor health detection method for soft-rail wheels as described in any one of claims 1-8, characterized in that, include: The feature data acquisition and processing module is used to collect wheel diameter distance, surface morphology, ultrasonic echo and operating condition information in real time, preprocess the multi-source data, and construct a wheel health feature dataset indexed by rotation angle. The ultrasonic penetration attenuation feature extraction module is used to extract the penetration attenuation features that reflect the internal propagation characteristics of the material based on the comparison of ultrasonic echo energy in two time windows, and to form a robust attenuation characterization through time stabilization processing. The internal defect structural anomaly discrimination module is used to comprehensively evaluate the structural properties of internal anomalies at corner positions based on ultrasonic echo analysis of ultrasonic echo reflection characteristics and spectral structural stability, and to determine the anomaly response type and its spatial distribution state based on the structural properties of the internal anomalies. The remaining life prediction and maintenance module is used to analyze the surface degradation trend and internal evolution behavior of wheels in abnormal locations using wheel health feature datasets to conduct life constraint assessments, and output maintenance warnings and safety decision guidance based on the life constraint assessment.

Images