Hub run-out detection method and system based on multi-modal data fusion

Abstract

The invention relates to the technical field of vehicle fault detection, in particular to a hub runout detection method and system based on multi-modal data fusion, and the method comprises the following steps: 1, collecting hub data in real time through a vibration sensor, a vehicle speed sensor, a tire pressure sensor and a steering angle sensor; step 2, carrying out synchronous processing on the collected data; according to the scheme, multi-dimensional data are synchronously collected through multiple types of sensors such as a vibration sensor, a vehicle speed sensor and a tire pressure sensor, preprocessing means such as band-pass filtering, feature extraction and time-frequency domain analysis are combined, and then the fault probability of each modal is fused through the evidence theory. According to the method, the limitation of a single sensor is avoided, for example, interference of abnormal air pressure on vibration data is eliminated through a tire pressure correction coefficient, speed sensitive faults are recognized through vehicle speed-vibration correlation analysis, multi-dimensional accurate judgment of radial / axial runout of the hub is achieved, and the detection accuracy is improved by 30% or above compared with a traditional single-mode scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle fault detection, and in particular to a wheel hub runout detection method and system based on multimodal data fusion. Background Art

[0002] Wheel runout detection is a key technology for ensuring vehicle safety and stability. It is primarily used to identify radial or axial deviations of the wheel during rotation. Traditional detection methods rely primarily on a single sensor (such as a vibration sensor) to collect data, which has the following shortcomings:

[0003] Single detection dimension: The vibration fault is judged only by the vibration signal, which cannot distinguish between interference factors such as road bumps and engine vibration, and is prone to misjudgment.

[0004] Insufficient data correlation: Failure to combine multiple sources of information, such as vehicle speed, tire pressure, and steering angle, makes it difficult to identify speed-sensitive faults or vibration changes caused by abnormal tire pressure.

[0005] Low level of intelligence: Lack of fault mode library and multi-modal fusion algorithm, unable to automatically distinguish radial / axial runout types, and the early warning mechanism lags, making it difficult to accurately locate faults in the early stages.

[0006] To this end, the present invention proposes a wheel hub runout detection method and system based on multimodal data fusion. By synchronously collecting data from multiple types of sensors such as vibration, vehicle speed, and tire pressure, and combining bandpass filtering, evidence theory fusion algorithm, and fault pattern matching technology, it achieves multi-dimensional precise detection of wheel hub runout, automatic identification of fault types, and graded warning, effectively improving detection accuracy and safety, and reducing misjudgment rate and maintenance costs. Summary of the Invention

[0007] Technical problems solved: single detection dimension, insufficient data correlation, and low level of intelligence.

[0008] In view of the deficiencies in the prior art, the present invention provides a wheel hub runout detection method and system based on multimodal data fusion, thereby solving the technical problems mentioned in the background technology.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0010] A wheel hub runout detection method based on multimodal data fusion comprises the following steps:

[0011] Step 1: Collect wheel hub data in real time through vibration sensors, vehicle speed sensors, tire pressure sensors, and steering angle sensors;

[0012] Step 2: Synchronize the collected data to ensure that the timestamps of each modal data are aligned and the error is within the specified range;

[0013] Step 3: Preprocess the data using bandpass filtering and feature extraction to obtain time domain, frequency domain and correlation features;

[0014] Step 4: Normalize the preprocessed data, use evidence theory to assign basic probability distribution of each modal data, and calculate the fusion fault probability through synthesis rules;

[0015] Step 5: Determine whether the wheel hub has a runout fault based on the fused fault probability. If the fault probability exceeds the set threshold, further identify the fault type and output the result.

[0016] In a possible implementation, in the data synchronization processing step, a high-precision clock module is used to add a timestamp to each sensor data, and timestamp alignment is achieved through linear interpolation or synchronization signal calibration.

[0017] In one possible implementation, in the data preprocessing, the bandpass filter is set to 5-50 Hz to remove high-frequency noise and low-frequency interference; the feature extraction includes calculating the root mean square (RMS) and kurtosis of the vibration data, the stability index of the vehicle speed data, and the correlation coefficient between tire pressure and vibration.

