Abnormality detection method and device for vehicle and vehicle

Abstract

This invention provides a method, device, and vehicle for anomaly detection. It collects strain voltage signals representing the torque experienced by the vehicle's windshield wipers during wiping motion; performs feature processing on the strain voltage signals to extract multi-dimensional features; and applies a pre-defined anomaly detection model to assess wiper blade wear and glass contamination using these multi-dimensional features, obtaining anomaly detection results. The anomaly detection model employs a dual-branch network structure to simultaneously detect wiper blade wear and glass contamination. By performing multi-scale decomposition on the strain voltage signals, feature vectors reflecting wear, contamination, and impact are extracted, thereby assessing wiper blade wear and glass contamination. This method accurately obtains anomaly detection results and proactively alerts the user to replacement before significant performance degradation, enabling early fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of vehicle condition monitoring technology, specifically to a method, device, and vehicle for detecting vehicle anomalies. Background Technology

[0002] Traditional vehicles rely on experience to determine when to replace wipers, and cannot distinguish between different types of pollution such as rainwater, oil film, insects, and mud. Furthermore, they can only detect the wear of wiper blades when they are severely worn or when the mechanism is stuck and makes abnormal noises. They cannot quantify the wear status of wiper blades, lack status perception, have poor environmental adaptability, and lack early warning. Summary of the Invention

[0003] In view of this, the present invention aims to provide a method, device and vehicle for detecting vehicle anomalies. By performing multi-scale decomposition on the strain voltage signal, feature vectors reflecting wear, contamination and impact are extracted, and then wear and contamination are assessed. This can accurately obtain the abnormality detection results of the vehicle, and proactively remind replacement before significant performance degradation, thus achieving early fault diagnosis.

[0004] According to one aspect of the present invention, an embodiment of the present invention provides a vehicle anomaly detection method, comprising: acquiring a strain voltage signal of the magnitude of the torque experienced by the vehicle's windshield wiper during wiping motion; performing feature processing on the strain voltage signal to extract multi-dimensional features; applying a preset anomaly detection model to evaluate the wear of the windshield wiper blade and the contamination of the glass opposite the windshield wiper using the multi-dimensional features, thereby obtaining anomaly detection results for the vehicle, wherein the anomaly detection model adopts a dual-branch network structure for simultaneously performing wear detection on the wiper blade and contamination detection on the glass.

[0005] In this embodiment, by performing multi-scale decomposition on the strain voltage signal, feature vectors reflecting wear, contamination, and impact are extracted, and then wear and contamination assessment is performed. This can accurately obtain abnormal vehicle detection results and proactively remind replacement before performance deteriorates significantly, thus achieving early fault diagnosis.

[0006] Optionally, the step of performing feature processing on the strain voltage signal to extract multidimensional features includes: performing time-domain processing on the strain voltage signal to extract time-domain statistical features; performing a fast Fourier transform on the strain voltage signal and extracting spectral features; performing multi-level wavelet packet decomposition on the strain voltage signal to obtain multiple sub-frequency bands, and extracting time-frequency domain wavelet packet features based on the multiple sub-frequency bands; and combining the time-domain statistical features, the spectral features, and the time-frequency domain wavelet packet features to obtain multidimensional features.

[0007] In this embodiment of the application, by analyzing the time-domain, frequency-domain, and time-frequency-domain characteristics of the strain voltage signal with high frequency and high precision, it is possible to easily separate the load changes of two different sources, scraper wear and glass contamination, from a single strain signal, and achieve independent quantitative evaluation.

[0008] Optionally, the application of a preset anomaly detection model to assess the wiper blade wear and the contamination of the glass opposite the wiper using the multidimensional features, and to obtain the anomaly detection results of the vehicle, includes: applying a wear assessment branch network in the anomaly detection model to process the multidimensional features and the strain voltage signal to obtain the wiper blade wear features; applying a contamination assessment branch network in the anomaly detection model to process the multidimensional features and the strain voltage signal to obtain the glass contamination features; and applying a feature cross-attention module in the anomaly detection model to share information between the wear features and the contamination features to obtain the wiper blade wear detection results and the glass contamination detection results of the vehicle.

[0009] In this embodiment, load changes from two different sources—blade wear and glass contamination—are separated from a single strain signal to achieve independent quantitative assessment. Furthermore, the acquired wear and contamination characteristics are decoupled and jointly judged under two states, thereby accurately detecting vehicle anomalies.

[0010] Optionally, the application of the feature cross-attention module in the anomaly detection model to share information between the wear features and the contamination features to obtain the wiper wear detection results and glass contamination detection results of the vehicle includes: using the wear features as a query, retrieving a first relevant information from the contamination features, and fusing the first relevant information with the wear features to obtain the wiper wear status of the vehicle, wherein the first relevant information is a context vector generated from feature information associated with the wear features in the contamination features; using the contamination features as a query, retrieving a second relevant information from the wear features, and fusing the second relevant information with the contamination features to obtain the glass contamination status of the vehicle.

[0011] In this embodiment of the application, by performing dual-state joint decoupling of wear characteristics and contamination characteristics, the wear condition of the vehicle's wiper blades and the contamination condition of the glass can be accurately obtained, so as to facilitate vehicle adjustments.

[0012] Optionally, the method further includes: adjusting the control parameters of the vehicle based on the anomaly detection result; and / or, predicting the remaining lifespan of the vehicle based on the anomaly detection result, obtaining the remaining lifespan of the vehicle, and issuing graded warnings to the vehicle based on the anomaly detection result or the remaining lifespan.

[0013] In this embodiment, the vehicle's control parameters can be adjusted based on the anomaly detection results to achieve the best control effect. The remaining lifespan can also be predicted, and the vehicle can be given graded warnings based on the anomaly detection results, so that the driver can understand the actual situation of the vehicle in a timely manner and take corresponding countermeasures.

[0014] Optionally, adjusting the vehicle's control parameters based on the anomaly detection results includes: obtaining candidate control parameters for the vehicle based on the anomaly detection results; applying a model predictive control method to obtain a predicted cleanliness of the glass based on the anomaly detection results and the candidate control parameters; determining an objective function based on the predicted cleanliness and the anomaly detection results; and selecting the vehicle's control parameters that minimize the objective function from the candidate control parameters.

[0015] In this embodiment of the application, by applying the model predictive control method to determine the control parameters from the candidate control parameters determined based on the anomaly detection results, the water spraying strategy, brush speed and pressure can be dynamically optimized according to the real-time identified working conditions, so as to achieve on-demand cleaning and full-condition adaptive operation.

