Wheel speed self-checking method based on big data platform, electronic equipment and storage medium

Abstract

The invention provides a wheel speed self-checking method based on a big data platform, electronic equipment and a storage medium, and belongs to the technical field of vehicle safety detection. The wheel speed self-checking method comprises the following steps: acquiring an original signal of a wheel speed sensor and vehicle state data in real time; extracting wheel speed characteristic data from the original signal; inputting the wheel speed feature data and the vehicle state data into the trained wheel speed anomaly recognition model for anomaly detection and scoring to obtain an anomaly score; the wheel speed abnormity identification model is a model which is constructed by adopting an isolated forest model and is obtained by training based on wheel speed data under a normal working condition; determining abnormal wheel speed preliminary screening data based on the abnormal score and a preset abnormal detection threshold value; and performing multi-source verification on the abnormal wheel speed preliminary screening data in combination with IMU data to obtain abnormal wheel speed data. According to the method, overall analysis and evaluation are carried out based on the wheel speed data, dynamic and efficient self-inspection can be achieved, and the wheel speed anomaly recognition precision and the fault positioning efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of vehicle safety inspection technology, specifically relating to a wheel speed self-inspection method, electronic equipment, and storage medium based on a big data platform. Background Technology

[0002] During vehicle operation, wheel speed is one of the core parameters for monitoring and controlling vehicle operating status. Accurate wheel speed information is crucial for safe driving, power control, and fault diagnosis. Specifically, accurate wheel speed information is not only the basis for real-time adjustments by active safety systems such as Anti-lock Braking System (ABS) and Electronic Stability Program (ESP); in vehicle fault diagnosis, abnormal changes in wheel speed data can serve as important clues for identifying potential problems in the transmission and braking systems, playing an irreplaceable role in ensuring the safety and reliability of the vehicle throughout its entire lifecycle.

[0003] However, traditional wheel speed detection methods have significant technical limitations, making it difficult to meet the comprehensive and timely requirements of modern vehicles for wheel speed monitoring. Traditional methods typically only collect wheel speed data for a single vehicle at a specific moment, failing to track long-term trends in wheel speed changes. They also struggle to combine wheel speed data from multiple operating conditions and environments for overall analysis and evaluation, resulting in the inability to identify progressive performance degradation issues of wheel speed sensors or related components in advance.

[0004] Furthermore, when wheel speed sensors malfunction, or wheel speed data becomes abnormal due to factors such as tire slippage or wear, traditional detection methods lack effective real-time judgment and processing mechanisms. They often rely on manual inspection or the vehicle to be brought to a repair shop to locate the problem, which may pose potential risks to the vehicle's driving safety during this period.

[0005] Therefore, how to conduct overall analysis and evaluation based on wheel speed data, achieve dynamic and efficient self-inspection, and improve the accuracy of wheel speed anomaly identification and fault location efficiency has become an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to solve the problems existing in the prior art and provide a wheel speed self-testing method, electronic device and storage medium based on a big data platform, which is effective.

[0007] This invention is achieved through the following technical solution:

[0008] The first aspect of the invention provides a wheel speed self-checking method based on a big data platform, the wheel speed self-checking method comprising:

[0009] The system acquires raw signals from wheel speed sensors and vehicle status data in real time, including IMU data.

[0010] Wheel speed feature data are extracted from the original signal;

[0011] The wheel speed feature data and the vehicle state data are input into the trained wheel speed anomaly recognition model for anomaly detection and scoring to obtain an anomaly score; the wheel speed anomaly recognition model is constructed using an isolated forest model and trained based on wheel speed data under normal operating conditions.

[0012] Based on the anomaly score and the preset anomaly detection threshold, the initial screening data for abnormal wheel speeds are determined;

[0013] The abnormal wheel speed data is obtained by multi-source verification of the initial screening data using the IMU data.

[0014] Optionally, before extracting wheel speed feature data from the original signal, the wheel speed self-testing method further includes: performing hardware error correction on the original signal;

[0015] The hardware error correction includes:

[0016] The physical gap between the Hall sensor and the trigger wheel is adjusted by using shims, and the original signal is compensated for the physical gap.

[0017] A thermistor is connected in parallel with the Hall sensor, and the circuit gain is adjusted according to the temperature.

[0018] A temperature-sensitivity lookup table is established. Based on the ambient temperature collected in real time by the on-chip temperature sensor or the estimated ambient temperature, the signal amplitude is corrected in real time by referring to the temperature-sensitivity lookup table.

[0019] Optionally, wheel speed feature data is extracted from the original signal, including:

[0020] The wheel speed is calculated based on the pulse frequency of the original signal using the frequency measurement method, period measurement method, or M / T method.

[0021] The wheel speed data is corrected using Kalman filtering.

[0022] Optionally, after extracting wheel speed feature data from the original signal, the wheel speed self-testing method further includes:

[0023] Based on a comparison of GPS speed and wheel speed, tire diameter deviations are periodically corrected; or...

[0024] By fusing four-wheel speed data using Kalman filtering, the tire diameter deviation of each wheel is dynamically corrected in real time.

[0025] Optionally, before acquiring the raw signals from the wheel speed sensors and vehicle status data in real time, the wheel speed self-testing method further includes: training the isolated forest model;

[0026] The training process of the isolated forest model includes:

[0027] Collect normal wheel speed data of multiple vehicles under various operating conditions;

[0028] Construct a multidimensional feature vector that includes wheel speed basic features, temporal features, and correlation features;

[0029] Set the number of trees, the size of the sample subset, and the feature sampling method, and then train the model.

[0030] Generate a threshold mapping table and calculate the statistical quantiles of the abnormal scores of all normal samples as the initial abnormal detection threshold.

[0031] Optionally, the setting and adjustment of the anomaly detection threshold is dynamic, specifically including: selecting the corresponding threshold from a preset threshold mapping table based on the operating condition interval to which the real-time vehicle speed belongs; and / or, periodically updating based on the statistical quantile of the anomaly scores of recent normal sample data; and / or, co-evolving with the incremental learning process of the isolated forest model.

