Brake leaf spring crack data analysis method and system

Abstract

The invention provides a brake leaf spring crack data analysis method and system, and relates to the field of vehicle fault diagnosis and data analysis, and the method comprises the steps: receiving a maintenance report of a third-party maintenance station, vehicle-mounted vibration sensor data and user feedback data; when the contradiction among the three is recognized and a vehicle engine flames out, a standardized brake instruction is issued to drive a brake caliper to clamp, so that a brake plate assembly containing a brake plate spring is excited to generate vibration response; synchronously triggering an accelerometer to collect a target vibration response signal at a high sampling rate; performing advanced feature extraction on the signal to obtain a vehicle feature vector; and comparing the feature vector with a preset standard sample library feature set by adopting a multi-dimensional feature vector similarity comparison algorithm, and determining the real crack state of the brake leaf spring according to a comparison result. According to the method, standardized excitation is triggered through multi-source data contradiction, high-fidelity vibration signal feature analysis is combined, the problem of brake leaf spring crack misjudgment under a traditional single data source is solved, and the detection accuracy under complex working conditions is improved.

Description

Technical Field

[0001] The present application relates to the field of vehicle fault diagnosis and data analysis, and in particular, to a brake leaf spring crack data analysis method and system. Background Art

[0002] As a critical safety component of a vehicle's braking system, the integrity of brake leaf springs is directly linked to driving safety. Currently, shared vehicle maintenance services typically rely on numerous third-party repair shops. These shops vary significantly in size, technical expertise, and equipment configuration. For example, large chain repair centers may be equipped with advanced diagnostic equipment capable of providing highly accurate, structured data. For example, detection systems based on ultrasonic arrays or eddy current imaging can accurately measure the length, width, and depth of cracks. Smaller independent repair shops, however, may be constrained by cost and generally still rely on traditional magnetic particle inspection equipment. While magnetic particle inspection technology is relatively inexpensive, its ability to detect small cracks is limited, and test results are often qualitative, such as "presence or absence of crack indication." This disparity in equipment and technical expertise leads to significant inconsistencies in the accuracy, format, and information dimensions of brake leaf spring crack detection data uploaded to the shared vehicle platform's central database by different repair shops.

[0003] Furthermore, human error exacerbates data inaccuracies. For example, when faced with stringent vehicle turnover targets at repair shops, technicians may, to save time, skip critical equipment calibration steps, such as the mandatory daily calibration of magnetic particle inspection equipment. This oversight significantly reduces the sensitivity of the inspection equipment, rendering microscopic metal fatigue cracks imperceptible to the naked eye invisible. When these potentially hazardous brake leaf springs are mistakenly marked as "passed" and uploaded to the platform database, erroneous "normal" data is introduced.

[0004] These erroneous maintenance data had a serious impact on the shared car platform's crack condition assessment system. This system was designed to predict the health of brake leaf springs by analyzing maintenance data and onboard sensor data. However, when the system received real crack samples that were incorrectly labeled "normal," its internal assessment logic and parameters were inaccurately adjusted. This led to the system misjudging vehicles with similar characteristics in subsequent risk assessments, mistakenly deeming their brake systems to be less risky.

[0005] After the vehicle returned to service, the microscopic cracks in the brake leaf spring continued to expand under frequent braking and complex road conditions, causing changes in its structural integrity and vibration patterns. The onboard vibration sensors began to detect abnormal vibration waveforms with specific frequency characteristics. However, because the platform's backend analysis system had been affected by the erroneous information, its evaluation logic was unable to accurately identify the actual crack information represented by these abnormal vibration waveforms. The system may have mistakenly identified these real crack warning signals as "transient impacts caused by speed bumps" or "normal vibrations caused by road bumps," marking them as benign events that did not require human intervention, thereby ignoring potential safety hazards.

[0006] Furthermore, user feedback, a crucial source of information, presents new challenges due to its unstructured and subjective nature. When users perceive fault signs such as "soft brakes" or "strange noises" and submit feedback through the platform, these natural language descriptions differ significantly from structured sensor data and maintenance records. The platform's vehicle dispatch system is faced with conflicting information from various sources: structured maintenance data from the repair station indicates "normal," while onboard vibration sensor data is misclassified as "normal" by the system, while user feedback clearly indicates a fault. Lacking a mechanism to effectively integrate, calibrate, and interpret these heterogeneous and conflicting information sources, the system is unable to issue recall orders based on any single source or simple logical rules. Consequently, vehicles with safety hazards are forced to continue operating online, without timely and effective intervention. Summary of the Invention

[0007] The present application provides a brake leaf spring crack data analysis method and system, which effectively solves the problem of misjudgment of brake leaf spring cracks caused by inconsistent and inaccurate data in the existing technology through active excitation, high-precision data acquisition and multi-source data fusion analysis, thereby having the advantage of improving diagnostic accuracy and reliability.

[0008] In one aspect, the present application provides a brake leaf spring crack data analysis method, comprising: Receive maintenance reports from third-party repair stations, vehicle vibration sensor data, and user feedback data; Upon identifying a discrepancy between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data, and when the vehicle engine is turned off, issuing a standardized braking command to the braking system of the target vehicle to drive the brake caliper to perform a clamping action, causing the brake leaf spring to receive a standardized excitation to generate a vibration response; wherein the brake caliper clamps a brake pad assembly, and the brake pad assembly includes a brake leaf spring and a brake friction pad; When the standardized braking instruction is issued, a synchronous trigger signal is sent to the accelerometer, and the accelerometer collects the target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate; Performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information; A multi-dimensional feature vector similarity comparison algorithm is used to compare the vehicle feature vector information with a feature set in a preset standard sample library to obtain a comparison result, and the actual crack state of the brake leaf spring is determined based on the comparison result.

[0009] Optionally, before the step of issuing a standardized braking instruction to the braking system of the target vehicle to drive the brake caliper to perform the clamping action, the method further includes: A standardized physical surface pre-processing instruction is issued to the brake system to drive the brake caliper to perform multiple clamping and releasing operations on the brake leaf spring according to a set pressure and duration; wherein the clamping and releasing operations cause the brake friction pad in the brake pad assembly to rub against the brake disc to remove an unstable oxide film on the friction surface.

[0010] Optionally, the step of performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information includes: Performing three-level wavelet packet decomposition processing on the target vibration response signal to obtain a sub-band signal; Extracting time domain characteristic parameters and frequency domain characteristic parameters of each sub-band signal, wherein: the time domain characteristic parameters include: root mean square value, kurtosis coefficient and impulse factor; the frequency domain characteristic parameters include: main frequency band energy proportion and frequency band energy entropy; The time domain feature parameters and the frequency domain feature parameters are combined into a multi-dimensional feature vector as the vehicle feature vector information.

[0011] Optionally, when the standardized braking instruction is issued, a synchronous trigger signal is sent to an accelerometer, and the accelerometer collects the target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate, including the following steps: Repeating the application of the standardized stimulus a preset number of times; Each time the standardized excitation is applied, a vibration response signal generated by the brake leaf spring under the standardized excitation is synchronously collected to obtain a plurality of vibration response signals; Multiple vibration response signals are time-aligned and superimposed to suppress noise to obtain the target vibration response signal.

[0012] Optionally, the feature sets in the preset standard sample library include a standard healthy brake leaf spring feature set and a standard cracked brake leaf spring feature set; the steps of comparing the vehicle feature vector information with the feature sets in the preset standard sample library using a multi-dimensional feature vector similarity comparison algorithm to obtain a comparison result, and determining the true crack state of the brake leaf spring according to the comparison result specifically include: Calculate the average distance DH between the vehicle feature vector information and all vectors in the standard healthy brake leaf spring feature set, and the average distance DC between the vehicle feature vector information and all vectors in the standard cracked brake leaf spring feature set; If DC < DH and DC is less than a preset threshold, it is determined that the brake leaf spring is in a cracked state; If DH < DC and DH is less than a preset threshold, it is determined that the brake leaf spring is in a healthy state.

[0013] Optionally, after the step of receiving the maintenance report from a third-party repair station, the on-vehicle vibration sensor data, and the user feedback data, it includes: Extract the status of the maintenance report, the on-vehicle vibration sensor data, and the user feedback data to obtain status indicators corresponding to each data source; Assign reliability weights to the status indicators according to the types and historical performances of the data sources;<00000​​​​​​​​​​​​​​​​​ Extracting time-domain or frequency-domain features from the vehicle-mounted vibration sensor data, and comparing the extracted features with a preset feature pattern to obtain a status indication of the brake leaf spring; Furthermore, natural language processing is performed on the user feedback data to identify keywords or phrases related to the state of the brake leaf spring, and the identification results are mapped into a state indication of the brake leaf spring.