[0016] Optionally, the anomaly detection result includes a wiper blade health index; the step of classifying and issuing warnings to the vehicle based on the anomaly detection result or the remaining lifespan includes: if the anomaly detection result determines that the frame used to install the wiper blade has a risk of contacting the glass, then outputting a first prompt message, which suggests stopping the use of the current wiper blade; if the wiper blade health index is less than a first health threshold, or if unilateral wear of the wiper blade is detected, then outputting a second prompt message each time the wiper is used, which indicates that the current wiper blade is severely worn; if the wiper blade health index is less than a second health threshold and greater than or equal to the first health threshold, or the remaining lifespan is less than the first lifespan, then outputting a third prompt message when the wiper is activated, which suggests replacing the current wiper blade; if the wiper blade health index is less than a third health threshold and greater than or equal to the second health threshold, or the remaining lifespan is less than the second lifespan and the second lifespan is greater than the first lifespan, then outputting a fourth prompt message, which indicates that the current wiper blade is performing well.

[0017] In this embodiment, the vehicle is given multiple levels of warnings based on abnormal detection results and remaining lifespan, and is proactively reminded to replace it before performance deteriorates significantly, thus enabling early fault diagnosis.

[0018] Optionally, the abnormal detection results include a wiper blade health index, a glass cleanliness index, and a contamination type. Obtaining candidate control parameters for the vehicle based on the abnormal detection results includes: if the wiper blade health index is greater than a fourth health threshold, or the glass cleanliness index is less than or equal to a first cleaning threshold, then determining the candidate control parameters for the vehicle includes at least one of: reducing the wiper blade pressure, delaying the water spray time, and reducing the wiper blade return speed; if the wiper blade health index is less than or equal to the fourth health threshold and greater than a fifth health threshold, or the glass cleanliness index is greater than a second cleaning threshold and less than or equal to a third cleaning threshold, then determining the candidate control parameters includes: reducing the wiper blade pressure, delaying the water spray time, and reducing the wiper blade return speed. The candidate control parameters for the vehicle include at least one of the following: increasing the water spray volume, adding cleaning agent, increasing the wiping pressure of the wiper blade, and increasing the number of wiping strokes; if the glass cleanliness index is greater than the third cleaning threshold and the contamination type is spot stains, then the candidate control parameters for the vehicle include: point spraying and / or multiple short-stroke wiping based on the location of the spot stains, wherein the short-stroke wiping refers to wiping the location of the spot stains in the wiping stroke; if the wiper blade health index is less than or equal to the sixth health threshold, then the candidate control parameters for the vehicle include: water lubrication during each wiping stroke and / or limiting the maximum wiping speed.

[0019] In this embodiment of the application, different candidate control parameters are obtained based on different wiper blade health indices, glass cleanliness indices and pollution types, so that drivers can minimize wiper blade wear and keep the glass as clean as possible when operating the wipers.

[0020] According to another aspect of the present invention, an embodiment of the present invention provides a vehicle anomaly detection device, comprising: an information acquisition module for acquiring strain voltage signals of the torque magnitude experienced by the vehicle's windshield wipers during wiping motion; a feature extraction module for performing feature processing on the strain voltage signals to extract multi-dimensional features; and an anomaly detection module for applying a preset anomaly detection model to assess the wear of the windshield wiper blades and the contamination of the glass opposite the windshield wipers using the multi-dimensional features, thereby obtaining anomaly detection results for the vehicle. The anomaly detection model adopts a dual-branch network structure to simultaneously perform wear detection on the wiper blades and contamination detection on the glass.

[0021] According to another aspect of the present invention, one embodiment of the present invention provides a vehicle including a vehicle body and a controller for performing the steps of the method described above.

[0022] According to another aspect of the present invention, one embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed, implements the method as described above.

[0023] This invention provides a vehicle anomaly detection method, device, and vehicle. The method involves collecting strain voltage signals representing the torque experienced by the vehicle's windshield wipers during wiping motion; performing feature processing on the strain voltage signals to extract multi-dimensional features; and applying a preset anomaly detection model to assess wiper blade wear and glass contamination using these multi-dimensional features, thereby obtaining the vehicle's anomaly detection results. The anomaly detection model employs a dual-branch network structure to simultaneously detect wiper blade wear and glass contamination. By performing multi-scale decomposition on the strain voltage signals, feature vectors reflecting wear, contamination, and impact are extracted, enabling wiper blade wear and glass contamination assessments. This method accurately obtains the vehicle's anomaly detection results and proactively reminds users to replace the wiper before significant performance degradation, achieving early fault diagnosis.

[0024] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, specific embodiments of this disclosure are described below. Attached Figure Description

[0025] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0026] Figure 1 The diagram shown is a flowchart illustrating a vehicle anomaly detection method according to an embodiment of this application.

[0027] Figure 2 The diagram shown is a structural schematic of a vehicle anomaly detection device provided in an embodiment of this application.

[0028] Figure 3 The diagram shown is a structural schematic of a vehicle according to an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Furthermore, in exemplary embodiments, since the same reference numerals denote the same components having the same structure or the same steps of the same method, if one embodiment has been described by way of example, then in other exemplary embodiments only structures or methods different from those described in the embodiment will be described.

[0031] Throughout the specification and claims, when a component is described as being “connected” to another component, that component may be “directly connected” to the other component or “electrically connected” to the other component via a third component. Furthermore, unless explicitly stated otherwise, the term “comprising” and its corresponding terms should be understood only to include the stated component and not to exclude any other component.

[0032] In vehicle-related technologies, there is a need for independent and quantitative assessment of the health status of the wiper blades and the cleanliness of the windshield surface, in order to achieve accurate condition perception and predictive maintenance, and to provide a basis for adaptive cleaning control.

[0033] Monitoring windshield wiper status typically relies on the driver's subjective visual judgment or a replacement strategy based on fixed time intervals, making quantification and automation difficult. To sense wiper load, some solutions attempt to collect current or speed signals at the drive motor. These methods can sense changes in the overall system load torque, but the information they acquire is limited. Since both wiper blade wear and glass contamination cause significant changes in wiper load, and these two factors are coupled and occur simultaneously in actual operation, it is difficult to distinguish from a single total load signal whether the load change originates from progressive wear of the wiper blade rubber or from contamination such as oil film, silt, or hard particles adhering to the glass surface. This fundamental coupling limits the ability of related technologies to independently and quantitatively assess wiper blade health and glass cleanliness, making precise on-demand cleaning control and predictive maintenance based on condition difficult to achieve. This situation means that, since vehicles cannot directly perceive their own mechanical condition and the state of glass contamination, the contribution of wiper blade wear and glass contamination to changes in wiper load is coupled and difficult to decouple. Current technology cannot independently and quantitatively assess wiper blade health and glass cleanliness, hindering accurate vehicle condition perception and predictive maintenance. In other words, traditional vehicle users rely on experience to determine replacement timing, unable to quantify wiper blade wear, resulting in a lack of condition perception and problems with premature replacement (waste) or premature replacement (incomplete wiping, glass damage). Furthermore, it cannot distinguish between different types of contamination such as rainwater, oil film, insects, and mud, leading to low cleaning efficiency, wasted cleaning fluid, and poor environmental adaptability. Additionally, traditional vehicles only detect wiper blade wear when it is severely worn or when the mechanism jams, producing abnormal noises, lacking early warning.