[0032] Optionally, before extracting wheel speed feature data from the original signal, the wheel speed self-checking method further includes:

[0033] Cluster analysis is performed on the anomalies to identify at least one typical anomaly pattern.

[0034] Each anomaly pattern is associated with a potential cause of failure, forming fault mode knowledge;

[0035] Based on the aforementioned fault mode knowledge, a fault mode library containing abnormal characteristics and fault causes is generated.

[0036] Optionally, after obtaining the abnormal wheel speed data, the wheel speed self-checking method further includes:

[0037] Extract abnormal features from the abnormal wheel speed data;

[0038] The cause of the fault is obtained by matching the abnormal characteristics from the fault mode library, and maintenance suggestions are generated.

[0039] Optionally, after obtaining the abnormal wheel speed data, the wheel speed self-checking method further includes:

[0040] Based on a preset number of wheel speed feature data and vehicle status data collected during the wheel speed self-inspection process, the wheel speed anomaly identification model is trained and its parameters are updated.

[0041] In a second aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program, when executed by the processor, causes a device equipped with the processor to perform the wheel speed self-test method based on a big data platform as described in any of the preceding claims.

[0042] A third aspect of the present invention provides a storage medium storing a computer program that runs on a computer and, when running, causes the computer to perform the wheel speed self-checking method based on a big data platform as described in any of the preceding claims.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. Through multi-source data fusion and machine learning optimization, the accuracy of wheel speed anomaly detection has been effectively improved, the fault identification accuracy has been increased to over 98%, and the false alarm rate has been reduced to below 1%.

[0045] 2. Faster response speed, higher fault location efficiency, edge preprocessing can achieve millisecond-level response, and cloud analysis supports long-term trend prediction;

[0046] 3. Predict potential faults in advance, reduce the frequency of manual inspection, extend the sensor replacement cycle, and reduce operation and maintenance costs by more than 30%;

[0047] 4. Applicable to various wheel speed sensors such as Hall effect and magnetoelectric, supports data access from different vehicle models, and has high compatibility. Attached Figure Description

[0048] The above and other objects, features, and advantages of the present invention will become more apparent from the more detailed description of the embodiments of the invention in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0049] Figure 1 A flowchart illustrating a wheel speed self-checking method based on a big data platform, provided in an embodiment of the present invention;

[0050] Figure 2 A logic diagram of a wheel speed self-checking method based on a big data platform provided in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0053] To address the limitations of existing methods in tracking long-term trends in vehicle wheel speed and incorporating wheel speed data from multiple operating conditions and environments for comprehensive analysis and evaluation, this invention proposes a wheel speed self-checking method based on a big data platform. This method performs comprehensive analysis and evaluation based on wheel speed data, achieving dynamic and efficient self-checking and improving the accuracy of wheel speed anomaly identification and fault location efficiency. The invention will be further described in detail below with reference to the accompanying drawings.

[0054] To facilitate understanding of this invention, a detailed description of a wheel speed self-checking method based on a big data platform, as disclosed in this embodiment, is provided first. The main execution entities of this wheel speed self-checking method based on a big data platform primarily include the vehicle-side Electronic Control Unit (ECU) and the cloud-based big data platform. The two interact and coordinate functions via a network. The vehicle-side ECU, as an edge computing node, is responsible for tasks with extremely high real-time and security requirements and is the core execution entity enabling the method to respond in real time. The cloud-based big data platform, as a background computing and optimization center, is responsible for tasks with high computing power and storage requirements and non-real-time requirements. It is the execution entity enabling continuous optimization and intelligent diagnostics of the method.

[0055] The following is for reference Figure 1 This invention describes a wheel speed self-checking method based on a big data platform according to an embodiment of the present invention.

[0056] Figure 1 This is a flowchart illustrating the wheel speed self-checking method based on a big data platform according to an embodiment of the present invention. Figure 1 As shown, the wheel speed self-checking method based on a big data platform in this invention includes at least the following steps S100 to S500.

[0057] Step S100: Acquire the raw signals from the wheel speed sensors and vehicle status data in real time, wherein the vehicle status data includes IMU data.

[0058] The wheel speed sensor can be a Hall effect sensor, and the raw signals collected by the Hall effect sensor, such as the pulse width modulation (PWM) signal, as well as vehicle status data, such as vehicle speed and acceleration, can be acquired in real time via a Controller Area Network (CAN) bus.

[0059] Specifically, the signal acquisition circuit that acquires signals using a Hall effect sensor includes:

[0060] Power supply regulation and protection circuit: The vehicle power supply (12V / 24V) needs to be regulated to the operating voltage of the Hall element (such as 5V or 3.3V) through a low dropout regulator (LDO), and a transient voltage suppressor (TVS) diode should be added to prevent surge voltage (such as transient interference in the ISO 7637-2 standard).

[0061] Signal conditioning circuit: An RC low-pass filter (e.g., 100Ω + 0.1μF) is used to suppress high-frequency noise (such as PWM interference and electromagnetic radiation); a Schmitt trigger converts the analog signal output by the Hall element into a standard square wave to eliminate false triggering caused by noise. When the voltage of the back-end MCU interface does not match the sensor output (e.g., the sensor outputs 0-12V, and the MCU receives 0-5V), a voltage divider resistor or a level conversion chip is also provided.

[0062] The sensor cable uses twisted-pair shielded cable, with the shielding layer grounded at one end (usually connected to the ECU housing) to reduce electromagnetic interference.

[0063] Preferably, the Hall effect sensor can be a differential Hall effect sensor, which transmits differential signals through two signal lines to suppress common-mode interference.

[0064] The ideal gap between the Hall sensor and the trigger wheel is generally 0.5-2mm. If the gap is too large, the signal amplitude will decrease, and if it is too small, it will be easily affected by mechanical vibration. Therefore, the gap can be adjusted by using shims.

[0065] Step S200: Extract wheel speed feature data from the original signal.