[0016] Optionally, the step of calculating the propensity score of each data source for the brake leaf spring state includes: The status indications of the data sources are converted into numerical values, and weighted summation or weighted averaging of the numerical values ​​is performed according to the reliability weights to obtain the propensity score.

[0017] On the other hand, the present application provides a brake leaf spring crack data analysis system, the system comprising: Data receiving module, used to receive maintenance reports from third-party repair stations, vehicle vibration sensor data and user feedback data; an excitation control module, configured to, upon identifying a discrepancy between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data, and when the vehicle's engine is turned off, issue a standardized braking command to the target vehicle's braking system to drive the brake caliper to perform a clamping action, causing the brake leaf spring to receive a standardized excitation to generate a vibration response; wherein the brake caliper clamps a brake pad assembly, which includes a brake leaf spring and a brake friction pad; a vibration response acquisition module, configured to send a synchronous trigger signal to an accelerometer when the standardized braking instruction is issued, so that the accelerometer acquires a target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate; Feature extraction module: performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information; The crack diagnosis module uses a multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with the feature sets in the preset crack sample library and the healthy sample library. When the similarity with the crack sample library exceeds a first threshold, a crack diagnosis conclusion is output; when the similarity with the healthy sample library exceeds a second threshold, a healthy diagnosis conclusion is output.

[0018] The present application provides a brake leaf spring crack data analysis method and system, which effectively solves the problem of misjudgment of brake leaf spring cracks caused by inconsistent and inaccurate data in the existing technology through active excitation, high-precision data acquisition and multi-source data fusion analysis, thereby having the advantage of improving diagnostic accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to explain the present application more clearly, the following briefly introduces the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 hereinafter is a flow chart showing a method for analyzing crack data of a brake leaf spring according to an embodiment of the present invention; Figure 2 Schematic diagram of a module configuration of a brake leaf spring crack data analysis system in an embodiment is shown in FIG.

[0021] Reference numerals: 100, brake leaf spring crack data analysis system; 10, data receiving module; 20, excitation control module; 30, vibration response acquisition module; 40, feature extraction module; 50, crack diagnosis module. DETAILED DESCRIPTION

[0022] The technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments. The components of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0023] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0024] Traditional, existing shared vehicle brake leaf spring condition assessment systems, when integrating maintenance reports from third-party repair stations, on-board vibration sensor data, and user feedback, suffer from heterogeneous data sources, conflicting information, and a lack of a single reliable source. Specifically, differences in testing equipment and operating procedures across repair stations lead to inconsistent maintenance data accuracy and format. For example, small repair stations using magnetic particle inspection equipment may be unable to detect microscopic cracks, while large repair stations using high-precision image scanning equipment can provide detailed structured data. At the same time, on-board vibration sensor data can be misjudged due to the evaluation model being contaminated by erroneous maintenance data, misidentifying real crack signals as benign events. Furthermore, user feedback data is typically unstructured text descriptions, making it difficult to standardize and quantify. These issues prevent the system from accurately identifying and confirming the true crack condition of the brake leaf spring, which in turn affects decisions regarding safe vehicle operations.

[0025] If these issues are not addressed, ride-sharing platforms will continue to face the risk of misjudging the status of brake leaf spring cracks. Such misjudgments could result in vehicles with safety hazards continuing to operate online, posing a threat to user safety. Furthermore, failure to promptly detect and repair cracks could exacerbate damage to the brake leaf spring, ultimately leading to component failure and more serious traffic accidents. Furthermore, erroneous assessment results will continue to contaminate the platform's risk assessment model, causing cascading misjudgments in subsequent vehicle status assessments and reducing the reliability and accuracy of the entire system. This long-term trend will not only damage the platform's user trust and brand reputation, but also increase operating costs, such as accident compensation, vehicle repair costs, and potential legal liabilities.

[0026] like Figure 1 The figure shows a schematic diagram of a brake leaf spring crack data analysis method. By integrating multiple data sources and applying standardized excitation to obtain reliable vibration responses, the crack state can be analyzed. The present application proposes a brake leaf spring crack data analysis method, which includes: S10 receives maintenance reports from third-party repair stations, vehicle vibration sensor data, and user feedback data.

[0027] S20, when a contradiction is identified among the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data, and when the engine of the vehicle is turned off, a standardized braking instruction is issued to the braking system of the target vehicle to drive the brake caliper to perform a clamping action, so that the brake leaf spring receives standardized excitation to generate a vibration response; wherein the brake caliper clamps the brake pad assembly, and the brake pad assembly includes a brake leaf spring and a brake friction pad.

[0028] Among them, standardized braking instructions refer to preset, repeatable braking operation instructions, which can be implemented by the vehicle control unit (ECU) issuing specific brake pressure instructions through bus communication, or by simulating brake pedal signals through external test equipment. It can ensure that the brake leaf spring receives consistent excitation under controlled conditions, thereby eliminating the impact of environmental factors or operational differences on vibration response data.

[0029] Driven by standardized braking commands, the brake leaf spring is subjected to a predetermined pattern of force or displacement, thereby generating predictable vibrations. This can be achieved by precisely clamping and releasing the brake pad assembly with a brake caliper, simulating the stress state of the brake leaf spring under actual braking conditions and inducing it to produce a specific vibration mode when cracks are present.

[0030] The brake pad assembly includes the brake spring and brake friction pad. It refers to the component in the braking system that directly contacts the brake disc and generates friction. It is used to clarify the physical object of standardized excitation.

[0031] S30, while the standardized braking instruction is being issued, sending a synchronous trigger signal to the accelerometer, and the accelerometer collects and obtains the target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate.

[0032] The synchronous trigger signal refers to a time synchronization signal sent to the accelerometer at the same time as the standardized braking command is issued. It can be implemented by connecting a hardware trigger line or through software timestamp synchronization to ensure that the accelerometer starts data acquisition at the precise moment of excitation application and ensures that the vibration response signal is aligned with the time of the excitation action.

[0033] S40: performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information.

[0034] The original vibration response signal is subjected to complex mathematical transformation and processing to achieve advanced feature extraction, so as to extract characteristic parameters with lower dimension but richer information that can characterize the state of the brake leaf spring. It can adopt various signal processing technologies such as time domain analysis, frequency domain analysis, time-frequency analysis, etc. to realize the conversion of raw, huge vibration data into structured information that can be used for pattern recognition and comparison.

[0035] Vehicle feature vector information refers to a multidimensional numerical set obtained after advanced feature extraction that can represent the vibration characteristics of the brake leaf spring of the current vehicle, and serves as the input for subsequent similarity comparison.

[0036] S50, using a multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with a feature set in a preset standard sample library to obtain a comparison result, and determining the actual crack state of the brake leaf spring based on the comparison result.

[0037] The multidimensional feature vector similarity comparison algorithm refers to a calculation method used to quantify the similarity between two or more multidimensional feature vectors. It can be implemented using a variety of distance or similarity measurement methods such as Euclidean distance, cosine similarity, and Mahalanobis distance. It can objectively evaluate the degree of similarity between the characteristics of the brake leaf spring to be tested and the known healthy or cracked state characteristics.

[0038] The core innovation of this application lies in that, when inconsistencies in multi-source data are identified, a controlled standardized physical excitation test is introduced, combined with high-precision vibration response acquisition and advanced feature comparison, thereby overcoming the unreliability and inconsistency of traditional data sources, achieving accurate judgment of the actual crack state of the brake leaf spring, and achieving the effect of improving diagnostic reliability and avoiding the continued operation of vehicles with safety hazards.