[0034] To address the problems of traditional vehicles lacking state perception, having poor environmental adaptability, and lacking early warning, this application provides a method for detecting vehicle anomalies. Figure 1 The diagram shown is a flowchart illustrating a vehicle anomaly detection method according to an embodiment of this application. Figure 1 As shown, the methods for detecting vehicle anomalies include: Step S11: Collect the strain voltage signal of the torque experienced by the vehicle during the brushing process.

[0035] A vehicle's windshield wiper system typically includes a drive source (e.g., a wiper motor), a linkage mechanism (e.g., a drive lever, a driven lever, a wiper arm shaft, and other transmission components), and a wiper blade that performs the cleaning function. This application's embodiments acquire and analyze strain voltage signals reflecting the magnitude of the load torque experienced by the wiper during its wiping motion. From these signals, characteristic information representing the wear state of the wiper blade itself and the contamination state of the contacted glass surface is decoupled, thereby achieving an independent quantitative assessment of two different but physically coupled state changes.

[0036] When the windshield wipers are working, the friction between the wiper blade and the glass, along with the contaminants on the glass surface, collectively create a resistance torque on the drive mechanism. This resistance torque is transmitted through the wiper arm and linkage mechanism, causing minute mechanical deformations at key stress points of the linkage mechanism that are positively correlated with the load. In this embodiment, a deformation sensing unit can be deployed at selected locations in the linkage mechanism to convert this minute deformation into a strain voltage signal, thereby mapping the dynamic load information of the entire wiper system into a continuous electrical signal. Thus, one to two strain gauges can be arranged at the hinge point between the driving and driven rods and at the wiper arm pivot point of the linkage mechanism, forming redundant measurements. By employing redundant measurements, even if one sensor node fails due to aging or damage, it can still continue to operate based on signals from other nodes, avoiding the loss of the entire condition assessment function due to a single point of failure, significantly improving the vehicle's functional safety level and fault tolerance. The strain gauges are temperature-compensated strain gauges with an operating temperature range covering -40 degrees Celsius to 125 degrees Celsius. All strain voltage signals are synchronously acquired at a preset sampling frequency to ensure time-domain alignment. The preset frequency can be set as needed, preferably 2kHz.

[0037] The strain voltage signal carries rich state information. Specifically, when the scraper is brand new and the glass surface is clean, the coefficient of friction between them is low and stable. For example, if the coefficient of friction μ is approximately between 0.1 and 0.3, the load torque is stable, and the strain voltage signal exhibits a relatively smooth curve. When the scraper wears down, its blade edge is no longer sharp and straight, and the contact pressure is unevenly distributed along the length of the blade, resulting in intermittent sliding-viscous vibrations during scraping. This microscopic "jitter" is reflected in the signal as a high-frequency pulsating component. Different types of contamination modulate the coefficient of friction in drastically different ways, thus affecting the load signal. For example, uniformly distributed oil film contaminants create a boundary lubrication effect, leading to a significant increase in the coefficient of friction, manifested as an overall rise in the signal baseline; while hard particles scattered on the glass surface, such as sand grains and insect remains, cause localized, transient, and severe impacts when the scraper passes over them, reflected as spike pulses in the signal. Thus, from a physical mechanism perspective, the progressive wear of the scraper and the contamination of the glass surface are both coupled and recorded in a single strain voltage signal.

[0038] Step S12: Perform feature processing on the strain voltage signal to extract multidimensional features.

[0039] The strain voltage signal V_strain(t) represents the real-time voltage signal output by the Wheatstone bridge measurement circuit, which measures the resistance change of a micro-electro-mechanical system (MEMS) strain gauge attached to a key node of the wiper linkage due to the slight deformation caused by the force applied to the linkage. In step S12, the original one-dimensional time series signal of the strain voltage signal V_strain(t) is transformed and abstracted into a set of feature quantities that can characterize the intrinsic physical properties of the signal from different dimensions through various signal processing techniques. These dimensions cover the signal's time-domain statistical regularity, frequency-domain energy distribution, and time-frequency domain localization characteristics.

[0040] Step S13: Apply a preset anomaly detection model to evaluate the wear of the wiper blade and the contamination of the glass opposite the wiper blade to obtain the anomaly detection results of the vehicle. The anomaly detection model adopts a dual-branch network structure to simultaneously perform wear detection on the wiper blade and contamination detection on the glass.

[0041] The anomaly detection model can learn and identify which changes in feature patterns are caused by the wear of the wiper blade itself and which are caused by contamination of the glass surface. Although the effects of both on the original signal may be intertwined—for example, a worn wiper blade passing over a hard contaminant will generate an unusually strong impact signal that incorporates both the "irritation" characteristic of the wiper blade wear and the "impact" characteristic of the contaminant on the glass—the anomaly detection model can decouple these two states through joint analysis of multidimensional features, outputting independent, quantitative assessment results, such as an index characterizing wiper blade health and an index characterizing glass cleanliness.

[0042] This application embodiment extracts feature vectors reflecting wear, contamination, and impact by multi-scale decomposition of strain voltage signals; constructs a dual-branch deep learning network based on an attention mechanism to output the wiper blade health index and the glass cleanliness index respectively; dynamically optimizes water spraying and wiping parameters through model predictive control (MPC) based on the state recognition results; the system has end-to-end self-learning capability, can learn the optimal control strategy from each effective cleaning, and adaptively adjust the baseline as the wiper blade ages.

[0043] The vehicle anomaly detection method of this application collects strain voltage signals of the torque magnitude experienced by the vehicle's windshield wipers during wiping motion; performs feature processing on the strain voltage signals to extract multi-dimensional features; and applies a preset anomaly detection model to evaluate the wiper blade wear and the glass contamination relative to the wipers, thereby obtaining the vehicle anomaly detection results. By using the vehicle as a sensing element and collecting its strain voltage signals, and utilizing the model to analyze the multi-dimensional features of the signals, the method achieves joint decoupling and independent quantitative evaluation of the mutually coupled wiper blade wear state and glass contamination state, enabling accurate vehicle condition perception and predictive maintenance.

[0044] To more clearly illustrate the technical solutions provided in the embodiments of this application, the following further describes a method for detecting vehicle anomalies provided in this application.

[0045] To accurately detect vehicle anomalies, it is advisable to extract time-domain and / or frequency-domain features from the acquired strain voltage signal to facilitate comprehensive analysis. Based on this, in some embodiments of this application, optionally, the feature processing of the strain voltage signal to extract multi-dimensional features includes: performing time-domain processing on the strain voltage signal to extract time-domain statistical features; performing a Fast Fourier Transform on the strain voltage signal and extracting spectral features; performing multi-level wavelet packet decomposition on the strain voltage signal to obtain multiple sub-frequency bands, and extracting time-frequency domain wavelet packet features based on the multiple sub-frequency bands; and combining the time-domain statistical features, the spectral features, and the time-frequency domain wavelet packet features to obtain multi-dimensional features.