[0018] In one possible implementation, during the evidence theory fusion process, a failure probability model of each modal data is constructed through historical data and expert experience to determine the initial value of the basic probability distribution.

[0019] In one possible implementation, a wheel hub runout detection system based on multimodal data fusion, which is implemented in the above method, includes a perception layer, a transmission layer, a processing layer, and an application layer;

[0020] Perception layer: includes vibration sensors, vehicle speed sensors, tire pressure sensors, and steering angle sensors, used to collect wheel-related data in real time;

[0021] Transport layer: CAN bus is used to realize vehicle intranet data transmission, 4G / 5G module is used for remote data communication, and SD card is used for local data storage;

[0022] Processing layer: It has a data preprocessing module for filtering and feature extraction of collected data; a multimodal fusion module for data normalization and evidence theory fusion; and a fault diagnosis module for determining the fault type based on the fusion results.

[0023] Application layer: includes on-board terminals for real-time display of test results and graded warnings; cloud platform for storing diagnostic reports and performing data analysis.

[0024] In a possible implementation, the sensors of the perception layer have an automatic calibration function, and perform self-inspection and calibration when the vehicle is started to ensure the accuracy of data collection.

[0025] In one possible implementation, the multimodal fusion module and fault diagnosis module of the processing layer use hardware acceleration chips or cloud computing platforms for calculation to improve detection efficiency, achieve processing of more than 1,000 groups of data points per second, and a single detection takes less than 200ms.

[0026] In one possible implementation, two-way data interaction is achieved between the vehicle-mounted terminal and the cloud platform of the application layer. The vehicle-mounted terminal can receive optimization algorithms and update information from the cloud platform, and the cloud platform can obtain the detection data of the vehicle-mounted terminal in real time for continuous optimization of the diagnosis model.

[0027] Beneficial effects compared with existing technologies:

[0028] 1. In this solution, multi-dimensional data (such as vibration acceleration, wheel speed, and tire pressure) is collected synchronously through multiple sensors, including vibration sensors, vehicle speed sensors, and tire pressure sensors. Preprocessing methods such as bandpass filtering, feature extraction, and timely frequency domain analysis are then used to fuse the fault probabilities of each modality using evidence theory. This method avoids the limitations of a single sensor. For example, the tire pressure correction coefficient is used to eliminate the interference of abnormal air pressure on vibration data. Speed-sensitive faults are identified through vehicle speed-vibration correlation analysis. This allows for accurate multi-dimensional determination of wheel hub radial / axial runout, improving detection accuracy by over 30% compared to traditional single-modal solutions.

[0029] 2. This solution automatically identifies specific fault types, such as radial runout (RMS > 0.5g and main frequency matching wheel speed frequency) or axial runout (kurtosis > 3.5 and stable steering angle), by establishing a fault pattern library (including typical vibration feature thresholds and similarity matching algorithms). Based on the fused fault probability (> 0.6 triggers an early warning), a graded response is implemented: a mild early warning (yellow icon + sound prompt) guides the driver to conduct a prompt inspection, while a severe early warning (red icon + SMS push) forces the driver to stop the vehicle for repairs. This intelligent mechanism can identify more than 90% of potential wheel hub failures in advance, significantly reducing the safety risks caused by wheel hub runout during high-speed driving.

[0030] 3. In this solution, the entire process from data acquisition (automatic sensor calibration, 1kHz high-frequency sampling), pre-processing (5-50Hz Butterworth filtering) to fusion diagnosis (fast calculation of Dempster synthesis rules) is automated, and more than 1,000 sets of data points can be processed per second, with a single detection time of less than 200ms. At the same time, the system supports automatic generation of diagnostic reports containing data trend charts, fault location and maintenance suggestions every 100 kilometers, and synchronizes them to the cloud for remote viewing by car owners. For example, in the case, the detection and warning of the left front wheel radial runout fault only took 8 seconds, and the report was directly located to the specific wheel, which greatly shortened the troubleshooting time and improved maintenance efficiency by more than 50%. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings.