[0039] This application builds a multi-dimensional data input foundation by receiving maintenance reports, vehicle-mounted vibration sensor data, and user feedback data from third-party repair stations. When the system identifies a contradiction between these heterogeneous data sources, such as a maintenance report showing normal operation but an abnormality in the vehicle data or user feedback, and the vehicle is in a quiet state with the engine turned off, the system will proactively issue a standardized braking command to the target vehicle's braking system. This command drives the brake caliper to perform a precise clamping action, so that the brake leaf spring receives a standardized excitation. This controlled excitation method eliminates the impact of uncertain factors such as vehicle operating status, road conditions, or human operation on the vibration response, ensuring the consistency of conditions for each test. At the same time as the standardized braking command is issued, the system sends a synchronous trigger signal to the accelerometer, ensuring that the accelerometer begins to collect the target vibration response signal generated by the brake leaf spring at a preset high sampling rate at the precise moment the excitation is applied. The high sampling rate ensures the ability to capture vibration details and provides high-quality raw data for subsequent analysis. Subsequently, the collected target vibration response signal is subjected to advanced feature extraction and converted into vehicle feature vector information. This process compresses the complex raw signal into a more representative set of values, reducing the data dimensionality while preserving key vibration characteristics. Finally, a multi-dimensional feature vector similarity comparison algorithm is used to compare the extracted vehicle feature vector information with a set of features in a pre-set standard sample library. This standard sample library contains the features of brake leaf springs with known healthy and known crack states. By comparing the results, the system can objectively determine the actual crack state of the current brake leaf spring, thus overcoming the limitations of a single data source and the interference of conflicting information.

[0040] For example, data reception can be achieved through the vehicle's remote diagnostic interface or cloud platform API. Maintenance reports, onboard vibration sensor data, and user feedback are aggregated into a central processing unit (CPU). The CPU runs a data consistency check module, which uses pre-set logic rules or machine learning models to evaluate whether there are significant discrepancies in the brake leaf spring status indications from various data sources. Once a discrepancy is identified, and the vehicle's engine status sensor confirms that the engine is off, the CPU sends a standardized braking command to the brake control unit via the vehicle communication network. This command can be configured to cause the brake caliper to clamp once or multiple times at a specific pressure and duration. Simultaneously, a hardware trigger or software synchronization mechanism sends a synchronization signal to an accelerometer mounted near the brake caliper or brake pad assembly. This accelerometer can be a microelectromechanical system (MEMS) sensor that continuously collects vibration data at a sampling rate of tens of thousands of times per second. The collected raw vibration signal is transmitted to a signal processing module, which performs advanced feature extraction, such as extracting the energy of the main frequency band through frequency domain analysis or calculating the root mean square value through time domain analysis. These parameters are then combined into a multi-dimensional feature vector. This feature vector is then fed into a pattern recognition module, which stores pre-trained feature sets for standard healthy and cracked samples. Using a similarity comparison algorithm, the module calculates the distance or similarity between the feature vector of the vehicle under test and each feature set in the standard sample library. Based on the comparison results, it outputs a diagnosis of the brake leaf spring's health or crack condition.

[0041] Through the above-mentioned technical solution, this application can effectively integrate and calibrate heterogeneous and conflicting information from different sources. By introducing controlled, standardized physical excitation, the inherent flaws and uncertainties of traditional data sources are overcome, ensuring the reliability and consistency of vibration response data. Furthermore, by performing advanced feature extraction on the high-precision vibration response signal and comparing its similarity with a standard sample library, an accurate judgment of the actual crack state of the brake leaf spring is achieved. This significantly improves the reliability and accuracy of brake leaf spring crack diagnosis, preventing the continued operation of vehicles with safety hazards, thereby ensuring vehicle operation safety.

[0042] In some embodiments, before the step of issuing a standardized braking instruction to the braking system of the target vehicle to drive the brake caliper to perform a clamping action, the method further includes: A standardized physical surface pre-processing instruction is issued to the brake system to drive the brake caliper to perform multiple clamping and releasing operations on the brake leaf spring according to a set pressure and duration; wherein the clamping and releasing operations cause the brake friction pad in the brake pad assembly to rub against the brake disc to remove an unstable oxide film on the friction surface.

[0043] Among them, the brake friction surface is cleaned and prepared before the formal brake actuation, which can be achieved by using pre-programmed electronic control unit (ECU) instructions or specific commands issued through the on-board diagnostic system (OBD) interface to ensure the stability and repeatability of subsequent braking operations.

[0044] This application adds a physical surface pretreatment step before issuing a standardized braking instruction to the braking system of the target vehicle to drive the brake caliper to perform the clamping action. Specifically, before performing the formal standardized braking excitation, the system will issue a standardized physical surface pretreatment instruction to the braking system to drive the brake caliper to perform multiple clamping and releasing operations on the brake leaf spring according to the pre-set pressure and duration. This series of clamping and releasing operations causes repeated friction between the brake friction pad and the brake disc in the brake pad assembly. This friction can effectively remove unstable oxide films or other contaminants that may exist on the friction surface. The presence of these unstable substances will make the clamping force and friction coefficient of the brake caliper unstable when performing the clamping action, thereby causing deviations in the vibration response signal generated by the brake leaf spring when subjected to standardized excitation, affecting the accuracy of subsequent crack state analysis.

[0045] By pre-emptively eliminating these unstable factors, the contact between the brake pad and disc is ensured to be clean and stable. Consequently, when standardized braking commands are subsequently issued, the brake caliper's clamping action becomes more stable and consistent, and the excitation of the brake leaf spring is more standardized, resulting in a purer and more reliable vibration response signal. This pre-processing mechanism, combined with the subsequent standardized excitation acquisition and data analysis steps, creates a more robust brake leaf spring crack diagnosis process.

[0046] By eliminating the uncertainty of the initial friction surface, the signal-to-noise ratio and repeatability of the collected vibration response signal can be significantly improved. This in turn improves the accuracy of advanced feature extraction of the target vibration response signal and the use of a multi-dimensional feature vector similarity comparison algorithm to determine the true crack state of the brake leaf spring, thereby avoiding misjudgments caused by unstable friction surfaces and ensuring the reliability of the diagnostic results.

[0047] For example, before issuing standardized braking commands to the target vehicle's braking system, physical surface pretreatment can be performed. Specifically, the system can send a pretreatment command to the vehicle's electronic brake control unit (EBCU). This command includes a preset brake pressure value, such as 500 kPa, the duration of each clamping operation, such as 2 seconds, and the number of repetitions, such as 5. Upon receiving the command, the EBCU controls the brake hydraulic system, actuating the brake caliper to clamp the brake pad assembly at 500 kPa, hold for 2 seconds, and then release, repeating this process 5 times. During each clamping process, the brake pad rubs against the brake disc. This friction effectively removes or peels away trace amounts of rust, oil, or old oxide layers adhering to the friction surface. For example, if a vehicle is parked for an extended period, a thin layer of rust may form on the brake disc surface. Repeated, gentle braking can remove this rust, restoring the friction surface to a relatively clean and stable state. After completing the pretreatment, the system issues the official standardized braking command for subsequent vibration response signal acquisition and crack analysis.

[0048] The above technical solution effectively removes unstable oxide films and other contaminants from the friction surfaces between the brake pad and the disc through standardized physical surface pretreatment before analyzing brake leaf spring crack data. This ensures more stable and consistent clamping of the brake caliper, and produces a purer and more repeatable vibration response signal when subjected to standardized excitation. This eliminates interference from friction surface instability on the vibration response signal, improving the standardization of the braking process and ultimately enhancing the accuracy of brake leaf spring crack status analysis, thereby avoiding misjudgments.

[0049] In some embodiments, the step of performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information includes: Performing three-level wavelet packet decomposition processing on the target vibration response signal to obtain a sub-band signal; Extracting time domain characteristic parameters and frequency domain characteristic parameters of each sub-band signal, wherein: the time domain characteristic parameters include: root mean square value, kurtosis coefficient and impulse factor; the frequency domain characteristic parameters include: main frequency band energy proportion and frequency band energy entropy; The time domain characteristic parameters and the frequency domain characteristic parameters are combined into a multi-dimensional characteristic vector as the vehicle characteristic vector information.

[0050] The original signal is decomposed at different frequency scales to produce a series of sub-signals with different frequency ranges, achieving a three-level wavelet packet decomposition process. The "three-level" here refers to the number of decomposition levels. The higher the number of levels, the more detailed the signal decomposition, allowing for more precise capture of the signal's local features. This effectively separates the complex information in the original vibration response signal, facilitating subsequent feature extraction for specific frequency ranges while also helping to reduce noise interference with feature extraction.

[0051] The sub-band signal is a component signal representing the original signal within a specific frequency range obtained through wavelet packet decomposition. Each sub-band signal contains the energy and waveform information of the original signal in the corresponding frequency interval, and is used to distribute the energy and information of the original signal to different frequency channels, so that the characteristics of different frequency components can be analyzed in a targeted manner, thereby more effectively identifying weak signals related to brake leaf spring cracks.