[0046] Each wiper action can be used as an analysis window to record the complete load curve. A single wiper action can be a one-way wipe (moving from one side of the windshield to the other) or a reciprocating wiper action. The strain voltage signal can be processed in the time domain to extract time-domain statistical features, such as mean, variance, and peak value. These time-domain statistical features reflect the overall load level and fluctuations. The strain voltage signal can also be subjected to a Fast Fourier Transform (FFT), and then the spectrum within a preset frequency band can be extracted. Spectral features are extracted based on the spectrum within the selected preset frequency band. These spectral features reflect periodic wear and vibration. Furthermore, the strain voltage signal can be decomposed into multiple sub-bands using multi-level wavelet packet decomposition. For example, a 4-level wavelet packet decomposition of the strain voltage signal yields 16 sub-bands. When performing multi-level wavelet packet decomposition on the strain voltage signal, a binary tree decomposition can be used, where the wavelet packet decomposition resembles a continuously branching binary tree. Layer 0: The root, i.e., the original full-band strain voltage signal V_strain(t), assuming its highest analysis frequency is Fs / 2, where Fs is the previous sampling frequency, for example, a 2kHz sampling frequency corresponds to an analysis frequency of 0-1000Hz. Layer 1: The Layer 0 signal is decomposed into two sub-bands by passing it through a set of high-pass and low-pass filters: Node (1,0): Low-frequency sub-band (approximation coefficients). Node (1,1): High-frequency sub-band (detail coefficients). Layer 2: Each sub-band of Layer 1 is decomposed again using the same high-pass and low-pass filtering, resulting in four sub-bands: Nodes (2,0), (2,1), (2,2), (2,3). Layer 3: Further decomposition yields eight sub-bands: Nodes (3,0) to (3,7). Layer 4: Final decomposition yields sixteen sub-bands: Nodes (4,0) to (4,15). Each layer of decomposition divides the frequency band of the previous layer in two, thus increasing the frequency band resolution (i.e., narrowing the bandwidth of each sub-band), but correspondingly reducing the time resolution. Then, time-frequency domain wavelet packet features are extracted based on these multiple sub-bands. Finally, the time-domain statistical features, the spectral features, and the time-frequency domain wavelet packet features are combined to obtain multi-dimensional features. This embodiment of the application, through high-frequency and high-precision analysis of the time-domain, frequency-domain, and time-frequency domain features of the strain voltage signal V_strain(t), can easily separate the load changes from two different sources—blade wear and glass contamination—from a single strain signal, achieving independent quantitative evaluation.

[0047] Considering the need to effectively identify the characteristic parameters of hard stain impact and signal fluctuation intensity to distinguish different types of contamination and wear, relevant features can be extracted in the time domain, frequency domain, and time-frequency domain respectively. Based on this, in some embodiments of this application, optionally, the time-domain statistical features include: mean μ_V, standard deviation σ_V, skewness γ1, kurtosis γ2, peak-to-peak value V_pp, root mean square V_rms, impact factor IF = V_peak / V_rms, number of impacts exceeding the threshold V_th N_impact, and average impact energy. The threshold V_th can be set as needed and is not specifically limited here. The spectral features include: multiple spectral peak frequencies and their corresponding amplitudes, spectral centroid, frequency variance, and the energy proportion of multiple frequency bands. For example, the spectral features include: the first 5 significant spectral peak frequencies f_peak1-5 and their amplitudes, spectral centroid FC, frequency variance FV, the energy proportion of the 2-10Hz frequency band used to characterize the elastic vibration of the scraper, and the energy proportion of the 10-50Hz frequency band used to characterize uneven wear of the rubber strip. Wherein, the spectral centroid FC = Σ(f_i A_i) / ΣA_i, where f_i corresponds to a specific vibration frequency generated by the wiper system during operation. For example, localized wear on the rubber strip may cause periodic vibrations of 5Hz, while clearance in the linkage mechanism may induce impact vibrations of 20Hz. A_i reflects the severity of this specific vibration mode. For instance, the more severe the wear, the larger the amplitude A_i of its corresponding characteristic frequency is typically. The formula FC = Σ(f_i) A_i) / ΣA_i represents the "average frequency" for calculating the total vibration energy, reflecting whether the overall load vibration is too high or too low. Frequency variance FV = Σ[(f_i - FC)² [A_i] / ΣA_i. Frequency variance FV describes the degree of concentration or dispersion of vibrational energy at frequency, and is used to distinguish different types of wear or contamination modes.

[0048] The time-frequency domain wavelet packet features include the energy E_j and energy entropy H_we of each sub-band. After multi-layer wavelet packet decomposition of the strain voltage signal, different layers (sub-bands) capture vibration features from different physical sources. Vibrations caused by uneven wear of the rubber strip or edge breakage are neither a gradual low-frequency trend, as described in layers 1-2, nor purely random noise (higher layers). They manifest as sudden, transient vibrations with specific frequency characteristics. The sub-bands of layers 3-4 have moderate bandwidth, precisely "focusing" on these characteristic frequency ranges (e.g., tens to hundreds of hertz). When wear occurs, the vibration energy (E_j) in these specific frequency ranges significantly increases, leading to a larger proportion (p_j) of the total energy. Therefore, by observing which specific sub-bands in layers 3-4 suddenly increase in energy E_j, the wear characteristics can be accurately located, enabling early and accurate fault diagnosis. In short, layers 3-4 are the key microscopes for capturing "wear fingerprints." Energy entropy H_we = -Σ(p_j ln p_j), where p_j = E_j / ΣE. The above time-domain statistical features, spectral features, and time-frequency wavelet packet features are combined to form a 65-dimensional feature set.

[0049] The embodiments of this application, by introducing the spectral centroid and significant spectral peak frequencies, can effectively describe the frequency distribution of vibration energy and the dominant vibration mode, which can be used to trace the wear source; by using the sub-band energy and energy entropy obtained by wavelet packet decomposition, the localized information of transient and non-stationary vibration events caused by factors such as uneven wear of rubber strips can be accurately captured, thereby realizing an earlier and more detailed diagnosis of the wear state.

[0050] Considering that vehicle anomalies may include wiper blade wear or glass contamination, anomaly detection requires taking into account both of these different physical characteristics. Therefore, in some embodiments of this application, optionally, the application of a preset anomaly detection model to evaluate the wiper blade wear and the glass contamination relative to the wiper blade using the multi-dimensional features to obtain the vehicle anomaly detection results includes: applying a wear evaluation branch network in the anomaly detection model to process the multi-dimensional features and the strain voltage signal to obtain wiper blade wear characteristics; applying a contamination evaluation branch network in the anomaly detection model to process the multi-dimensional features and the strain voltage signal to obtain glass contamination characteristics; and applying a feature cross-attention module in the anomaly detection model to share information between the wear characteristics and the contamination characteristics to obtain the vehicle's wiper blade wear detection results and glass contamination detection results.