[0032] Figure 1 is a flow chart of the method steps of the present invention;

[0033] Figure 2 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0034] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can be implemented in various forms, so the present invention is not limited to the embodiments described below.

[0035] The technical solution in the embodiments of the present application is to solve the problems of the above-mentioned background technology, and the overall idea is as follows:

[0036] Example 1:

[0037] Please refer to Figure 1 As shown, this embodiment introduces a wheel hub runout detection method based on multimodal data fusion, and the method includes the following steps:

[0038] 1. System Architecture and Data Acquisition: Data acquisition is achieved through multiple types of sensors: triaxial MEMS accelerometers (range ±16g, sampling frequency ≥1kHz) are installed on the steering wheel and wheel hubs, acquiring vehicle speed and wheel speed pulses through the OBD interface. Tire pressure sensors monitor the pressure (accuracy ±0.05 bar) and temperature of all four tires. A steering angle sensor (resolution ≤1°) and a brake signal sensor record driving status. After the vehicle is started, the sensors perform self-test and calibration, collecting data at a set frequency (e.g., 1kHz for vibration and every 5 minutes for tire pressure). Data synchronization is achieved through a hardware clock or timestamp (error ≤1ms), as follows:

[0039] 1. Sensor selection and layout: To achieve multimodal data collection, different types of sensors need to be reasonably arranged on the vehicle;

[0040] 1.1. Vibration sensor: A three-axis MEMS accelerometer is installed on the inner side of the steering wheel frame and the wheel hub bearing seat. The steering wheel sensor is used to collect steering wheel vibration information, with a range of ±16g and an accuracy of 0.001g. The wheel hub sensor directly measures wheel hub vibration, with a sampling frequency of no less than 1kHz.

[0041] 1.2. Vehicle speed sensor: obtains CAN bus speed data through the vehicle's OBD interface in km / h with an accuracy of ±1km / h; it also records wheel speed pulse signals to calculate the actual wheel speed;

[0042] 1.3. Tire pressure sensor: TPMS tire pressure sensor, installed on the tire valve stem, monitors the tire pressure and temperature of all four tires in real time; tire pressure measurement accuracy is ±0.05Bar, and temperature accuracy is ±2℃;

[0043] 1.4. Steering angle sensor, installed in the steering column electronic module, with a resolution of ≤1°, is used to record the vehicle's steering angle and distinguish the difference in vibration between straight-line driving and turning;

[0044] 1.5. Brake signal sensor, connected to the brake light switch circuit, obtains the brake status analog signal and identifies the vehicle's braking condition;

[0045] 2. Data Collection Process

[0046] 2.1. Initialization: After the vehicle is started, each sensor performs self-test and calibration to ensure the accuracy of data collection;

[0047] 2.2 Real-time Collection

[0048] The vibration sensor collects triaxial acceleration data at a frequency of 1kHz, denoted as a x (t), a y (t), a z (t);

[0049] The vehicle speed sensor updates the vehicle speed v (km / h) and wheel speed n (rev / min) once per second;

[0050] The tire pressure sensor collects tire pressure P every 5 minutes. i (i=1,2,3,4, corresponding to four tires) and temperature T i ;

[0051] The steering angle sensor records the steering angle θ in real time, and the brake signal sensor records the brake status S (0 is released, 1 is braked);

[0052] 2.3 Data synchronization: Use hardware clock or timestamp to align all sensor data to ensure the time error is ≤ 1ms;

[0053] Second, data preprocessing: De-noise the vibration data using a 5-50Hz second-order Butterworth bandpass filter. For example, an original acceleration of 2g is filtered to 1.2g. Time domain features (peak difference, RMS, kurtosis), frequency domain features (main frequency and frequency multiplication analysis after FFT transformation), and associated features (vehicle speed-vibration mapping, tire pressure correction factor, such as corrected RMS value when tire pressure deviates from the standard by ≥0.2bar) are extracted. Multi-dimensional features are used to characterize the wheel vibration state, providing a foundation for subsequent fusion analysis. The details are as follows:

[0054] 1. Denoising: Denoise the raw acceleration data collected by the vibration sensor and use a bandpass filter to filter out interference signals:

[0055] Filter Design: Design a 5-50Hz second-order Butterworth bandpass filter;

[0056] Filtering formula: For the original acceleration data a(t), the filtered data a f (t) can be calculated by the following formula: Among them, w c1 =2π×5(rad / s), w c2 =2π×50(rad / s), where s is the Laplace operator; H(s) is converted into a digital filter for calculation by discretization (such as bilinear transformation method);

[0057] Example: Assume that the value of the original acceleration data a(t) at a certain moment is 2g. After the above bandpass filter calculation, the filtered data a is obtained. f (t) = 1.2 g;

[0058] 2. Feature extraction

[0059] 2.1 Time Domain Features

[0060] Vibration peak value: Calculate the maximum value of the filtered acceleration within a unit time window (such as 1 second) A max and minimum value A min , peak difference ΔA=A max -A min ;

[0061] Root mean square (RMS): Where N is the number of sampling points in 1 second (N = 1000, because the sampling frequency is 1kHz), a f (t i ) is the acceleration value of the i-th sampling point after filtering;

[0062] Kurtosis value: in, for a f The mean of (t);

[0063] 2.2 Frequency Domain Characteristics

[0064] Fast Fourier transform (FFT): Perform FFT on the filtered acceleration data of 1 second to convert the time domain signal into the frequency domain signal X(f);

[0065] Main frequency and sideband analysis: Calculate the frequency corresponding to the wheel speed n is the wheel speed (r / min); f in the frequency domain signal n and its frequency multiples (2f n 、3f n The amplitude of the components (e.g., [A], [B], [C], [D ...E]) is used to determine whether there is an abnormal peak;

[0066] 2.3. Related Features

[0067] Vehicle speed-vibration amplitude relationship: Establish a mapping relationship between vehicle speed v and vibration RMS value; for example, record the RMS value every 10 km / h and analyze its growth trend;

[0068] Tire pressure influencing factor: Calculate the correlation between tire pressure and vibration characteristics; set the average tire pressure like Then the vibration RMS value is corrected:

[0069] Third, multimodal data fusion: First, perform minimum-maximum normalization on vibration, vehicle speed, tire pressure, and other data to unify them into the range [0, 1]. Evidence theory is used to assign basic probabilities to each modality (for example, when vibration RMS is greater than 0.5g, the probability of failure is 0.7). Fusion probabilities are then calculated using the Dempster synthesis rule. A decision threshold is set: after fusion, a failure probability greater than 0.6 is considered a fault, and a normal probability greater than 0.7 is considered normal. This allows for comprehensive confidence assessment of multi-source data and improves detection accuracy. The details are as follows:

[0070] 1. Data normalization

[0071] Unify different types of data into the same numerical range and use the minimum-maximum normalization method: Among them, x is the original data, x min and x max are the minimum and maximum values ​​of this type of data, respectively. norm is the normalized data;

[0072] 3.2 Evidence-theory integration

[0073] Basic Probability Assignment (BPA): Assign a probability of failure to each modal data; for example, assign a BPA for a wheel runout fault based on the comparison of the vibration RMS value with a threshold: if RMS>0.5g, m 振动 (fault) = 0.7, m 振动 (normal) = 0.3; if RMS ≤ 0.5g, m 振动 (fault) = 0.2, m 振动 (normal) = 0.8; similarly, the corresponding BPA is assigned according to the vehicle speed-vibration correlation, tire pressure and other data, and is recorded as m 车速 、m 胎压 wait;

[0074] Evidence fusion: Dempster's synthesis rule is used to calculate the probability after fusion: Where A is the proposition (such as "wheel hub runout failure" or "normal"), m i (A i ) is the i-th modal data pair A i of BPA;

[0075] Decision rule: If m(fault)>0.6 after fusion, it is determined that there is a wheel hub runout fault; if m(normal)>0.7, it is determined to be normal;