[0052] Time-domain characteristic parameters are numerical values ​​extracted from a signal's timeline representation, used to quantify its transient characteristics, energy intensity, or waveform shape. Specifically, the root mean square (RMS) value measures the signal's effective energy or intensity, reflecting its overall vibration level. The kurtosis coefficient describes the degree to which the signal waveform deviates from a normal distribution. It is sensitive to shock or sudden changes and can indicate the presence of abnormal shock components in the signal. The pulse factor reflects the relative intensity of the shock component in the signal and is useful for identifying periodic shocks or transient fault signals. The purpose of extracting these parameters is to capture the energy changes, shock characteristics, and waveform sharpness of the brake leaf spring's vibration response signal over time. These characteristics are often closely related to the initiation and propagation of cracks.

[0053] Frequency domain characteristic parameters refer to the numerical values ​​extracted from the signal's performance on the frequency axis, which are used to quantify the signal's frequency distribution, energy concentration or spectral complexity. Specifically, the proportion of main frequency band energy measures the concentration of signal energy in the main frequency range and can reflect the main vibration mode of the signal.

[0054] Band energy entropy describes the uniformity or complexity of a signal's energy distribution across different frequency components. The more uneven the energy distribution, the lower the entropy value, and vice versa. These parameters are extracted to analyze the energy distribution and complexity of the brake leaf spring's vibration response signal from a frequency perspective. The presence of cracks alters the component's natural frequency and vibration modes, resulting in specific changes in the frequency domain.

[0055] A multidimensional feature vector is a numerical sequence formed by combining multiple different types of feature parameters. It is used to integrate signal features extracted from different dimensions to form a comprehensive and discriminative mathematical representation to facilitate subsequent pattern recognition and classification algorithms, thereby more accurately characterizing the true state of the brake leaf spring.

[0056] This application effectively solves the problem of inaccurate feature extraction caused by noise and redundant information in the original signal by performing multi-level and multi-dimensional processing on the target vibration response signal.

[0057] Specifically, the target vibration response signal is first subjected to a three-level wavelet packet decomposition process, thereby breaking the original signal into multiple sub-band signals. This decomposition allows for detailed analysis of the signal at different frequency scales, effectively isolating key information related to the brake leaf spring crack. It also suppresses any noise and interference components that may be present in the signal, providing a cleaner and more focused data foundation for subsequent feature extraction.

[0058] Based on this, time-domain and frequency-domain feature parameters are extracted from these sub-band signals. Time-domain feature parameters capture the signal's energy, impact characteristics, and waveform morphology in the temporal dimension. These characteristics are crucial for identifying transient or nonlinear vibration modes caused by cracks. Meanwhile, frequency-domain feature parameters reveal the signal's energy distribution and complexity in the frequency dimension, as cracks in brake leaf springs alter their natural frequencies and vibration response spectrum. By simultaneously extracting these two types of features, a more comprehensive and in-depth understanding of the vibration signal can be achieved, thereby enhancing the discriminative power of the feature vector.

[0059] Finally, these time-domain and frequency-domain feature parameters are combined into a multidimensional feature vector, which serves as the vehicle feature vector information. This combination fully utilizes the complementarity between different types of features, making the final feature vector more robust and accurate in representing the true vibration characteristics of the brake leaf spring.

[0060] This advanced feature extraction method makes the vehicle feature vector information obtained from the target vibration response signal more representative and discriminative. When these optimized vehicle feature vectors are compared with the feature set in a pre-set standard sample library, the accuracy of the comparison results is significantly improved, thereby more accurately determining the true crack condition of the brake leaf spring. This method avoids the inaccuracies that may arise from direct analysis of the raw signal, providing reliable data support for brake leaf spring crack diagnosis, effectively solving the problem of inaccurate feature extraction that affects the accuracy of crack status judgment.

[0061] Exemplarily, high-level feature extraction of a target vibration response signal can be implemented as follows: First, the target vibration response signal can be input into a signal processing unit, which can be an embedded processor or an onboard diagnostic computer. Within this signal processing unit, a wavelet packet transform algorithm can be used to perform a three-level decomposition of the signal. For example, the Daubechies series of wavelet basis functions or the Symlets series of wavelet basis functions can be used to perform a three-level decomposition of the original vibration signal, thereby generating eight different sub-band signals, each corresponding to a specific frequency range. Subsequently, for each generated sub-band signal, its time-domain and frequency-domain feature parameters are calculated. The time-domain feature parameters can be calculated by calculating the root mean square (RMS) value, which can be obtained by averaging the square root of the signal; the kurtosis coefficient, which can be calculated by calculating the ratio of the signal's fourth-order central moment to the square of its second-order central moment; and the impulse factor, which can be calculated by calculating the ratio of the signal's peak value to its RMS value. The calculation of frequency domain characteristic parameters can include: the main frequency band energy proportion, which can be obtained by identifying the frequency band with the most concentrated energy in the signal spectrum and calculating the proportion of the energy of this frequency band to the total energy; and the frequency band energy entropy, which can be obtained by calculating the Shannon entropy of the probability distribution of the energy of each sub-band to quantify the uniformity of the energy distribution. Finally, all time domain characteristic parameters and frequency domain characteristic parameters extracted from all sub-band signals are combined. Specifically, these calculated values ​​can be spliced ​​in a predetermined order to form a single, high-dimensional numerical sequence, namely a multidimensional feature vector. For example, the root mean square value, kurtosis coefficient, impulse factor, main frequency band energy proportion and frequency band energy entropy of all sub-bands can be arranged in sequence to form a comprehensive feature vector. This multidimensional feature vector can then be used as vehicle feature vector information for subsequent crack status diagnosis.

[0062] The above-mentioned technical solution can effectively overcome the problem of inaccurate feature extraction caused by the large amount of noise and redundant information contained in the signal when directly using the original vibration response signal for analysis. Through three-level wavelet packet decomposition processing, the vibration response signal is effectively denoised and decomposed into multiple sub-band signals with specific frequency ranges, so that the weak features related to the brake leaf spring cracks can be highlighted. Furthermore, by comprehensively extracting the time domain characteristic parameters and frequency domain characteristic parameters of each sub-band signal, key information such as the signal's energy, impact characteristics, waveform shape, and frequency distribution can be captured from different dimensions, thereby forming a more comprehensive and more discriminative feature set. Ultimately, combining these multi-dimensional feature parameters into vehicle feature vector information can significantly improve the representativeness and robustness of the feature vector, providing a more reliable and accurate basis for subsequent crack status judgment, thereby improving the diagnostic accuracy of the brake leaf spring crack status.

[0063] In some embodiments, when the standardized braking instruction is issued, a synchronous trigger signal is sent to an accelerometer, and the accelerometer collects the target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate, including the following steps: Repeating the application of the standardized stimulus a preset number of times; Each time the standardized excitation is applied, a vibration response signal generated by the brake leaf spring under the standardized excitation is synchronously collected to obtain a plurality of vibration response signals; Multiple vibration response signals are time-aligned and superimposed to suppress noise to obtain the target vibration response signal.

[0064] Among them, multiple vibration response signals collected at different time points are accurately synchronized or aligned on the time axis through a certain algorithm or technical means to achieve time alignment. Specifically, this can be achieved by using a cross-correlation algorithm, peak detection or a method based on specific event marking to ensure the phase consistency of the effective signal during subsequent superposition processing, thereby avoiding signal cancellation.

[0065] The technology of superposition noise suppression is realized by arithmetically superposing or averaging multiple time-aligned vibration response signals to enhance the effective components in the signal and weaken the random noise. Specifically, it can be achieved by using methods such as coherent superposition or integrated averaging to improve the signal-to-noise ratio of the signal, thereby obtaining a clearer and more reliable target vibration response signal.

[0066] The present application can obtain multiple vibration response signals by repeatedly applying standardized excitation a preset number of times. Compared with a single excitation, multiple excitations can accumulate more effective signal energy and improve the signal-to-noise ratio. Each time the standardized excitation is applied, the vibration response signal generated by the brake leaf spring under the standardized excitation is synchronously collected to obtain multiple vibration response signals. Synchronous collection ensures that each collected signal accurately corresponds to the standardized excitation, providing a basis for subsequent time alignment and superposition processing. The multiple vibration response signals are time-aligned and superposition noise suppression processing is performed to obtain the target vibration response signal. Time alignment ensures the correspondence between the various vibration response signals on the time axis, so that the superposition processing can effectively enhance the common components in the signal while suppressing random noise. The superposition noise suppression processing utilizes the coherence of the signal, that is, the effective signal has a similar waveform and phase in multiple measurements, while the noise is random. Through superposition, the amplitude of the effective signal is enhanced, while the amplitude of the noise is reduced due to random cancellation, thereby improving the signal-to-noise ratio of the target vibration response signal.