[0051] The anomaly detection model is a deep learning model comprising two parallel branches: wiper blade wear detection and glass contamination feature detection. Specifically, a dual-branch network structure can be used, focusing on the wiper blade wear characteristics and the glass contamination characteristics, respectively. Wear features characterize the wiper blade wear, while contamination features characterize the glass contamination. The input to the anomaly detection model is the previously extracted 65-dimensional features and the strain voltage signal. The strain voltage signal can be obtained based on a preset number of sampling points. The preset number can be set as needed, preferably 500. These 500 sampling points are located in the core working interval where the wiper blade is in full contact with the glass and the load is stable. The specific determination method for each sampling point is as follows: determine the time period during which the wiper blade moves to the middle region of the glass; within the time period, find a high-load interval where the strain voltage signal amplitude is stable and continuous; extract a preset number of sampling points from the high-load interval. Within this high-load interval, the strain voltage signal V_strain(t) amplitude is stable and significantly higher than the return no-load baseline. Specifically, the time period during which the wiper blade moves to the middle region of the glass is determined based on the motor Hall position signal. For example, consider the time interval from the left A-pillar to the middle 50% of the travel distance from the right A-pillar. Within this defined time interval, find a continuous range where the amplitude of the strain voltage signal V_strain(t) is stable and significantly higher than the return no-load baseline. The return no-load baseline is the steady-state mean baseline Vbase of the strain voltage when the equipment is returning without load and without workpiece contact. A strain voltage signal whose amplitude is significantly higher than the baseline can be selected based on a preset condition: |Vstrain(t)−Vbase|≥ΔVth. The deviation threshold ΔVth is higher than the peak-to-peak noise level of the return no-load baseline, effectively eliminating random jitter. Amplitude stability means that the signal does not experience large jumps or trend drift within the interval, and the fluctuation range is within the allowable steady-state bandwidth. Continuity refers to an uninterrupted continuous time interval that meets the above preset condition, without falling back to within the deviation threshold ΔVth in the middle. Then, continuously extract 500 sampling points from this stable high-load interval (corresponding to 0.25 seconds at a 2kHz sampling rate). The strain voltage signal obtained from 500 sampling points contains the most representative and least disturbed load information in this scraping action.

[0052] The anomaly detection model includes a wear assessment branch network, a contamination assessment branch network, and a feature cross-attention module. The wear assessment branch network comprises a one-dimensional (1D) convolutional neural network (CNN) and a long short-term memory (LSTM) neural network. The 1D CNN is used to extract local wear texture features, such as periodic pulses caused by scratches. The LSTM is used to capture the load evolution pattern throughout the entire brush stroke. The wear assessment branch network is applied to process the multi-dimensional features and the strain voltage signal to obtain the wear characteristics of the blade. The CNN and LSTM are cascaded within the wear branch network, working together to extract the wear features. The wear features output by the wear assessment branch network include the blade health index H_wear and wear pattern classification. The blade health index H_wear ranges from 0 to 100, where 0 indicates old and needs replacement, and 100 indicates brand new. Wear patterns are categorized, including but not limited to: normal uniform wear, unilateral wear / uneven wear, edge defects / localized peeling, hardening / aging of the wiper blade rubber strip, and frame interference / wiper blade rubber strip curling. Normal uniform wear refers to a stable load signal, uniform energy distribution in the frequency domain, and no abnormal energy concentration in the time and frequency domains. If the rubber strip ages normally, it indicates uniform edge wear, which is the expected progressive wear. The vehicle's wiper system can operate in standard or eco mode until the wiper blade health index drops to a threshold. Unilateral wear / uneven wear refers to a significantly higher load on the wiper blade in one wiping direction (e.g., from left to right) than in the other, with directional differences in time-domain statistics. This may be due to unbalanced wiper arm pressure, uneven glass curvature, or improper installation angle, causing excessive force on one side of the wiper blade rubber strip. Users should be advised to check the installation; in severe cases, use should be restricted or adjustments recommended. Edge defects / localized peeling refers to periodic or non-periodic peaks in the time domain, spectral peaks at specific frequencies in the frequency domain, and energy concentration in the corresponding sub-bands in the time and frequency domains. The scraper blade may be scratched by gravel, partially torn, or have small pieces peeling off due to uneven material, causing jumping and abnormal noise during scraping. In this case, the "protection and suppression mode" can be triggered to increase lubrication and prevent damage to the glass. Scraper blade hardening / aging refers to an overall increase in load baseline and signal stiffness (increased high-frequency components), but the impact characteristics are not obvious. This may be due to abnormal changes in properties with temperature variations, causing the rubber to lose elasticity due to ozone, ultraviolet rays, and high temperatures, becoming hard and brittle, resulting in poor adhesion to the glass, especially at low temperatures. In this case, the water spray volume needs to be increased to compensate. Frame interference / scraper blade curling refers to severe distortion of the load signal, possibly accompanied by periodic large fluctuations or abnormal motor current. The scraper blade may become loose or curled from the frame, or the metal frame may begin to contact the glass after wear to its limit. In this case, a "serious" warning can be triggered, and use should be stopped immediately to prevent scratching the glass.

[0053] The contamination assessment branch network can employ a Transformer encoder, which includes an embedding layer and a multi-head self-attention layer. The embedding layer transforms the signal segments into sequences, while the multi-head self-attention layer captures global dependencies. The contamination assessment branch network is applied to process the multidimensional features and the strain voltage signal to obtain the contamination characteristics of the glass. Optionally, the contamination assessment branch network embeds the strain voltage signal segments into multidimensional features, utilizing a self-attention mechanism to capture global dependencies, particularly adept at distinguishing between oil films (global load increase) and dispersed stains (local impact). Signal segmentation refers to dividing the strain voltage signal into fixed-length time windows. For example, 500 sampling points are divided into 10 non-overlapping sub-segments, each containing 50 consecutive sampling points (corresponding to a 25-millisecond signal duration). During embedding, each sub-segment (e.g., 50 points) is mapped to a fixed-dimensional vector (i.e., a "token" or "embedding vector") through a learnable linear projection layer or a one-dimensional convolutional layer. This vector represents the local features of the strain voltage signal within that short time period. The embedding vectors of all segments are then arranged in chronological order, forming a token sequence of length 10. This token sequence serves as the direct input to the Transformer encoder. The output of the contamination assessment branch network includes: the glass cleanliness index C_dirt and the contamination type probability distribution. The glass cleanliness index C_dirt ranges from 0 to 100, where 0 represents completely clean and 100 represents completely clean.