[0076] 4. Fault diagnosis and determination, combining threshold determination and pattern recognition: Radial runout must meet the following requirements: RMS vibration > 0.5g and the deviation between the main frequency and the wheel speed frequency ≤ ±10%. Axial runout must meet the following requirements: kurtosis > 3.5 and steering angle change < 5°. A fault pattern library containing typical features is established. Current data is matched using a weighted similarity algorithm (such as Euclidean distance). When the similarity is > 0.8, the corresponding fault type is determined, achieving accurate identification and location of the runout type. The details are as follows:

[0077] 1. Threshold determination

[0078] Vibration threshold

[0079] Radial runout: When the vibration RMS value is greater than 0.5g, and the main frequency f is equal to the wheel speed frequency f n If the deviation is within ±10%, it is determined that there may be a radial runout fault;

[0080] Axial runout: When the kurtosis value is greater than 3.5 and the steering angle θ changes slightly (|Δθ|<5°), an axial runout fault may occur.

[0081] Vehicle speed associated threshold: For every 10 km / h increase in vehicle speed, if the vibration RMS value increases by more than 0.1g, the fault judgment weight is increased;

[0082] 2. Pattern Recognition

[0083] Establish a fault mode library: Based on a large amount of experimental data, establish a vibration characteristic mode library of different degrees of radial runout and axial runout, including the typical value range of time domain and frequency domain characteristics;

[0084] Matching judgment: Match the characteristics of the current collected data with the pattern library and calculate the similarity Sim: Among them, k is the number of features (such as RMS value, kurtosis value, main frequency, etc.), w j is the weight of the jth feature, is the current eigenvalue x j Corresponding eigenvalues ​​in the standard mode similarity (which can be calculated using methods such as Euclidean distance); if Sim>0.8, it is determined to be a corresponding type of beating fault;

[0085] 5. Result Output and Early Warning: The vehicle terminal displays vibration parameters, vehicle speed, tire pressure, and fault results (including confidence levels) in real time. A graded early warning system is set: a fault probability of 0.6-0.8 triggers a yellow warning (prompting inspection), while a probability of ≥0.8 triggers a red warning (forcing the vehicle to stop for repair). A report is generated every 100 kilometers driven or when a fault is detected, including data trends, fault type, and repair recommendations (such as "Check left front wheel radial runout"). The report is stored locally and synchronized to the cloud, allowing the owner to view it remotely. The details are as follows:

[0086] 1. Real-time display: the following information is displayed in real time on the vehicle terminal:

[0087] Steering wheel and wheel hub vibration RMS value and kurtosis value;

[0088] Data such as vehicle speed, tire pressure, steering angle, etc.

[0089] Wheel hub runout fault determination results and confidence level;

[0090] 2. Early warning mechanism

[0091] Mild warning: When the fused fault probability is 0.6≤m(fault)<0.8, the vehicle terminal displays a yellow warning icon and a voice prompts "The wheel hub may have slight vibration, please check as soon as possible";

[0092] Serious warning: When m(fault) ≥ 0.8, the vehicle terminal displays a red warning icon, the buzzer continues to alarm, and a message is pushed to the owner's mobile phone via SMS or APP, including the message "The wheel hub is beating seriously, please stop the car immediately for repairs";

[0093] 5.3 Diagnostic Report Generation

[0094] A wheel hub health diagnostic report is generated every 100 kilometers or when a fault is detected, including:

[0095] Recent data trend charts of vibration, vehicle speed, tire pressure, etc.;

[0096] Fault type, occurrence time and severity;

[0097] Maintenance recommendations, such as "It is recommended to check the radial runout of the left front wheel. The wheel hub may need to be corrected or replaced";

[0098] The report is stored locally on the vehicle device and synchronized to the cloud. The car owner can view it at any time through the mobile phone APP.

[0099] To summarize, let's take a specific example: a family car with 16-inch wheels (standard tire pressure 2.5 bar) experiences steering wheel vibration while driving at 80 km / h (corresponding to a wheel speed of 850 rpm). The detection system collects multi-source data: the RMS raw value of the left front wheel vibration is 0.8 g, the average tire pressure of all four tires is 2.45 bar, and the steering angle is 0°.