[0067] The higher-quality target vibration response signal obtained through this processing provides a purer data source for subsequent advanced feature extraction. When the target vibration response signal with a higher signal-to-noise ratio is subjected to three-level wavelet packet decomposition and the time-domain and frequency-domain characteristic parameters of each sub-band signal are extracted, these characteristic parameters more accurately reflect the actual vibration characteristics of the brake leaf spring, reducing noise interference in feature calculation. Furthermore, these more accurate time-domain and frequency-domain characteristic parameters are combined into a multidimensional feature vector, enabling the vehicle feature vector information to more accurately characterize the crack condition of the brake leaf spring.

[0068] Ultimately, when the vehicle feature vector information is compared with the feature set in the preset standard sample library using a multi-dimensional feature vector similarity comparison algorithm, the accuracy and reliability of the comparison results will be significantly improved, thereby being able to more accurately determine the true crack state of the brake leaf spring, effectively avoiding misjudgment or missed judgment due to signal quality problems, and improving the robustness and diagnostic accuracy of the entire crack data analysis method.

[0069] Exemplarily, the preset number of repeated applications of standardized excitation can be set to 10 to 20 times to control the detection time while ensuring the signal enhancement effect. Each time the standardized excitation is applied, the vibration response signal generated by the brake leaf spring under the standardized excitation is synchronously collected. The brake system control unit can send a synchronous pulse signal to the accelerometer while issuing a braking instruction. The pulse signal serves as the starting trigger point for accelerometer data collection to ensure that each collected vibration response signal is accurately aligned with the starting moment of the excitation. When time-aligning multiple vibration response signals, a method based on a cross-correlation function can be used. Specifically, the first collected vibration response signal is selected as the reference signal, and then the cross-correlation function between each of the remaining vibration response signals and the reference signal is calculated to find the peak position of the cross-correlation function. The peak position is the time delay. The signal is shift-compensated according to this delay, thereby achieving time alignment of all signals. When performing superposition noise suppression processing, a coherent averaging method can be used. The multiple time-aligned vibration response signals are arithmetic averaged point by point on the time axis. That is, for each time point, the amplitudes of all signals at that time are added together and then divided by the number of signals. In this way, random noise components will cancel each other out due to their randomness during multiple superpositions, while the effective vibration response signals generated by the standardized excitation will be cumulatively enhanced due to their coherence, thus obtaining a target vibration response signal with a significantly improved signal-to-noise ratio.

[0070] Through the above technical solution, the problem that the vibration response signal generated by a single excitation is weak and vulnerable to noise interference, resulting in low signal quality, which in turn affects the accuracy of subsequent feature extraction and crack state judgment, is solved. By repeatedly applying standardized excitation and synchronously collecting multiple vibration response signals, and then performing time alignment and superposition to suppress noise, the effective signal components can be effectively enhanced, and the noise interference can be significantly reduced, so as to obtain the target vibration response signal with a higher signal-to-noise ratio. This enables more accurate and reliable vehicle feature vector information to be obtained when performing advanced feature extraction on the vibration response signal subsequently. Furthermore, through the multi-dimensional feature vector similarity comparison algorithm, the true crack state of the brake leaf spring can be determined more precisely, improving the accuracy and reliability of crack diagnosis.

[0071] In some embodiments, the feature sets in the preset standard sample library include a standard healthy brake leaf spring feature set and a standard cracked brake leaf spring feature set; the steps of using the multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with the feature sets in the preset standard sample library, obtaining a comparison result, and determining the true crack state of the brake leaf spring according to the comparison result specifically include: Calculate the average distance DH between the vehicle feature vector information and all vectors in the standard healthy brake leaf spring feature set, and the average distance DC between the vehicle feature vector information and all vectors in the standard cracked brake leaf spring feature set; If DC < DH and DC is less than a preset threshold, it is determined that the brake leaf spring is in a cracked state; If DH < DC and DH is less than a preset threshold, it is determined that the brake leaf spring is in a healthy state.

[0072] Among them, the standard healthy brake leaf spring feature set refers to the set of feature vectors representing the healthy state constructed by collecting vibration response signals and performing advanced feature extraction on a large number of known healthy brake leaf springs. It can be constructed by means of clustering analysis, principal component analysis or expert experience annotation, etc., providing a clear healthy reference benchmark for subsequent crack state determination.

[0073] The standard cracked brake leaf spring feature set refers to the set of feature vectors representing the cracked state constructed by collecting vibration response signals and performing advanced feature extraction on a large number of known brake leaf springs with cracks of different degrees. It can be constructed by means of machine learning classification, anomaly detection or manual annotation, etc., and is used to provide a clear cracked reference benchmark for subsequent crack state determination.

[0074] DH and DC can be measured by distance calculation methods such as Euclidean distance, Manhattan distance or cosine similarity.

[0075] The preset threshold refers to the critical value used to compare the average distance when judging the status of the brake leaf spring. It can be determined based on historical data analysis, expert experience, or through machine learning model training optimization, and can provide an adjustable judgment standard to balance the false alarm rate and missed alarm rate.

[0076] This application provides a clear classification basis for subsequent judgment by subdividing the preset standard sample library into a standard healthy brake leaf spring feature set and a standard cracked brake leaf spring feature set. After receiving the vehicle feature vector information obtained through advanced feature extraction, the system will calculate the average distance DH between the vehicle feature vector information and all vectors in the standard healthy brake leaf spring feature set, as well as the average distance DC between the vehicle feature vector information and all vectors in the standard cracked brake leaf spring feature set. This average distance calculation method can effectively reduce the impact of a single abnormal sample or local noise on the overall comparison results, thereby enhancing the robustness of the comparison. Subsequently, the system will make a dual judgment based on the relative size relationship between DH and DC, combined with a preset threshold. Specifically, when the average distance DC of the crack state is smaller than the average distance DH of the healthy state, and DC is also smaller than the preset threshold, the system determines that the brake leaf spring is in a cracked state, which indicates that the vehicle feature vector information is closer to the crack sample set, and the degree of closeness reaches the preset crack standard. Conversely, when the average distance DH for the healthy state is less than the average distance DC for the cracked state, and DH is also less than a preset threshold, the system determines that the brake leaf spring is in a healthy state. This indicates that the vehicle feature vector information is closer to the healthy sample set and the degree of closeness meets the preset health standard. This judgment mechanism, combining relative distance and absolute thresholds, effectively avoids misjudgments that could result from relying solely on a single similarity metric, significantly improving the accuracy and reliability of the brake leaf spring crack state judgment. Furthermore, this technical solution, combined with previous methods for acquiring vehicle feature vector information, can deliver even greater advantages. Previous methods, by receiving conflicting data from multiple sources, issuing standardized braking commands, collecting vibration responses at a high sampling rate, and performing advanced feature extraction, ensure that the vehicle feature vector information input to this judgment step is high-quality data that has undergone standardized excitation, denoising, and rich discriminative capabilities. It is precisely because of this high-quality and highly reliable vehicle feature vector information that the judgment logic based on average distance and dual thresholds employed in this technical solution can fully demonstrate its effectiveness and accurately capture subtle changes in the brake leaf spring's state. This combination enables the system to not only overcome the limitations of traditional similarity comparison, but also effectively deal with the noise and data deviations that exist in actual applications. In the case of conflicting information from multiple sources, it can provide a more accurate and reliable judgment of the actual crack status of the brake leaf spring, avoiding the safety hazards caused by misjudgment and the continued operation of the vehicle.