[0066] The wheel speed characteristic data can include the original wheel speed value, the difference between the left and right wheel speeds, and the wheel speed change rate (Δv / Δt); the time series characteristics include statistics such as mean, variance, standard deviation, and peak value calculated based on a sliding window.

[0067] Step S300: Input the wheel speed feature data and the vehicle state data into the trained wheel speed anomaly recognition model for anomaly detection and scoring to obtain an anomaly score; the wheel speed anomaly recognition model is a model constructed using the isolated forest algorithm and trained based on wheel speed data under normal operating conditions.

[0068] The Isolation Forest model works by learning the distribution characteristics of wheel speed data under normal operating conditions during the training phase. It assesses the anomaly level by calculating the average path length of the input sample within the set of "isolated trees" constructed by the model. The more easily a data point is isolated (i.e., the shorter the path length), the more likely it is to be an anomaly.

[0069] The model outputs an anomaly score for each input feature vector. This score is a value between 0 and 1. The closer the score is to 1, the higher the probability that the wheel speed data at that moment is abnormal; the closer the score is to 0, the higher the probability that it belongs to the normal mode.

[0070] Step S400: Based on the anomaly score and the preset anomaly detection threshold, determine the initial screening data for abnormal wheel speeds.

[0071] The anomaly score obtained in step S300 is compared with a preset anomaly detection threshold. This threshold can be a fixed value, but is preferably a dynamic threshold that can be adaptively adjusted according to operating parameters such as real-time vehicle speed and road surface adhesion coefficient. For example, under high-speed conditions, the allowable range of normal fluctuations is wider, so the threshold can be increased accordingly to avoid false alarms. When the anomaly score of a sample point exceeds the preset threshold, the wheel speed data at that moment is determined to be initially anomaly, and it is marked together with the timestamp, corresponding vehicle status, and other information as preliminary screening data for abnormal wheel speeds.

[0072] Step S500: Combine IMU data to perform multi-source verification on the initial screening data of abnormal wheel speeds to obtain abnormal wheel speed data.

[0073] This step involves the final confirmation and correction of the initial screening results. By introducing independent data sources for cross-validation, the accuracy of fault diagnosis is greatly improved, and the false positive rate is effectively reduced. This includes:

[0074] First, data synchronization and acquisition are performed: acquire inertial measurement unit data that is synchronized with the initial screening data of the abnormal wheel speed, mainly including longitudinal acceleration, lateral acceleration and yaw rate.

[0075] Then, logical consistency verification is performed: the wheel speed data is compared with the IMU data to analyze their physical consistency. Specific verification logic includes, but is not limited to:

[0076] Verify slippage condition: If the initial screening data indicates that the wheel speed of a certain wheel has increased abnormally, but the longitudinal acceleration of the IMU does not show a corresponding drastic change, then the abnormality is determined to be caused by wheel slippage, rather than sensor failure.

[0077] Verify road impact: If the initial screening data shows a sudden change in wheel speed, and the IMU also detects a significant impact signal in the vertical acceleration, then the anomaly is determined to be caused by road bumps (such as driving over potholes) and is excluded.

[0078] Verifying steering conditions: When the vehicle is turning, if the initial screening data indicates an abnormal difference in the left and right wheel speeds, the reasonableness of this difference is verified by measuring the yaw rate of the IMU. If it matches the yaw rate, it is considered a normal steering condition and is excluded.

[0079] Finally, the final abnormal data is generated: After the above multi-source verification, the preliminary screening data that cannot be explained by normal vehicle dynamics, i.e., after excluding conditions such as slippage, impact, and normal steering, is finally confirmed as abnormal wheel speed data. This part of the data corresponds to real faults with high confidence (such as sensor damage, wiring faults, bearing damage, etc.) and can be used to trigger the final fault alarm and maintenance suggestions.

[0080] This invention provides a general-purpose wheel speed self-checking method based on a big data platform. By fusing wheel speed signals, vehicle inertial measurement data, and historical operating condition information for cross-dimensional collaborative analysis, the system can effectively distinguish between genuine sensor malfunctions and signal anomalies caused by complex driving conditions (such as slippage and bumps). This mechanism fundamentally reduces the system's false alarm rate, ensuring the accuracy and high reliability of alarm information, and providing drivers with more reliable safety assurance.

[0081] In some embodiments of the present invention, before extracting wheel speed feature data from the original signal, the wheel speed self-testing method further includes: performing hardware error correction on the original signal;

[0082] The hardware error correction process includes:

[0083] The physical gap between the Hall sensor and the trigger wheel is adjusted by using shims, and the original signal is compensated for the physical gap.

[0084] A thermistor is connected in parallel with the Hall sensor, and the circuit gain is adjusted according to the temperature.

[0085] A temperature-sensitivity lookup table is established. Based on the ambient temperature collected in real time by the on-chip temperature sensor or the estimated ambient temperature, the signal amplitude is corrected in real time by referring to the temperature-sensitivity lookup table.

[0086] Specifically, a compensation coefficient is written into the ECU, such as a 5% signal amplitude attenuation for every 0.1mm increase in the gap. This compensation coefficient enables software gain compensation, thereby calibrating the installation gap between the Hall sensor and the trigger wheel.

[0087] The sensitivity of a Hall element varies with temperature, with a typical temperature drift coefficient of -0.1% / ℃. Therefore, temperature drift compensation can be achieved through the following methods:

[0088] Hardware: Parallel thermistor, which adjusts the circuit gain according to temperature.

[0089] Software: Establish a temperature-sensitivity lookup table and adjust the signal amplitude in real time by estimating the signal using the on-chip temperature sensor or ambient temperature.

[0090] By systematically correcting the hardware errors of the Hall effect sensor, this solution improves the accuracy and reliability of the data from the signal source. Specifically, gap compensation eliminates signal attenuation caused by mechanical installation deviations, and temperature compensation overcomes the inherent temperature drift characteristics of electronic components. These measures together ensure the quality of the original signal upon which subsequent processing is based, laying a solid foundation for high-precision anomaly detection and significantly reducing misjudgments caused by uncertainties in the hardware itself.