[0054] Contamination types include, but are not limited to: oil films / hydrophobic films, mud spots / slurry, insect remains / resin, hard particles (sand, ash), water films (pure rainwater), and mixed contamination. Oil films / hydrophobic films refer to oil stains or hydrophobic coatings uniformly adhered to the glass surface, causing a smooth, overall increase in the load baseline with minimal signal fluctuations and concentrated frequency domain energy at extremely low frequencies. Mud spots / slurry refer to discrete or patchy stains formed after the splashing and drying of muddy water containing particles, causing an overall increase in load and superimposed with numerous low-to-medium frequency, wide pulses, resulting in "sawtooth" fluctuations in the time domain. Insect remains / resin refer to locally adhered, sticky organic contaminants, causing a "step-like" increase in load (during passage) that may be sustained, accompanied by mid-to-high frequency vibrations. Hard particles (sand, ash) refer to scattered dust, sand particles, brake pad dust, etc., resulting in a strain voltage signal filled with random, transient, extremely high-amplitude spikes (micro-impacts) and a wide frequency domain. Water film (pure rainwater) refers to clean rainwater forming a uniform film on the surface. At this point, the load is stable and low, the signal is "clean," and it represents the system's baseline cleanliness state. Mixed pollution refers to a combination of the above types, exhibiting a complex set of characteristics. The output of the pollution assessment branch network corresponds to the probability distribution of each category. Mixed pollution is the most common type of pollution.

[0055] After acquiring the wear characteristics of the wiper blade and the contamination characteristics of the glass, a feature cross-attention module is applied to share information between the wear and contamination characteristics to obtain the wiper blade wear detection results and glass contamination detection results for the vehicle. The wear branch and the contamination branch are two "collaborative experts" that engage in real-time dialogue and information sharing through the feature cross-attention module, thereby achieving accurate and joint judgment of the two coupled states of wiper blade wear and glass contamination. For example, when hard contamination is identified, it can help the wear branch distinguish whether the impact spike originates from the stain rather than a defect in the wiper blade itself.

[0056] This application embodiment separates the load changes from two different sources—blade wear and glass contamination—from a single strain signal to achieve independent quantitative evaluation. It also performs dual-state joint decoupling and joint judgment on the separately acquired wear characteristics and contamination characteristics, thereby accurately detecting vehicle anomalies.

[0057] Considering that some wear features may be caused by glass contamination, while some contamination features may be caused by wiper blade wear, in some embodiments of this application, optionally, the application of the feature cross-attention module in the anomaly detection model to share information between the wear features and the contamination features to obtain the wiper blade wear detection results and glass contamination detection results of the vehicle includes: using the wear feature as a query, retrieving a first relevant information from the contamination features, and fusing the first relevant information with the wear feature to obtain the wiper blade wear detection result of the vehicle, wherein the first relevant information is a context vector generated from feature information associated with the wear feature in the contamination feature; using the contamination feature as a query, retrieving a second relevant information from the wear feature, and fusing the second relevant information with the contamination feature to obtain the glass contamination detection result of the vehicle, wherein the second relevant information is a context vector generated from feature information associated with the contamination feature in the wear feature.

[0058] When fusing wear and contamination features, the attention weights calculated in the previous step are weighted and fused with each other's information to generate a "context vector." This context vector is then added to or concatenated with the original features of each feature to update their respective feature representations. The attention weights calculated in the previous step can refer to the attention weights used when fusing the previously acquired wear and contamination features. To determine which contamination features are more relevant based on the current wear state, the wear features can be used as a query to retrieve first relevant information (Key, Value) from the contamination features. This first relevant information is then fused with the wear features to obtain the vehicle's scraper wear detection result.

[0059] To determine which parts of the wear characteristics are more critical based on the current contamination level, the contamination characteristics can be used as a query to retrieve second relevant information (key, value) from the wear characteristics. This second relevant information is then fused with the contamination characteristics to obtain the vehicle's glass contamination detection result. This embodiment of the application achieves accurate acquisition of vehicle wiper wear detection results and glass contamination detection results by performing dual-state joint decoupling of wear characteristics and contamination characteristics, facilitating vehicle adjustments.

[0060] After accurately obtaining the anomaly detection results of the vehicle, relevant processing can be performed on the vehicle based on the anomaly detection results. Therefore, in this embodiment, the method may optionally further include: adjusting the control parameters of the vehicle based on the anomaly detection results; and / or, predicting the remaining lifespan of the vehicle based on the anomaly detection results, obtaining the remaining lifespan of the vehicle, and issuing graded warnings to the vehicle based on the anomaly detection results or the remaining lifespan.

[0061] The Model Predictive Control (MPC) method can be applied to obtain the predicted cleanliness C_dirt_pred of the glass based on the anomaly detection results. An objective function can be constructed based on the predicted cleanliness C_dirt_pred, the wiper blade health index, and the glass cleanliness index. The control parameters of the vehicle can then be adjusted based on this objective function.

[0062] This application embodiment can also construct a wiping blade health index decay model based on the acquired anomaly detection results; and predict the remaining lifespan of the wiping blade health index when it drops to a threshold index according to the decay model. Specifically, a decay model of the wiping blade health index H_wear over time or wiping frequency is established based on the anomaly detection results. This decay model is expressed as dH / dt = -k·f(usage intensity, contamination type), where k is the decay coefficient, and f is a function that is related to both usage intensity and contamination type, usually designed as the product of the two: f(usage intensity, contamination type) = g(usage intensity) × h(contamination type), used to quantify the relative wear rate under actual working conditions, with a value of 1 under the baseline working condition. The baseline working condition refers to the normal condition of the vehicle, such as standard wiping intensity and clean rainwater. g(usage intensity) is calculated comprehensively from wiping frequency, speed, pressure, load torque, etc. For example, based on the average wiping conditions of the vehicle (such as moderate rain, intermittent mode), let g=1; for high-speed continuous wiping, dry wiping, or high load, g>1 (e.g., 1.2~1.5); for low speed and good lubrication, g<1 (e.g., 0.8); it is usually estimated in real time from data such as motor current and wiping time. h (contamination type) is calculated based on the contamination type identification results (probability distribution). Each contamination type corresponds to an experimentally calibrated wear acceleration factor. For example, when the contamination type is clean rainwater / water film, h=1.0 (baseline); when the contamination type is oil film / hydrophobic film, h=1.2 (poor lubrication, high frictional heat); when the contamination type is mud spots / mud, h=1.5 (containing fine abrasive particles); when the contamination type is hard particles (sand, ash), h=2.5~3.0 (severe abrasive wear); when the contamination type is insect remains / resin, h=1.8 (adhesive wear). The actual h value is a weighted average of the probabilities of each type. The remaining days or number of wipes required for the wiper blade health index H_wear to drop to the health threshold H_th, based on the constructed attenuation model, are predicted to obtain the remaining lifespan of the vehicle's wiper blades. The health threshold H_th can be set as needed, preferably 20. When predicting the remaining lifespan, seasonal factors also need to be considered; for example, high summer temperatures accelerate the aging of the wiper blade rubber strips, requiring an increase in the attenuation coefficient of the attenuation model.