[0077] Exemplarily, the present application is implemented as follows: First, a preset standard sample library can be stored in a distributed database. The standard healthy brake leaf spring feature set can be a collection of thousands of vibration feature vectors of healthy brake leaf springs, obtained through standardized testing and feature extraction of brake leaf springs from a large number of normally operating vehicles. The standard cracked brake leaf spring feature set can be a collection of thousands of vibration feature vectors of brake leaf springs known to have cracks of varying types and severity, obtained through testing and feature extraction of faulty vehicles or laboratory simulated crack samples. After the system obtains the vehicle feature vector information of the vehicle to be analyzed, for example, a 10-dimensional feature vector, the system invokes a distance calculation module. This module uses Euclidean distance as a distance metric to calculate the Euclidean distance between the vehicle feature vector information and each vector in the standard healthy brake leaf spring feature set, then averages these distances to obtain the average distance DH. Simultaneously, the module also calculates the Euclidean distance between the vehicle feature vector information and each vector in the standard cracked brake leaf spring feature set, and averages these distances to obtain the average distance DC. For example, the preset threshold can be set to 0.5, which is obtained through cross-validation and optimization of a large amount of historical data. Subsequently, a decision logic unit receives DH, DC, and the preset threshold. If the calculated DC is 0.3, DH is 0.8, and the preset threshold is 0.5, since DC (0.3) is less than DH (0.8), and DC (0.3) is less than the preset threshold (0.5), the decision logic unit outputs that the brake leaf spring is in a cracked state. Conversely, if DH is 0.2, DC is 0.7, and the preset threshold is 0.5, since DH (0.2) is less than DC (0.7), and DH (0.2) is less than the preset threshold (0.5), the decision logic unit outputs that the brake leaf spring is in a healthy state. This specific implementation ensures the accuracy and reliability of the brake leaf spring condition determination.

[0078] Through the above technical solution, the present application can effectively solve the problems of insufficient accuracy and susceptibility to noise in the traditional similarity comparison method in determining the crack status of brake leaf springs. By subdividing the standard sample library into two feature sets, healthy and cracked, and calculating the average distance between the feature vector information of the vehicle to be tested and these two types of feature sets, the overall attribution tendency of the sample to be tested can be measured more comprehensively. Combining the relative size comparison of the average distance and the absolute threshold judgment, a dual verification mechanism is formed, which significantly enhances the accuracy and reliability of the judgment. This method can effectively distinguish between the healthy state and the crack state of the brake leaf spring, reduce the misjudgment rate caused by noise or data deviation, thereby ensuring the accurate identification of the true crack state of the brake leaf spring and avoiding the continued operation of vehicles with safety hazards.

[0079] In some embodiments, after the step of receiving the maintenance report from the third-party repair station, the vehicle vibration sensor data, and the user feedback data: Performing status extraction on the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data to obtain a status indication corresponding to each data source; assigning reliability weights to the status indications based on the type and historical performance of the respective data sources; Calculating a propensity score of each data source for the brake leaf spring state according to the state indication and the reliability weight; The propensity scores are compared, and when a preset difference exists between the propensity scores, a contradiction between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data is identified.

[0080] This involves converting data of different formats and types (such as structured maintenance reports, time-series vehicle vibration sensor data, and unstructured user feedback data) into a unified, standardized state description or numerical representation to achieve state extraction. This can be achieved using technologies such as natural language processing, signal processing, or data parsing to unify heterogeneous data and lay the foundation for subsequent quantitative comparison. The status indication refers to the standardized description or quantitative value of the current status of the brake leaf spring from each data source after status extraction. It can be expressed as a discrete category (such as "healthy", "abnormal", "to be observed") or a continuous value, which is used to provide a unified and comparable brake leaf spring status assessment benchmark; Reliability weight refers to the degree of trust or influence coefficient assigned to status indications based on the inherent characteristics, historical accuracy, or data quality assessment results of each data source. It can be determined through expert experience assignment, statistical analysis of historical data, or dynamic adjustment of machine learning models. It is used to quantify the confidence level of different data sources, so as to give more reliable data greater influence in conflicting judgments. The propensity score refers to the quantitative value of each data source's judgment on the brake leaf spring status after comprehensively considering the status indication and its reliability weight. It can be calculated using weighted summation, weighted average or other aggregation algorithms to integrate the judgments and credibility of each data source into a single, directly comparable value. Among them, the preset difference refers to the threshold or rule set used to determine whether there is a contradiction between the propensity scores. It can be determined based on actual application scenarios, empirical values ​​or through historical data training to provide a judgment standard to avoid misjudging minor fluctuations as contradictions, while ensuring sensitivity to real contradictions.

[0081] This application effectively addresses the challenge of identifying discrepancies between heterogeneous data through a series of refined processing steps. First, status extraction is performed on maintenance reports, onboard vibration sensor data, and user feedback data. This process transforms data with diverse formats and ambiguous semantics into standardized status indicators. For example, text descriptions in maintenance reports, waveform features in sensor data, and natural language user feedback are all mapped into standardized status descriptions such as "healthy," "abnormal," or "degraded." This conversion forms the basis for subsequent quantitative analysis and ensures comparability across different data sources. Furthermore, these status indicators are assigned reliability weights based on the type and historical performance of each data source. This takes into account the inherent accuracy and historical performance differences between different data sources. For example, rigorously calibrated sensor data is generally more objective than subjective user feedback, while reports from large, professional repair shops may be more reliable than those from smaller ones. By assigning weights, the system intelligently assesses the confidence level of each data source, ensuring that more reliable information has a greater impact in subsequent judgments, thereby reducing the risk of misleading information caused by low-quality data. Subsequently, based on the acquired status indicators and reliability weights, a propensity score for each data source regarding the brake leaf spring condition is calculated. This step quantifies the status indications and weights them based on their reliability weights, resulting in a single numerical value that intuitively reflects each data source's "propensity" for judging the health of the brake leaf spring. For example, a highly reliable "healthy" indication yields a high positive score, while a highly reliable "abnormal" indication yields a low negative score. This quantification process enables the judgments of different data sources to be compared on a unified numerical scale. Ultimately, by comparing these propensity scores, the system can identify discrepancies between maintenance reports, onboard vibration sensor data, and user feedback when there are preset differences between them. This comparison mechanism can capture potential inconsistencies between data sources. For example, when a maintenance report leans toward "healthy" while sensor data or user feedback strongly leans toward "abnormal," the system can accurately identify a discrepancy. By setting the preset difference, the sensitivity of discrepancy detection can be flexibly adjusted, avoiding frequent triggering due to minor fluctuations while ensuring timely response to real safety hazards. Through this refined discrepancy detection mechanism, the present application can accurately determine inconsistencies between heterogeneous data sources, thereby providing accurate triggering conditions for the subsequent issuance of standardized braking commands. This means the system only initiates the time-consuming and resource-intensive standardized excitation and data acquisition process when a genuine data discrepancy exists and further verification of the brake leaf spring's true condition is required. This precise triggering mechanism avoids unnecessary testing, improves diagnostic efficiency, and ensures that the true crack condition of the brake leaf spring is determined based on more reliable and comprehensive information, significantly enhancing the accuracy and robustness of the entire brake leaf spring crack data analysis method.

[0082] In some embodiments, the step of assigning reliability weights to the status indications based on the types and historical performance of the respective data sources includes: assigning an initial reliability weight to the status indication based on the type and historical performance of each data source; continuously monitoring the consistency between the status indications of the respective data sources and the actual crack status of the brake leaf spring; According to the consistency, the initial reliability weight is adjusted to obtain an adjusted reliability weight, and the adjusted reliability weight is used as the reliability weight of the status indication.

[0083] Among them, the initial reliability weight refers to a preliminary assessment of the credibility of the data source based on the type and historical performance of each data source. It can be set by expert experience, statistical analysis based on historical data sets, or a preset rule base to provide a benchmark for subsequent dynamic adjustments.

[0084] The degree of conformity between the status indications given by each data source and the actual crack status of the brake leaf spring is compared in real time or periodically, and the consistency between the status indications of each data source and the actual crack status of the brake leaf spring is continuously monitored. This can be achieved by comparing the indications of the data source with manual detection results, higher-level diagnostic system results or subsequent actual fault occurrences, so as to obtain the performance data of the data source in actual operation and provide a basis for weight adjustment.

[0085] Based on the monitored consistency data, the preset initial reliability weights are dynamically revised. This can be achieved by using machine learning algorithms, adaptive filtering algorithms, or rule-based iterative update mechanisms. The purpose is to enable the weights to adapt to the real-time performance of the data source and environmental changes, thereby improving the accuracy of the assessment.