[0091] In some embodiments of the present invention, wheel speed feature data is extracted from the original signal, including:

[0092] The wheel speed is calculated based on the pulse frequency of the original signal using the frequency measurement method, period measurement method, or M / T method.

[0093] Kalman filtering is used to correct wheel speed data and remove high-frequency noise.

[0094] Specifically, when a Hall effect sensor is used for wheel speed, the sensor acquires a pulse signal. The pulse signal can be processed using the following three methods:

[0095] In high-wheel-speed scenarios, the pulse frequency is high, and the wheel speed can be calculated using the frequency measurement method (M method) to reduce counting errors. The calculation formula is as follows:

[0096]

[0097] Where n is the wheel speed, z is the number of trigger wheel teeth, t is a fixed time window, such as 100ms, and N is the pulse count within the fixed time window.

[0098] In low wheel speed scenarios, the pulse period is long. The wheel speed can be calculated using the period measurement method (T method) to improve the accuracy of time measurement. The calculation formula is as follows:

[0099]

[0100] Where T is the time interval between adjacent pulses.

[0101] When the rotational speed fluctuates within a wide range, the wheel speed can be calculated by combining the two methods mentioned above (M / T method). At this time, it is necessary to measure the number of pulses and the time interval simultaneously.

[0102] Preferably, before calculating the wheel speed, the collected data can be filtered by software: using moving average filtering (such as averaging of 5 samples) or median filtering to eliminate accidental pulse loss or false triggering (such as false pulses generated by a stone hitting the trigger wheel).

[0103] This solution integrates multiple wheel speed calculation methods and Kalman filtering technology to achieve high-precision, interference-resistant wheel speed estimation across the entire vehicle speed range. The combination of the frequency / period / M / T method ensures the optimal solution for calculations under various operating conditions from low to high speeds, while Kalman filtering utilizes the vehicle dynamics model to effectively smooth measurement noise and provide the optimal estimate, thereby improving the accuracy and stability of the wheel speed data itself and enhancing the reliability of subsequent analysis.

[0104] In some embodiments of the present invention, after extracting wheel speed feature data from the original signal, the wheel speed self-testing method further includes:

[0105] Based on a comparison of GPS speed and wheel speed, tire diameter deviations are periodically corrected; or...

[0106] By fusing the wheel speed data of the four wheels using Kalman filtering, the tire diameter deviation of each wheel is dynamically corrected in real time.

[0107] Specifically, every certain distance the vehicle travels (e.g., 1000km), the tire diameter parameters are updated via manual input or the On-Board Diagnostics (OBD) interface, or automatically calibrated by comparing GPS speed with wheel speed. For example, if the GPS speed is stable, the deviation between the wheel speed and the GPS speed is calculated, and the tire diameter coefficient is corrected accordingly.

[0108] By utilizing the consistency of four-wheel speeds during straight-line driving (non-slipping state), the wheel speed data of the four wheels is fused through Kalman filtering, and the tire diameter deviation of each wheel is dynamically corrected using the following formula:

[0109]

[0110] Among them, D i Let D0 be the diameter after the i-th round of correction, and v be the initial diameter. i The linear velocity v is the wheel speed calculated from the wheel speed. GPS The actual speed obtained from GPS.

[0111] This solution provides an active correction mechanism to address the dynamic factor of tires. By comparing GPS speed benchmarks or fusing data from multiple tires, it can correct for tire diameter changes caused by factors such as tire pressure and wear in real time or periodically. This directly eliminates a significant source of systematic error in wheel speed calculation, ensuring the long-term accuracy of parameters such as vehicle speed derived from wheel speed and avoiding systematic false alarms caused by changes in tire condition.

[0112] In some embodiments of the present invention, before acquiring the raw signals from the wheel speed sensors and vehicle status data in real time, the wheel speed self-testing method further includes training the isolated forest model.

[0113] Isolation Forest is an efficient unsupervised anomaly detection algorithm suitable for handling dynamic data. The training process of the Isolation Forest model is explained in detail below using wheel speed sensor data.

[0114] A1. Collect normal wheel speed data of vehicles under various operating conditions.

[0115] Collect wheel speed data of vehicles under various normal operating conditions, such as constant speed, acceleration, deceleration, and turning at low, medium, and high speeds, to ensure coverage of different vehicle speeds (e.g., 5-120 km / h), road conditions (dry, wet), and driving modes.

[0116] The data duration must cover at least 200 kilometers of driving data, or continuously collect more than 10 hours of normal operation data.

[0117] A2. Data Cleaning and Denoising

[0118] The data cleaning and repair process mainly includes the following steps:

[0119] Outlier removal: Based on the principles of vehicle dynamics, a reasonable data range is set, and obvious erroneous data that exceeds physical limits (e.g., wheel speed greater than 300 km / h) is identified and removed.

[0120] Missing value handling: Differentiated handling strategies are adopted for missing data in the data sequence. For short-term missing data, sliding window mean or linear interpolation algorithms are used for patching; for missing segments with a longer duration, the missing data segments are directly removed.

[0121] Residual feature introduction: Given that Kalman filtering has been applied for data prediction, the residual between the predicted and measured values ​​is calculated and incorporated as an auxiliary feature into subsequent analysis. Generally, a larger residual indicates a higher probability of anomalies in the data point.

[0122] A3. Construct a multidimensional feature vector that includes wheel speed basic features, time sequence features, and correlation features.

[0123] The basic features include the original wheel speed value, the difference between the left and right wheel speeds, and the wheel speed change rate (Δv / Δt); the time-series features include statistics such as mean, variance, standard deviation, and peak value calculated based on a sliding window (e.g., 50 sampling points); and the associated features include combinations of vehicle speed (obtained from wheel speed integration or GPS), acceleration, and steering wheel angle. The resulting multidimensional feature vector can be represented, for example, as [left wheel speed, right wheel speed, vehicle speed, wheel speed difference, change rate].

[0124] The conversion from wheel speed to vehicle speed can be based on the relationship between wheel linear velocity v and wheel speed n, and can be calculated using the following formula:

[0125]

[0126] Where D is the tire diameter.