[0063] Furthermore, vehicle warnings can be categorized and graded based on anomaly detection results or remaining lifespan. Vehicles can be classified into multiple levels, from healthy to severely abnormal, based on anomaly detection results or remaining lifespan. The specific number of levels can be set as needed and is not limited here. Different warnings are issued for different levels.

[0064] This application embodiment can adjust the vehicle's control parameters based on the anomaly detection results to achieve the best control effect. It can also predict the remaining lifespan and then combine the anomaly detection results to provide graded warnings for the vehicle, so that the driver can understand the actual situation of the vehicle in a timely manner and take corresponding countermeasures.

[0065] Considering that multiple control parameters may be involved when a vehicle malfunctions, but not all of them need to be adjusted, in some embodiments of this application, optionally, adjusting the vehicle's control parameters based on the malfunction detection result includes: obtaining candidate control parameters for the vehicle based on the malfunction detection result; applying a model predictive control method to obtain a predicted cleanliness of the glass based on the malfunction detection result and the candidate control parameters; determining an objective function based on the predicted cleanliness and the malfunction detection result; and selecting the vehicle's control parameters that minimize the objective function from the candidate control parameters.

[0066] Based on the anomaly detection results, the wiper blade wear detection results and glass contamination detection results can be determined. Then, based on these results, all possible control parameters for reducing wiper blade wear and / or glass contamination can be identified, resulting in candidate control parameters. After determining the candidate control parameters, model predictive control methods can be applied to obtain the predicted cleanliness of the glass. Specifically, the glass cleanliness C_dirt_pred can be obtained based on the current state (H_wear, C_dirt) and the candidate control parameters. Then, an objective function is determined based on the predicted cleanliness and the anomaly detection results, and the control parameters that minimize the objective function are selected from the candidate control parameters. The objective function is (C_dirt_target - C_dirt_pred)² + α·wear rate. For example, the control parameters selected from the candidate control parameters to minimize the objective function include optimal spray duration, pressure setpoint, speed curve, etc. Here, the target cleanliness index C_dirt_target is the glass cleanliness index when no measures are needed to address glass contamination, the wear rate characterizes the speed of wiper blade wear and can be obtained from the wiper blade wear detection results, and α is the weight of the wear rate.

[0067] This application embodiment determines control parameters from candidate control parameters based on anomaly detection results by applying a model predictive control method. It can dynamically optimize the water spray strategy, brush speed and pressure according to the real-time identified working conditions, so as to achieve on-demand cleaning and full-condition self-adaptation.

[0068] Considering different anomaly detection results or different remaining lifespans, different levels of early warning are needed so that users can take relevant measures. Based on this, in some embodiments of this application, optionally, the anomaly detection result includes a wiper blade health index; the step of classifying and issuing early warnings to the vehicle based on the anomaly detection result or the remaining lifespan includes: if the anomaly detection result determines that the frame used to install the wiper blade has a risk of contacting the glass, then outputting a first prompt message, which suggests stopping the use of the current wiper blade; if the wiper blade health index is less than a first health threshold, or if unilateral wear of the wiper blade is detected, then outputting a second prompt message each time the wiper is used, which indicates that the current wiper blade is severely worn; if the wiper blade health index is less than a second health threshold and greater than or equal to the first health threshold, or the remaining lifespan is less than the first lifespan, then outputting a third prompt message when the wiper is activated, which suggests replacing the current wiper blade; if the wiper blade health index is less than the third health threshold and greater than or equal to the second health threshold, or the remaining lifespan is less than the second lifespan and the second lifespan is greater than the first lifespan, then outputting a fourth prompt message, which indicates that the current wiper blade is performing well.

[0069] The first, second, and third health thresholds can be set as needed. Preferably, the first health threshold is 20, the second health threshold is 40, and the third health threshold is 60. The first and second lifespans can also be set as needed. Preferably, the first lifespan is 30 days and the second lifespan is 90 days.

[0070] Table 1. Tiered Early Warning The windshield wiper consists of a metal arm, a frame, and a rubber strip that serves as the wiper blade, which is mounted on the frame. Referring to Table 1, if a risk of the frame used to mount the wiper blade contacting the glass is detected, it indicates a serious vehicle health problem. A strong warning can be issued, and a first alert message can be output to recommend immediately stopping the use of the current wiper blade. The frequency of wiper use can also be automatically limited. If the wiper blade health index is less than 20 or unilateral wear is detected, the wipers are at a warning level. A second alert message can be output each time the wipers are used, indicating "The wiper blade is severely worn," and a one-click appointment for on-site replacement can be made. If the wiper blade health index is less than 40 or the remaining lifespan is less than 30 days, the wipers are at a recommended level. A third alert message can be output when the wipers are started, indicating "It is recommended to replace the wiper blade soon," and nearby service providers and promotional information can be pushed to the system. If the wiper blade health index is less than 60 or the remaining lifespan is less than 90 days, the vehicle is at a caution level. A fourth alert message can be output once by the vehicle's infotainment system, indicating "The wiper blade performance is good, continuously monitored," and this can also be recorded in the maintenance reminder list.

[0071] This application embodiment provides multiple levels of early warning for vehicles based on abnormal detection results and remaining lifespan, proactively reminding drivers to replace the vehicle before performance significantly deteriorates, thus enabling early fault diagnosis.

[0072] Different wiper blade wear and glass contamination levels can yield different candidate control parameters for appropriate wiper control. Based on this, in some embodiments of this application, optionally, the anomaly detection result includes a wiper blade health index, a glass cleanliness index, and a contamination type. Obtaining candidate control parameters for the vehicle based on the anomaly detection result includes: if the wiper blade health index is greater than a fourth health threshold, or the glass cleanliness index is less than or equal to a first cleanliness threshold, then determining the candidate control parameters for the vehicle includes at least one of: reducing wiper blade wiping pressure, delaying water spray time, and reducing wiper blade return speed; if the wiper blade health index is less than or equal to the fourth health threshold and greater than a fifth health threshold, or the glass cleanliness index is greater than a second cleanliness threshold and less than or equal to a third... If the cleaning threshold is determined, the candidate control parameters for the vehicle include at least one of the following: increasing the water spray volume, adding cleaning agent, increasing the swiping pressure of the wiper blade, and increasing the number of swiping strokes. If the glass cleanliness index is greater than the third cleaning threshold and the contamination type is spot stains, the candidate control parameters for the vehicle include: targeted spraying and / or multiple short-stroke swiping strokes based on the location of the spot stains, wherein the short-stroke swiping strokes refer to swiping strokes that include the location of the spot stains. If the wiper blade health index is less than or equal to the sixth health threshold, the candidate control parameters for the vehicle include: water lubrication during each swiping stroke and / or limiting the maximum swiping speed.