[0086] This application addresses the potential inaccurate assessments that can result from relying solely on initial weights by introducing a mechanism for dynamically adjusting reliability weights. Specifically, the system first assigns an initial reliability weight to each data source's status indication based on its type and historical performance, such as the professionalism of maintenance reports, the objectivity of on-board vibration sensor data, and the subjectivity of user feedback. This initial weight reflects a priori confidence in the data source. However, given that the actual performance of a data source may vary over time or deviate, this technical solution further continuously monitors the consistency between the status indications of each data source and the actual crack status of the brake leaf spring. This means that the system continuously compares the status reported by the data source with the actual crack status obtained through more reliable means (such as actual testing or subsequent fault confirmation). If a data source's indications consistently differ from the actual status, the system adjusts the initial reliability weight of that data source based on this discrepancy. For example, if a repair station's maintenance reports repeatedly differ from the actual crack status, its weight will be reduced. Conversely, if a data source's indications consistently match the actual status, its weight will be increased. Through this continuous monitoring and dynamic adjustment, the system is able to obtain adjusted reliability weights that more accurately reflect the current credibility of the data source. These adjusted reliability weights are then used to calculate the propensity score of each data source regarding the brake leaf spring status. This allows for more accurate weighting when identifying conflicts between maintenance reports, on-board vibration sensor data, and user feedback data, significantly improving the accuracy of conflict identification and the reliability of determining the true crack status of the brake leaf spring. It is precisely because of this dynamic learning and adaptability that this technical solution can effectively overcome the limitations of single or fixed weight assessments, ensuring that the system can still make accurate judgments when conflicts exist between heterogeneous information from multiple sources.

[0087] Exemplarily, the present application is implemented as follows: The system first assigns initial reliability weights to status indications based on the type and historical performance of each data source. For example, maintenance reports from authoritative third-party repair stations may be assigned a higher initial weight, such as 0.8; vehicle-mounted vibration sensor data, due to its objectivity, may be assigned a medium weight, such as 0.6; and user feedback data, due to its subjectivity, may be assigned a lower initial weight, such as 0.4. The system then continuously monitors the consistency between the status indications from each data source and the actual crack condition of the brake leaf spring. This can be achieved by establishing a feedback loop. For example, after the vehicle undergoes manual inspection or more precise diagnostic equipment and confirms the actual crack condition, the system compares this actual condition with the status indications previously provided by each data source. If the indication from a data source differs from the actual condition, for example, a maintenance report may indicate "no cracks" but actual inspection reveals "cracks," the system will record this discrepancy. Based on this consistency or discrepancy, the system dynamically adjusts the initial reliability weight. For example, an iterative update algorithm can be employed. If a data source's indication is consistent with the true state, its weight is slightly increased; if it is inconsistent, its weight is correspondingly decreased. After multiple iterations and learning, the system obtains adjusted reliability weights that more accurately reflect the actual credibility of the current data source. Ultimately, the system uses these adjusted reliability weights as the reliability weights for the state indication in subsequent propensity score calculations and conflict identification, thereby improving the accuracy of brake leaf spring crack status determination.

[0088] In some embodiments, the step of extracting status from the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data to obtain status indications corresponding to each data source includes: performing structured information parsing and keyword extraction on the maintenance report, and mapping the parsing results into a status indication of the brake leaf spring; Extracting time-domain or frequency-domain features from the vehicle-mounted vibration sensor data, and comparing the extracted features with a preset feature pattern to obtain a status indication of the brake leaf spring; Furthermore, natural language processing is performed on the user feedback data to identify keywords or phrases related to the state of the brake leaf spring, and the identification results are mapped into a state indication of the brake leaf spring.

[0089] Among them, data with predefined formats or fields are identified, extracted and classified to realize structured information parsing, which can be achieved by using regular expression matching, template parsing, or entity recognition models based on machine learning.

[0090] The process of identifying words or phrases that can represent the core content of unstructured text to achieve keyword extraction can be implemented using sequence labeling models based on word frequency statistics (such as TF-IDF), graph models (such as TextRank), or deep learning.

[0091] Extract parameters that can characterize the characteristics of the original signal in the time domain or frequency domain, and extract time domain or frequency domain features. This can be achieved by using statistical analysis (such as mean, variance, peak), transform analysis (such as Fourier transform, wavelet transform), or feature learning methods based on deep learning.

[0092] A preset feature pattern refers to a set of typical features that are pre-defined based on historical data or expert experience before data analysis and used for comparison and classification. It can exist in the form of cluster centers, classifier model parameters, or expert rule bases.

[0093] This application utilizes a customized data extraction method tailored to the characteristics of different data sources to accurately and comprehensively extract effective information reflecting the true condition of brake leaf springs from heterogeneous data sources. Specifically, since maintenance reports typically contain structured test data and unstructured maintenance personnel descriptions, this application uses structured information parsing to accurately extract quantitative indicators such as wear level and replacement date. Furthermore, through keyword extraction, key descriptions such as "cracks" and "abnormal wear" can be identified from free text. The combination of these two approaches enables comprehensive and detailed capture of information from maintenance reports and uniform mapping of these information into brake leaf spring status indicators. Given that on-board vibration sensor data can directly reflect changes in the physical properties of brake leaf springs, this application uses time-domain or frequency-domain feature extraction to convert raw vibration signals into quantifiable feature vectors, such as energy distribution and frequency components. These extracted features are then compared with pre-determined characteristic patterns of healthy or cracked brake leaf springs to determine whether the current brake leaf spring vibration state is abnormal and obtain a corresponding status indication. This feature comparison-based approach effectively identifies subtle vibration pattern changes caused by cracks in brake leaf springs. Furthermore, considering that user feedback data, as unstructured and subjective descriptions, can provide intuitive perceptions and abnormalities that are difficult to capture in sensors and maintenance reports, this application utilizes natural language processing technology to identify key words or phrases related to brake leaf spring status, such as "soft brakes" and "unusual noise," from user feedback text and map them into status indicators. This processing approach overcomes the unstructured nature of user feedback data by transforming it into analyzable status information. By employing targeted and refined status extraction methods for maintenance reports, on-board vibration sensor data, and user feedback data, more accurate, comprehensive, and reliable brake leaf spring status indicators can be obtained from these heterogeneous and potentially conflicting data sources. These refined status indicators serve as the basis for subsequent data conflict identification, providing high-quality input for determining whether there are pre-defined discrepancies between maintenance reports, on-board vibration sensor data, and user feedback data. Furthermore, by combining reliability weights assigned to each data source based on its type and historical performance, and continuously monitoring the consistency between status indicators and actual crack status to adjust the reliability weights, the system can more accurately identify data conflicts and provide solid data support for determining the actual crack status of the brake leaf spring.

[0094] In some embodiments, the step of calculating the propensity score of each data source for the brake leaf spring state includes: The status indications of the data sources are converted into numerical values, and weighted summation or weighted averaging of the numerical values ​​is performed according to the reliability weights to obtain the propensity score.

[0095] Among them, the status information expressed by different forms of data sources is uniformly quantified into a numerical form that can be used for mathematical operations. It can be implemented using preset mapping rules, coding tables or through data processing algorithms to provide a unified data basis for subsequent comprehensive calculations.

[0096] The converted values ​​are multiplied by the corresponding reliability weights and then accumulated, or the weighted values ​​are averaged to achieve weighted summation or weighted average. This can be achieved using linear weighting, exponential weighting or other nonlinear weighting algorithms, so that data sources with higher reliability make a greater contribution to the final result, thereby reducing the negative impact of unreliable data sources.

[0097] This application unifies the status indications from heterogeneous data sources such as maintenance reports, vehicle-mounted vibration sensors, and user feedback into a unified numerical process, thereby eliminating the differences between different data types and enabling quantitative comparison and calculation of this information. On this basis, these values ​​are weighted summed or averaged according to predetermined or dynamically adjusted reliability weights, so that data sources with more reliable historical performance and higher data quality occupy a larger proportion in the final propensity score calculation, while the influence of data sources with lower reliability is effectively weakened. It is precisely because of this numerical and weighted processing that the propensity scores of each data source for the brake leaf spring status can more accurately reflect its true status and effectively reduce the interference of a single unreliable data source on the overall judgment. The propensity scores obtained in this way can then be used for comparison. When there are preset differences between these propensity scores, the system can more accurately identify the contradictions between maintenance reports, vehicle-mounted vibration sensor data, and user feedback data, thereby triggering further detection of the brake leaf spring status. This processing mechanism ensures that even when faced with conflicting data of varying reliability, the system can make judgments based on more comprehensive, weighted information, significantly improving the accuracy of conflict identification and the reliability of subsequent crack status assessment.