[0127] It should be noted that tire pressure and wear will affect the tire diameter D, so regular calibration is necessary. For specific methods, please refer to the description above, which will not be elaborated here.

[0128] A4. Set the number of trees, the size of the sample subset, and the feature sampling method, and then train the model.

[0129] In the training phase of the isolated forest model, the model parameters are designed and configured first. Key parameters include:

[0130] Number of trees (n_estimators): Usually set to 100 to 200 trees, but can be reduced to 50 trees when the amount of training data is small (less than 10,000 samples);

[0131] Sample subset size (max_samples): It is recommended to set it to 256 or 512 to ensure that the training data for each tree is both representative and diverse, while not exceeding one-third of the total sample size.

[0132] Feature sampling is performed randomly and is automatically processed by the algorithm, eliminating the need for manual feature pre-selection.

[0133] The model construction and training process is as follows: The preprocessed n×d multidimensional feature matrix (where n is the number of samples and d is the feature dimension) is used as the model input. During training, each isolated tree randomly selects a feature and its split point to recursively partition a subset of samples until each sample is isolated or the partition reaches a preset maximum depth (usually set to log2(max_samples)). The model calculates the average path length of each sample across all trees and converts it into an anomaly score s(x,m) according to the following formula. The closer the score is to 1, the higher the probability that the sample is an anomaly.

[0134]

[0135] Where h(x) is the path length of sample x in the isolated tree, E(h(x)) is the average path length of sample x in all isolated trees, c(m) is the average path length of the tree (theoretical correction value) when given the number of samples m, and m is the number of samples used to construct a single isolated tree.

[0136] A5. Generate a threshold mapping table and calculate the statistical quantiles of the abnormal scores of all normal samples as the initial abnormal detection threshold.

[0137] Specifically, the process of generating the threshold mapping table is as follows:

[0138] First, the normal wheel speed data is segmented according to vehicle speed. For example, each segment is divided into different speed ranges, with 10 km / h as the interval.

[0139] Then, for all historical data samples within each speed range, perform the following operations:

[0140] The feature engineering method is exactly the same as that used in the real-time detection module to extract the feature vector (such as wheel speed, wheel speed difference, rate of change, etc.) for each sample.

[0141] These feature vectors are input into the already trained Isolation Forest model to calculate the anomaly score for each historical normal sample.

[0142] Finally, within each speed range, a statistical analysis is performed on the set of abnormal scores for all normal samples, calculating the specified statistical quantiles (e.g., 95th or 99th percentiles) for all abnormal scores within that range. This means that at that speed, 95% or 99% of the normal samples have abnormal scores below this value. Thresholds are set based on speed segments, and each speed range is associated with the calculated thresholds, forming a mapping table of <speed range, dynamic threshold>. For example, the threshold is slightly lower for low-speed ranges and slightly higher for high-speed ranges, thus avoiding misjudgments of normal fluctuations.

[0143] Calculate the statistical quantiles of abnormal scores for all normal samples, for example, by initializing the threshold to the 95th percentile of scores for all normal samples.

[0144] If there are a small number of known abnormal samples, the abnormality detection threshold can be set to the maximum value or mean of the normal sample scores plus 3 times the standard deviation.

[0145] This solution details the scientific training process for the Isolation Forest model. Multi-condition and multi-vehicle data ensures comprehensive model learning, multi-dimensional feature engineering enables the model to capture complex anomaly patterns, and the setting of statistical quantile thresholds provides an objective, data-driven benchmark for anomaly detection. This entire process collectively guarantees that the trained model possesses high generalization ability and accurate initial detection performance.

[0146] Preferably, during actual testing, when a certain amount of new normal data accumulates (such as 1,000 new samples per week), the model can be retrained or an incremental isolated forest algorithm can be used to adapt to vehicle aging or environmental changes.

[0147] This solution endows the wheel speed self-monitoring system with the ability to continuously evolve. Through periodic or triggered model updates, the system can utilize the latest operational data to optimize its parameters, thereby adapting to vehicle aging, component performance degradation, or newly emerging fault modes. This mechanism ensures the long-term effectiveness of the system's detection performance, achieving an upgrade from "one-time deployment" to "lifelong learning."

[0148] In rare normal operating conditions (such as driving on snow), the data can be manually labeled and added to the training set to reduce false alarms.

[0149] In some embodiments of the present invention, before extracting wheel speed feature data from the original signal, the wheel speed self-checking method further includes:

[0150] Cluster analysis is performed on the anomalies to identify at least one typical anomaly pattern.

[0151] Each anomaly pattern is associated with a potential cause of failure, forming fault mode knowledge;

[0152] Based on the aforementioned fault mode knowledge, a fault mode library is generated that includes fault characteristics, diagnostic conclusions, and maintenance suggestions.

[0153] After the wheel speed anomaly identification model based on the isolated forest model outputs anomaly points, the system further performs in-depth analysis on the anomaly points. This process mainly includes two steps: anomaly pattern clustering and physical feature attribution.

[0154] B1. Anomaly Pattern Clustering Analysis

[0155] To identify typical failure modes from a massive number of anomalies, the system employs unsupervised machine learning methods to perform cluster analysis on the anomalies. The specific steps are as follows:

[0156] Anomaly Pattern Visualization: First, to facilitate manual analysis and understanding, the system uses dimensionality reduction algorithms (such as t-Distributed Stochastic Neighbor Embedding (t-SNE) or Principal Component Analysis (PCA)) to project high-dimensional anomaly feature data into a two- or three-dimensional space. By visualizing its distribution in this low-dimensional space, different anomaly pattern clusters can be intuitively identified. For example, discrete data clusters corresponding to instantaneous changes in wheel speed, continuous deviations in wheel speed, and excessive differences between left and right wheel speeds can be observed. This step also helps determine the appropriate number of categories for subsequent cluster analysis.