[0100] After the data was band-pass filtered at 5-50 Hz, the RMS value was corrected to 0.816 g, with a kurtosis value of 4.2. The FFT showed an abnormal amplitude at 14.2 Hz (the frequency corresponding to wheel speed). For every 20 km / h increase in vehicle speed, the RMS value increased by 0.4 g, exceeding the threshold. The tire pressure correction factor was 1.02.

[0101] Using evidence fusion theory, the vibration mode fault probability was 0.7, the vehicle speed mode was 0.8, and the tire pressure mode was 0.4, resulting in a fusion fault probability of 0.78. Compared with the fault mode library, the feature similarity reached 0.89, indicating a left front wheel radial runout fault.

[0102] The system triggered a yellow warning and generated a report recommending repairs; the repair shop measured the radial runout of the left front wheel to be 0.4mm (exceeding the standard by 0.2mm), and the vehicle returned to normal after correction.

[0103] Example 2:

[0104] Please refer to Figure 2 As shown, this embodiment implements the wheel hub runout detection method based on multimodal data fusion described in Example 1, and provides a detailed scheme of a wheel hub runout detection system based on multimodal data fusion. The system is specifically as follows:

[0105] 1. Perception layer: multi-source data collection

[0106] 1. Vibration sensor: A three-axis MEMS accelerometer is installed on the inner side of the steering wheel frame and the wheel hub bearing seat. The steering wheel sensor is used to collect steering wheel vibration information, with a range of ±16g and an accuracy of 0.001g. The wheel hub sensor directly measures wheel hub vibration, with a sampling frequency of no less than 1kHz.

[0107] 2. Vehicle speed sensor: obtains CAN bus speed data through the vehicle's OBD interface in km / h with an accuracy of ±1km / h; it also records wheel speed pulse signals to calculate the actual wheel speed;

[0108] 3. Tire pressure sensor: TPMS tire pressure sensor, installed on the tire valve stem, monitors the tire pressure and temperature of all four tires in real time; tire pressure measurement accuracy is ±0.05Bar, and temperature accuracy is ±2℃;

[0109] 4. Steering angle sensor, installed in the steering column electronic module, with a resolution of ≤1°, is used to record the vehicle's steering angle and distinguish the difference in vibration between straight-line driving and turning;

[0110] 5. Brake signal sensor, connected to the brake light switch circuit, obtains the brake status analog signal and identifies the vehicle's braking condition;

[0111] 2. Transport layer: data communication design

[0112] In-vehicle network: uses CAN bus protocol (250kbps baud rate) to realize data interaction between sensors and main control unit;

[0113] Remote communication: 4G / 5G modules (such as QuectelRM500Q) enable data upload to the cloud diagnostic platform;

[0114] Local storage: SD card stores original data (supports 256GB, cyclic overwriting storage);

[0115] 3. Processing layer: data processing and analysis

[0116] 1. Data preprocessing module

[0117] 1.1. Filtering algorithm: Second-order Butterworth bandpass filter (5-50Hz) to remove road vibration and engine noise interference

[0118] 1.2 Feature extraction

[0119] Time domain: vibration peak value, RMS value, kurtosis value

[0120] Frequency domain: FFT transform to extract main frequency and sidebands

[0121] Related features: vehicle speed-vibration amplitude curve, tire pressure correction coefficient

[0122] 2. Multimodal fusion module

[0123] 2.1. Normalization: Minimum-maximum normalization method to unify the data into the [0,1] interval

[0124] 2.2. Evidence-theory integration:

[0125] Assign a basic probability assignment (BPA) to each mode; calculate the fusion probability using Dempster's synthesis rule; decision threshold: failure probability > 0.6 triggers an early warning;

[0126] 2. Fault diagnosis module

[0127] 2.1. Threshold determination:

[0128] Radial runout: RMS>0.5g and the deviation between main frequency and wheel speed frequency<±10%

[0129] Axial runout: kurtosis>3.5 and steering angle change<5°

[0130] 2.2. Pattern recognition: Establish a fault feature pattern library and use a weighted similarity algorithm to match fault types;

[0131] 4. Application layer: function implementation

[0132] 1. Real-time display: The vehicle-mounted OLED screen displays vibration parameters and fault status;

[0133] 2. Graded warning:

[0134] Yellow warning: failure probability 0.6-0.8;

[0135] Red warning: failure probability > 0.8;

[0136] 3. Diagnostic report:

[0137] Local storage: Generate inspection reports every 100 kilometers;

[0138] Cloud analysis: Provides historical data query and trend forecasting services.