[0098] On the other hand, Figure 2 As shown, a brake leaf spring crack data analysis system is exemplarily shown. The present application further proposes a brake leaf spring crack data analysis system 100, which includes: The data receiving module 10 is used to receive maintenance reports from third-party repair stations, vehicle vibration sensor data, and user feedback data; an excitation control module 20 for, upon identifying a discrepancy between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data, and when the vehicle engine is turned off, issuing a standardized braking command to the brake system of the target vehicle to drive the brake caliper to perform a clamping action, causing the brake leaf spring to receive a standardized excitation to generate a vibration response; wherein the brake caliper clamps a brake pad assembly, which includes a brake leaf spring and a brake friction pad; a vibration response acquisition module 30 for sending a synchronous trigger signal to an accelerometer when the standardized braking instruction is issued, so that the accelerometer acquires a target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate; Feature extraction module 40: performs high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information; The crack diagnosis module 50 uses a multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with the feature sets in the preset crack sample library and the healthy sample library. When the similarity with the crack sample library exceeds a first threshold, a crack diagnosis conclusion is output; when the similarity with the healthy sample library exceeds a second threshold, a healthy diagnosis conclusion is output.

[0099] Through the above technical solution, the present application provides a brake leaf spring crack data analysis system, which solves the problem in the prior art that it is difficult to judge the status of the brake leaf spring due to the contradictions between maintenance reports, vehicle-mounted vibration sensor data and user feedback data. The system integrates multi-source heterogeneous data through the data receiving module, and when the data contradiction is identified, the excitation control module actively sends a standardized braking instruction to the braking system to drive the brake caliper to perform a clamping action, so that the brake leaf spring is subjected to standardized excitation and generates a vibration response. The vibration response acquisition module synchronously collects the target vibration response signal at a high sampling rate to ensure the reliability of the data. The feature extraction module performs high-level feature extraction on the signal to obtain vehicle feature vector information, reducing the data dimension and retaining the core diagnostic information. The crack diagnosis module uses a multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with the preset crack sample library and healthy sample library, so that it can output the crack or health diagnosis conclusion of the brake leaf spring. The collaborative work of this series of modules enables the system to obtain diagnostic basis through active and standardized stimulus and response collection when faced with complex and contradictory raw data, thereby improving the accuracy of diagnosis of brake leaf spring crack conditions and avoiding safety hazards caused by misjudgment.

[0100] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A brake leaf spring crack data analysis method, characterized in that: include: Receive maintenance reports from third-party repair stations, vehicle vibration sensor data, and user feedback data; Upon identifying a discrepancy between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data, and when the vehicle engine is turned off, issuing a standardized braking command to the braking system of the target vehicle to drive the brake caliper to perform a clamping action, causing the brake leaf spring to receive a standardized excitation to generate a vibration response; wherein the brake caliper clamps a brake pad assembly, and the brake pad assembly includes a brake leaf spring and a brake friction pad; When the standardized braking instruction is issued, a synchronous trigger signal is sent to the accelerometer, and the accelerometer collects the target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate; Performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information; A multi-dimensional feature vector similarity comparison algorithm is used to compare the vehicle feature vector information with a feature set in a preset standard sample library to obtain a comparison result, and the actual crack state of the brake leaf spring is determined based on the comparison result.

2. The brake leaf spring crack data analysis method according to claim 1, characterized in that: Before the step of issuing a standardized braking command to the braking system of the target vehicle to drive the brake caliper to perform the clamping action, the method further includes: A standardized physical surface pre-processing instruction is issued to the brake system to drive the brake caliper to perform multiple clamping and releasing operations on the brake leaf spring according to a set pressure and duration; wherein the clamping and releasing operations cause the brake friction pad in the brake pad assembly to rub against the brake disc to remove an unstable oxide film on the friction surface.

3. The brake leaf spring crack data analysis method according to claim 1, characterized in that: The step of performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information includes: Performing three-level wavelet packet decomposition processing on the target vibration response signal to obtain a sub-band signal; Extracting time domain characteristic parameters and frequency domain characteristic parameters of each sub-band signal, wherein: the time domain characteristic parameters include: root mean square value, kurtosis coefficient and impulse factor; the frequency domain characteristic parameters include: main frequency band energy proportion and frequency band energy entropy; The time domain feature parameters and the frequency domain feature parameters are combined into a multi-dimensional feature vector as the vehicle feature vector information.

4. The brake leaf spring crack data analysis method according to claim 1, characterized in that: The step of sending a synchronous trigger signal to an accelerometer at the same time as the standardized braking instruction is issued, and the accelerometer collecting a target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate includes: Repeating the application of the standardized stimulus a preset number of times; Each time the standardized excitation is applied, a vibration response signal generated by the brake leaf spring under the standardized excitation is synchronously collected to obtain a plurality of vibration response signals; Multiple vibration response signals are time-aligned and superimposed to suppress noise to obtain the target vibration response signal.

5. The brake leaf spring crack data analysis method according to claim 1, characterized in that: The feature set in the preset standard sample library includes a standard healthy brake leaf spring feature set and a standard cracked brake leaf spring feature set; using a multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with the feature set in the preset standard sample library to obtain a comparison result, and determining the actual crack state of the brake leaf spring based on the comparison result specifically includes the following steps: Calculating an average distance DH between the vehicle feature vector information and all vectors in the standard healthy brake leaf spring feature set, and an average distance DC between the vehicle feature vector information and all vectors in the standard cracked brake leaf spring feature set; If DC < DH, and DC is less than a preset threshold, it is determined that the brake leaf spring is in a crack state; If DH < DC, and DH is less than a preset threshold, it is determined that the brake leaf spring is in a healthy state.

6. The brake leaf spring crack data analysis method according to claim 1, characterized in that: After the step of receiving the maintenance report from the third-party repair station, the vehicle vibration sensor data and the user feedback data, the following steps are included: Performing status extraction on the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data to obtain a status indication corresponding to each data source; assigning reliability weights to the status indications based on the type and historical performance of the respective data sources; Calculating a propensity score of each data source for the brake leaf spring state according to the state indication and the reliability weight; The propensity scores are compared, and when a preset difference exists between the propensity scores, a contradiction between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data is identified.

7. The brake leaf spring crack data analysis method according to claim 6, characterized in that: The step of assigning reliability weights to the status indications according to the types and historical performances of the data sources includes: assigning an initial reliability weight to the status indication based on the type and historical performance of each data source; continuously monitoring the consistency between the status indications of the respective data sources and the actual crack status of the brake leaf spring; According to the consistency, the initial reliability weight is adjusted to obtain an adjusted reliability weight, and the adjusted reliability weight is used as the reliability weight of the status indication.

8. The brake leaf spring crack data analysis method according to claim 6, characterized in that: The step of extracting the status of the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data to obtain the status indication corresponding to each data source includes: performing structured information parsing and keyword extraction on the maintenance report, and mapping the parsing results into a status indication of the brake leaf spring; Extracting time-domain or frequency-domain features from the vehicle-mounted vibration sensor data, and comparing the extracted features with a preset feature pattern to obtain a status indication of the brake leaf spring; Furthermore, natural language processing is performed on the user feedback data to identify keywords or phrases related to the state of the brake leaf spring, and the identification results are mapped into a state indication of the brake leaf spring.

9. The brake leaf spring crack data analysis method according to claim 6, characterized in that: The step of calculating the propensity score of each data source for the brake leaf spring state includes: The status indications of the data sources are converted into numerical values, and weighted summation or weighted averaging of the numerical values ​​is performed according to the reliability weights to obtain the propensity score.

10. A brake leaf spring crack data analysis system, characterized in that: The system includes: Data receiving module, used to receive maintenance reports from third-party repair stations, vehicle vibration sensor data and user feedback data; an excitation control module, configured to, upon identifying a discrepancy between the maintenance report, the vehicle-mounted vibration sensor data, and the user feedback data, and when the vehicle's engine is turned off, issue a standardized braking command to the target vehicle's braking system to drive the brake caliper to perform a clamping action, causing the brake leaf spring to receive a standardized excitation to generate a vibration response; wherein the brake caliper clamps a brake pad assembly, which includes a brake leaf spring and a brake friction pad; a vibration response acquisition module, configured to send a synchronous trigger signal to an accelerometer when the standardized braking instruction is issued, so that the accelerometer acquires a target vibration response signal generated by the brake leaf spring under the standardized excitation at a preset high sampling rate; Feature extraction module: performing high-level feature extraction on the target vibration response signal to obtain vehicle feature vector information; The crack diagnosis module uses a multi-dimensional feature vector similarity comparison algorithm to compare the vehicle feature vector information with the feature sets in the preset crack sample library and the healthy sample library. When the similarity with the crack sample library exceeds a first threshold, a crack diagnosis conclusion is output; when the similarity with the healthy sample library exceeds a second threshold, a healthy diagnosis conclusion is output.

Images