[0157] Automatic anomaly pattern identification: Based on visual analysis, the system uses clustering algorithms (such as K-Means) to automatically classify outliers. This process can identify several typical anomaly types, such as:

[0158] Type 1: Wheel speed instantaneous jump anomaly. This type of anomaly is characterized by a sudden and drastic jump in wheel speed value. Its physical causes are usually related to factors such as intermittent poor contact of the wheel speed sensor or instantaneous strong electromagnetic interference to the signal line.

[0159] Type 2: Persistent Deviation Anomaly. This type of anomaly is characterized by a wheel speed difference between the left and right wheels of the same vehicle that consistently exceeds a reasonable threshold. The physical cause may correspond to mechanical or operational problems such as vehicle bearing failure or persistent slippage of one wheel.

[0160] Type 3: Abnormal Wheel Speed ​​Change Rate. This type of anomaly is characterized by a wheel speed change rate that deviates from the normal range under specific operating conditions (such as a sharp decrease in wheel speed when not braking). Its physical cause often points to a malfunction in the anti-lock braking system (ABS).

[0161] B2. Attribution of Abnormal Features

[0162] After completing the clustering of abnormal patterns, the system performs precise definition and physical root cause analysis for each type of abnormality, i.e., feature attribution.

[0163] Multi-source data correlation analysis: The system combines other state data of the vehicle at the same time (such as the ABS operating status signal, the intervention signal of the Electronic Stability Program (ESP), steering wheel angle, etc.) to cross-validate the clustering results and analyze the physical causes corresponding to anomalies. For example, when an "anomaly of rate of change" is identified, if the system simultaneously detects that the ABS is in active boost mode, the anomaly can be attributed to normal system intervention rather than sensor failure, thus effectively avoiding misjudgment.

[0164] Key Feature Statistics and Quantification: By extracting the feature vectors of outliers in each cluster and performing statistical analysis, the core features of this type of anomaly can be quantified. For example, for "persistent deviation anomalies," statistics show that the mean value of its "left and right wheel speed difference" feature is as high as 20 km / h, which is significantly higher than the statistical level of the wheel speed difference in normal samples (usually around 5 km / h). This quantified feature description not only accurately defines this type of anomaly but also provides reliable data support for setting precise anomaly detection thresholds and generating specific maintenance recommendations.

[0165] B3. Associate each abnormal pattern with potential causes of failure to form fault mode knowledge; based on the fault mode knowledge, generate a fault mode library containing fault characteristics, diagnostic conclusions and maintenance suggestions.

[0166] This solution clusters and summarizes anomalies and attributes faults, enabling the system to identify typical fault modes and understand their physical causes. The resulting fault mode library links data patterns with engineering knowledge, providing core knowledge support for subsequent intelligent diagnosis and serving as a prerequisite for precise maintenance.

[0167] In some embodiments of the present invention, the wheel speed self-checking method further includes:

[0168] Cascaded detection: First, wheel speed is predicted using Kalman filtering. The residual is then used as wheel speed feature data and input into an anomaly detection model trained on an isolated forest model to perform wheel speed detection, thereby improving the ability to detect sudden anomalies that the model has not captured.

[0169] Threshold fusion: Combining expert experience thresholds (such as direct alarm for wheel speed difference > 15km / h) and anomaly scores reduces the false judgment rate of a single algorithm.

[0170] Next, refer to Figure 2 This invention describes a wheel speed self-checking method based on a big data platform according to an embodiment of the present invention.

[0171] like Figure 2 As shown, the wheel speed self-checking process of a certain new energy vehicle includes:

[0172] S1, Data Acquisition.

[0173] After the vehicle starts, the PWM signals and vehicle speed data of the four wheels are acquired in real time via the CAN bus.

[0174] S2. Preprocess the signal.

[0175] The signal is filtered to remove instantaneous interference during rapid acceleration; and the signal period for each wheel is calculated (e.g., the signal period for the left front wheel is 5ms).

[0176] S3. Feature Extraction: Extract wheel speed feature data from the original signal.

[0177] This includes parameters such as the extracted signal frequency (200Hz for the left front wheel) and duty cycle (50%).

[0178] By analyzing the resonant frequency in the wheel speed signal using Fourier transform, it can be determined whether the tire pressure is normal.

[0179] In the process of extracting wheel speed feature data from the original signal, signal anomalies and faults are handled.

[0180] Compensation is provided for pulse loss. If multiple consecutive pulse losses are detected (e.g., exceeding a threshold), it is determined to be a sensor malfunction or slippage. The velocity interpolation from the previous moment is used (e.g., v(t) = v(t-1) + a*Δt, where a is the estimated acceleration) to avoid sudden data changes.

[0181] In the ABS / ESP system, data from sensors such as four-wheel speed, steering wheel angle, and yaw rate are combined to logically eliminate abnormal wheel speeds. For example, if the speed of a certain wheel is significantly lower than that of the other wheels, it is judged as slippage, and the data of that wheel is temporarily blocked.

[0182] S4. Model Inference: Input wheel speed feature data and vehicle state data into the trained wheel speed anomaly recognition model to perform anomaly detection and scoring, and calculate the anomaly score; for example, the left front wheel scores 0.8.

[0183] S5. Determine whether the abnormal score exceeds the abnormal detection threshold and determine the initial screening data of abnormal wheel speed.

[0184] If yes, combine the initial screening data of abnormal wheel speed with IMU data to correct the wheel speed and perform multi-source verification to obtain abnormal wheel speed data; if no, return to the beginning of the detection process.

[0185] For example, a threshold of 0.7 is initially considered abnormal. Comparing the data with other wheels reveals that the right front wheel signal is normal, ruling out road interference. As another example, if the duty cycle of a wheel speed signal deviates from the historical average by 20% while other wheels are normal, it is initially considered abnormal.

[0186] The setting and adjustment of the anomaly detection threshold are dynamic, specifically including: selecting the corresponding threshold from a preset threshold mapping table based on the operating condition interval to which the real-time vehicle speed belongs; and / or, periodically updating based on the statistical quantile of the anomaly scores of recent normal sample data; and / or, co-evolving with the incremental learning process of the isolated forest model.

[0187] S5. Perform fault diagnosis and update the model.