[0139] Finally, it should be noted that the above embodiments are merely examples for the purpose of illustrating the present invention and are not intended to limit the embodiments. Those skilled in the art will readily appreciate that other variations or modifications based on the above description are possible. It is not necessary and impossible to provide an exhaustive list of all embodiments. However, obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A wheel hub runout detection method based on multimodal data fusion, characterized in that: The following steps are involved: Step 1: Collect wheel hub data in real time through vibration sensors, vehicle speed sensors, tire pressure sensors, and steering angle sensors; Step 2: Synchronize the collected data to ensure that the timestamps of each modal data are aligned and the error is within the specified range; Step 3: Preprocess the data using bandpass filtering and feature extraction to obtain time domain, frequency domain and correlation features; Step 4: Normalize the preprocessed data, use evidence theory to assign basic probability distribution of each modal data, and calculate the fusion fault probability through synthesis rules; Step 5: Determine whether the wheel hub has a runout fault based on the fused fault probability. If the fault probability exceeds the set threshold, further identify the fault type and output the result.

2. The wheel hub runout detection method based on multimodal data fusion according to claim 1, characterized in that: In the data synchronization processing step, a high-precision clock module is used to add a time stamp to each sensor data, and the time stamp alignment is achieved through linear interpolation or synchronization signal calibration.

3. The wheel hub runout detection method based on multimodal data fusion according to claim 1, characterized in that: In the data preprocessing, the bandpass filter is set to 5-50 Hz to remove high-frequency noise and low-frequency interference; the feature extraction includes calculating the root mean square (RMS) and kurtosis of the vibration data, the stability index of the vehicle speed data, and the correlation coefficient between tire pressure and vibration.

4. The wheel hub runout detection method based on multimodal data fusion according to claim 1, characterized in that: In the process of evidence theory fusion, the failure probability model of each modal data is constructed through historical data and expert experience to determine the initial value of the basic probability distribution.

5. A wheel hub runout detection system based on multimodal data fusion, implemented by the method according to any one of claims 1 to 4, characterized in that: The system includes a perception layer, a transmission layer, a processing layer, and an application layer; Perception layer: includes vibration sensors, vehicle speed sensors, tire pressure sensors, and steering angle sensors, used to collect wheel-related data in real time; Transport layer: CAN bus is used to realize vehicle intranet data transmission, 4G / 5G module is used for remote data communication, and SD card is used for local data storage; Processing layer: It is equipped with a data preprocessing module for filtering and feature extraction of collected data; Multimodal fusion module, which performs data normalization and evidence theory fusion; Fault diagnosis module, determines the fault type based on the fusion results; Application layer: including vehicle-mounted terminals, used to display detection results and graded warnings in real time; Cloud platform to store diagnostic reports and perform data analysis.

6. The wheel hub runout detection system based on multimodal data fusion according to claim 5, characterized in that: The sensors in the perception layer have an automatic calibration function and perform self-inspection and calibration when the vehicle is started to ensure the accuracy of data collection.

7. The wheel hub runout detection system based on multimodal data fusion according to claim 5, characterized in that: The multimodal fusion module and fault diagnosis module of the processing layer use hardware acceleration chips or cloud computing platforms for calculation to improve detection efficiency, achieve processing of more than 1,000 groups of data points per second, and a single detection takes less than 200ms.

8. The wheel hub runout detection system based on multimodal data fusion according to claim 5, characterized in that: The vehicle-mounted terminal of the application layer and the cloud platform realize two-way data interaction. The vehicle-mounted terminal can receive the optimization algorithm and update information of the cloud platform, and the cloud platform can obtain the detection data of the vehicle-mounted terminal in real time for continuous optimization of the diagnosis model.

Images