[0188] After obtaining abnormal wheel speed data, abnormal features are extracted from the abnormal wheel speed data; based on the abnormal features, the cause of the fault can be obtained by matching with the fault mode library, thereby enabling rapid fault location and diagnosis.

[0189] Based on the anomaly type obtained from the matching (such as sensor failure, gear wear), repair suggestions are generated. For example, the system prompts "Left front wheel speed sensor signal is abnormal; it is recommended to check the sensor clearance or replace it." After inspection, the repair personnel found that the sensor mounting bolts were loose; after adjustment, the fault disappeared.

[0190] The diagnostic results are fed back to the big data platform to update model parameters and improve detection accuracy. For example, after the diagnostic results are uploaded to the big data platform, the model automatically updates the feature thresholds for sensor gap anomalies.

[0191] By continuously learning and updating the fault mode library, the fault mode library and detection model in this embodiment can continuously optimize themselves as the vehicle operates. This feature enables the system to adapt to changes in sensor characteristics caused by vehicle aging, component wear, seasonal changes, and different driving styles, ensuring the accuracy and effectiveness of the technical solution throughout its entire lifecycle.

[0192] In this embodiment, the machine learning-based intelligent diagnostic function achieves a leap from "detecting anomalies" to "locating the root cause of the fault" and "guiding repairs." Furthermore, the system can also provide fault classification, for example, dividing faults into the following levels:

[0193] • Level 1 Fault: This is the most serious fault level, which may directly affect vehicle driving safety, causing the vehicle to lose control or become inoperable. For example, if multiple wheel speed sensors malfunction simultaneously, the vehicle's anti-lock braking system (ABS), electronic stability program (ESP), and other safety control systems may completely fail. In this case, the big data platform should immediately issue an emergency alert, prompting the owner to stop the vehicle for inspection and repair, and simultaneously push information on nearby repair shops.

[0194] • Level 2 faults: These are second only to Level 1 faults in severity and can significantly impact some of the vehicle's important functions, potentially leading to a decrease in vehicle safety performance. For example, if a single wheel speed sensor malfunctions, causing the vehicle's ABS system to fail, but the vehicle can still brake using the conventional braking system, the big data platform should promptly issue a fault warning, advising the owner to arrange for repairs as soon as possible.

[0195] • Level 3 fault: This is a general fault that may affect some of the vehicle's auxiliary functions or comfort, but will not directly endanger driving safety. For example, a slight deviation in the wheel speed sensor signal may cause the vehicle's cruise control system to be unable to accurately maintain the set speed, but the vehicle's basic driving and safety performance will not be affected. The big data platform can prompt the owner to have the vehicle checked and repaired at their convenience.

[0196] • Level 4 Fault: This is a warning fault, indicating a potential problem with the wheel speed sensor or related components that has not yet significantly affected vehicle performance. For example, big data analysis might reveal an abnormally high operating temperature of the wheel speed sensor or a slight decrease in signal stability. Although the wheel speed data may still be within the normal range, this could indicate an impending sensor failure. In such cases, the big data platform can issue a warning to remind the owner to monitor the vehicle's status and perform regular checks.

[0197] • Level 5 Fault: This refers to a minor fault or an informational fault. It may be due to abnormalities in some minor parameters related to wheel speed, or the system detecting problems with some non-critical components. For example, the wheel speed sensor may be slightly loose, but this does not affect the accuracy of the wheel speed data. The big data platform can provide relevant prompts in the vehicle's routine self-check report, informing the owner to tighten it during the next vehicle maintenance.

[0198] It provides actionable and precise repair suggestions for different levels of faults, enabling repair personnel to quickly locate and eliminate faults, greatly shortening diagnosis time, reducing repair costs and reliance on professional experience, while providing users with a clear and reassuring driving experience.

[0199] In one embodiment of the present invention, when an automotive-grade embedded system (such as an electronic control unit) serves as the real-time execution terminal in a vehicle, and the cloud-based big data platform does not participate in real-time control but instead performs model training and optimization, construction and updating of the fault mode library, and model distribution and configuration management, the feature dimensions used in the real-time inference stage can be simplified, for example, to include only:

[0200] Raw wheel speed: A fundamental signal that directly reflects the rotational speed of each wheel.

[0201] Wheel speed difference: The speed difference between the left and right wheels or wheels on the same axle, used to detect asymmetrical abnormalities such as tire slippage, insufficient tire pressure or bearing failure.

[0202] Wheel speed change rate: The first derivative of wheel speed with time, used to identify dynamic anomalies such as rapid acceleration, rapid deceleration, and signal abrupt changes.

[0203] Through the above pruning, the dimensionality of the feature vectors is significantly reduced. This results in a substantial reduction in the number of floating-point operations and memory accesses required for anomaly score calculation, ensuring extremely low computational latency within a single detection cycle and laying a solid foundation for subsequent rapid response.

[0204] Preferably, during the vehicle power-on and system initialization phase, the key parameters of the pre-trained isolated forest model (e.g., the split feature index, split threshold, and standardized constant c(n) of the leaf node path length for each tree) are pre-loaded from non-volatile memory (such as Flash) into the cache or static random access memory of the embedded processor, thereby enabling millisecond-level anomaly detection response.

[0205] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program that is executed by the processor. When the computer program is executed by the processor, it causes the device equipped with the processor to perform the wheel speed self-test method based on a big data platform as described in any of the above embodiments.

[0206] Next, refer to Figure 3 This describes an example electronic device 100 for implementing the wheel speed self-testing method based on a big data platform according to embodiments of the present invention.

[0207] like Figure 3 As shown, the electronic device 100 includes a processor 110, a memory 120, and a communication interface 130. The processor 110, the memory 120, and the communication interface 130 can be interconnected and communicate via a communication bus 140 and / or other forms of connection mechanisms (not shown).

[0208] It should be noted that Figure 3 The components and structure of the electronic device 100 shown are merely exemplary and not limiting; the electronic device may also have other components and structures as needed.

[0209] Optionally, the communication interface 130 may also include a transmitter and / or a receiver.

[0210] The processor 110 may be a microcontroller unit (MCU), a central processing unit (CPU), a digital signal processor (DSP), a microcontroller, an embedded device, or other processing units with data processing and / or instruction execution capabilities.

[0211] The memory 120 can be various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM), cache memory, synchronous dynamic random access memory (SDRAM), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may also be stored on the computer-readable storage medium, and the memory 120 can execute the program instructions to implement the wheel speed self-test method based on a big data platform described in the above embodiments of the present invention.

[0212] This application also provides a storage medium storing a computer program that runs on a computer. When the computer program runs, it causes the computer to execute the wheel speed self-checking method based on a big data platform as described in any of the above embodiments.

[0213] In summary, the wheel speed self-checking method based on a big data platform provided by this invention is applicable to devices such as automobiles that require real-time monitoring of wheel speed sensor status. It can fully utilize the advantages of the big data platform to perform comprehensive, accurate, and real-time self-checking of vehicle wheel speed, promptly detect abnormalities in wheel speed data, and effectively handle them, thereby improving the safety and reliability of vehicle operation.

[0214] Finally, it should be noted that the above technical solution is only one embodiment of the present invention. For those skilled in the art, based on the application methods and principles disclosed in the present invention, it is easy to make various types of improvements or modifications, and not limited to the methods described in the above specific embodiments of the present invention. Therefore, the methods described above are only preferred and have no limiting significance.

Claims

1. A wheel speed self-checking method based on a big data platform, characterized in that: The wheel speed self-checking method includes: The system acquires raw signals from wheel speed sensors and vehicle status data in real time, including IMU data. Wheel speed feature data are extracted from the original signal; The wheel speed feature data and the vehicle state data are input into the trained wheel speed anomaly recognition model for anomaly detection and scoring to obtain an anomaly score; the wheel speed anomaly recognition model is constructed using an isolated forest model and trained based on wheel speed data under normal operating conditions. Based on the anomaly score and the preset anomaly detection threshold, the initial screening data for abnormal wheel speeds are determined; The abnormal wheel speed data is obtained by multi-source verification of the initial screening data using the IMU data.

2. The wheel speed self-checking method based on a big data platform according to claim 1, characterized in that: Before extracting wheel speed feature data from the original signal, the wheel speed self-testing method further includes: performing hardware error correction on the original signal; The hardware error correction includes: The physical gap between the Hall sensor and the trigger wheel is adjusted by using shims, and the original signal is compensated for the physical gap. A thermistor is connected in parallel with the Hall sensor, and the circuit gain is adjusted according to the temperature. A temperature-sensitivity lookup table is established. Based on the ambient temperature collected in real time by the on-chip temperature sensor or the estimated ambient temperature, the signal amplitude is corrected in real time by referring to the temperature-sensitivity lookup table.

3. The wheel speed self-checking method based on a big data platform according to claim 1, characterized in that: Wheel speed feature data is extracted from the original signal, including: The wheel speed is calculated based on the pulse frequency of the original signal using the frequency measurement method, period measurement method, or M / T method. The wheel speed data is corrected using Kalman filtering.

4. The wheel speed self-checking method based on a big data platform according to claim 1, characterized in that: After extracting wheel speed feature data from the original signal, the wheel speed self-checking method further includes: Based on a comparison of GPS speed and wheel speed, tire diameter deviations are periodically corrected; or... By fusing four-wheel speed data using Kalman filtering, the tire diameter deviation of each wheel is dynamically corrected in real time.

5. The wheel speed self-checking method based on a big data platform according to claim 1, characterized in that: Before acquiring the raw signals from the wheel speed sensors and vehicle status data in real time, the wheel speed self-testing method further includes: training the isolated forest model; The training process of the isolated forest model includes: Collect normal wheel speed data of multiple vehicles under various operating conditions; Construct a multidimensional feature vector that includes wheel speed basic features, temporal features, and correlation features; Set the number of trees, the size of the sample subset, and the feature sampling method, and then train the model. Generate a threshold mapping table and calculate the statistical quantiles of the abnormal scores of all normal samples as the initial abnormal detection threshold.

6. The wheel speed self-checking method based on a big data platform according to claim 5, characterized in that: The setting and adjustment of the anomaly detection threshold is dynamic, specifically including: selecting the corresponding threshold from a preset threshold mapping table based on the operating condition interval to which the real-time vehicle speed belongs; and / or, periodically updating based on the statistical quantile of the anomaly scores of recent normal sample data; and / or, co-evolving with the incremental learning process of the isolated forest model.

7. The wheel speed self-checking method based on a big data platform according to claim 1, characterized in that: Before extracting wheel speed feature data from the original signal, the wheel speed self-checking method further includes: Cluster analysis is performed on the anomalies to identify at least one typical anomaly pattern. Each anomaly pattern is associated with a potential cause of failure, forming fault mode knowledge; Based on the aforementioned fault mode knowledge, a fault mode library containing abnormal characteristics and fault causes is generated.

8. The wheel speed self-checking method based on a big data platform according to claim 7, characterized in that: After obtaining the abnormal wheel speed data, the wheel speed self-checking method further includes: Extract abnormal features from the abnormal wheel speed data; The cause of the fault is obtained by matching the abnormal characteristics from the fault mode library, and maintenance suggestions are generated.

9. The wheel speed self-checking method based on a big data platform according to claim 1, characterized in that: After obtaining the abnormal wheel speed data, the wheel speed self-checking method further includes: Based on a preset number of wheel speed feature data and vehicle status data collected during the wheel speed self-inspection process, the wheel speed anomaly identification model is trained and its parameters are updated.

10. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program executed by the processor, the computer program, when executed by the processor, causing the device on which the processor is installed to perform the wheel speed self-test method based on a big data platform as described in any one of claims 1-9.

11. A storage medium, characterized in that, The storage medium stores a computer program that runs on a computer. When the computer program runs, it causes the computer to perform the wheel speed self-checking method based on a big data platform as described in any one of claims 1-9.

Images