[0073] Based on the anomaly detection results, candidate control parameters for real-time vehicle optimization are determined. A fourth health threshold, a fifth health threshold, a sixth health threshold, a first cleaning threshold, a second cleaning threshold, and a third cleaning threshold can be set, with the fourth, fifth, and sixth health thresholds decreasing sequentially, and the first, second, and third cleaning thresholds increasing sequentially. The specific values ​​of each threshold can be set as needed and are not specifically limited here. Preferably, the fourth health threshold is 80, the fifth health threshold is 40, the sixth health threshold is 30, the first cleaning threshold is 20, the second cleaning threshold is 60, and the third cleaning threshold is 80. In other embodiments of this application, the fifth and sixth health thresholds can be set to the same value, such as 40 or 30.

[0074] Table 2 Candidate Control Parameters For example, referring to Table 2, if the blade health index H_wear is greater than the fourth health threshold, or the cleanliness index C_dirt is less than or equal to the first cleaning threshold, the control optimization strategy can adopt an energy-saving and quiet mode. Candidate control parameters include: reducing brush pressure, delaying the water spray preset time (utilizing rainwater), or reducing the return brush speed by 15% to extend blade life and reduce noise and power consumption. The preset time can be set as needed. If the blade health index H_wear is greater than the fifth health threshold and less than or equal to the fourth health threshold, or the cleanliness index C_dirt is greater than the second cleaning threshold and less than or equal to the third cleaning threshold, the control optimization strategy can adopt a deep cleaning mode. Candidate control parameters include: increasing the pre-spray water volume, increasing brush pressure, adding an extra brush stroke, or adding cleaning agent if it is an oil film. If the cleanliness index C_dirt is greater than the third cleaning threshold and it is a point stain, the control optimization strategy is a spot cleaning mode. Candidate control parameters include: triggering spot spraying at the corresponding stain location, or performing three short-stroke brush strokes in that area to efficiently remove stubborn stains and save cleaning fluid.

[0075] If the wiper blade health index H_wear is less than or equal to the sixth health threshold, it indicates that the wiper blade is severely worn. The control optimization strategy is the protection and suppression mode. Candidate control parameters include: prohibiting dry wiping, spraying water for lubrication with each wiping stroke, or limiting the maximum wiping speed to prevent damage to the glass and extend the wiper blade's service life.

[0076] This application embodiment obtains different candidate control parameters based on different wiper blade health indices, glass cleanliness indices, and pollution types, so that subsequent drivers can minimize wiper blade wear and keep the glass as clean as possible when operating the wipers.

[0077] This application transforms the vehicle's mechanical linkage structure into a highly sensitive "tactile sensor." Based on the strain-friction coupling effect, microelectromechanical system (MEMS) strain gauges are attached to key hinge points (such as ball joints and pivots) or the surface of the linkage body to form a Wheatstone bridge. When the wiper is running, the frictional torque between the wiper blade and the glass is transmitted through the linkage, causing micron-level deformation at the stress point. The resistance of the strain gauge changes accordingly, outputting a strain voltage signal V_strain(t) that is positively correlated with the load.

[0078] If the blade edge of the worn scraper is uneven, the contact pressure distribution will be uneven, resulting in intermittent sliding-viscous vibration and high-frequency pulsation of the load. If the oil film is contaminated, boundary lubrication will occur, and the friction coefficient μ between the scraper and the glass will increase significantly, manifesting as an overall rise in the load baseline. If the contaminant is hard dirt, sand particles, insect remains, etc., it will cause micro-impacts, resulting in transient spikes in the load signal. By analyzing the time domain, frequency domain, and time-frequency domain characteristics of the strain voltage signal V_strain(t) with high frequency and high precision, a digital twin model of scraper health and glass cleanliness can be constructed.

[0079] In this embodiment of the invention, after the user confirms the replacement of the wiper blades via the vehicle's infotainment system, a 3-day learning period is initiated to establish new baseline characteristics. Then, the system continuously learns signal characteristics under different weather and temperatures, automatically adjusting the recognition threshold to achieve environmental adaptation. Simultaneously, user habit learning is performed, recording frequently used user settings (such as intermittent frequency) and pre-adjusting control parameters in similar scenarios. The vehicle of this application can also connect with vehicle-to-everything (V2X) applications; for example, it can aggregate data in the cloud, anonymously uploading data such as wiper blade wear rate and common pollution types. On the manufacturer's side, it can analyze regional pollution characteristics (such as oil film in industrial areas or areas with many insects) to optimize cleaning agent formulations; evaluate the lifespan of different wiper blade models to improve product design; provide prior information on glass cleanliness for autonomous driving systems; and perform over-the-air (OTA) upgrades, such as updating the recognition model and control strategy via OTA when new pollution patterns or wear characteristics are detected.

[0080] The vehicle anomaly detection method of this invention separates load changes from two different sources—wiper blade wear and glass contamination—from a single strain signal, performs dual-state joint decoupling, and achieves independent quantitative assessment; it dynamically optimizes the water spray strategy, wiping speed, and pressure based on real-time identified operating conditions (heavy rain, light rain, oil film, mud spots) to achieve full-condition adaptive operation and realize "on-demand cleaning"; it establishes a wiper blade wear rate model and predicts the remaining lifespan based on usage frequency for predictive maintenance, proactively reminding users to replace the wiper before significant performance degradation; and it monitors the wiper mechanism's own condition (such as linkage wear and motor efficiency) for system-level health management, enabling early fault diagnosis.

[0081] Figure 2 The diagram shown is a structural schematic of a vehicle anomaly detection device according to an embodiment of this application. Figure 2 As shown, the anomaly detection device 200 for the vehicle includes: The information acquisition module 201 is used to acquire the strain voltage signal of the torque magnitude of the vehicle's windshield wipers during the wiping motion. Feature extraction module 202 is used to perform feature processing on the strain voltage signal and extract multi-dimensional features; Anomaly detection module 203 is used to apply a preset anomaly detection model to evaluate the wear of the wiper blade and the contamination of the glass opposite the wiper blade on the multi-dimensional features, so as to obtain the anomaly detection results of the vehicle. The anomaly detection model adopts a dual-branch network structure to simultaneously perform wear detection on the wiper blade and contamination detection on the glass.

[0082] In some embodiments, the feature extraction module 202 is used to: perform time-domain processing on the strain voltage signal to extract time-domain statistical features; perform fast Fourier transform on the strain voltage signal and extract spectral features; perform multi-level wavelet packet decomposition on the strain voltage signal to obtain multiple sub-frequency bands, and extract time-frequency domain wavelet packet features based on the multiple sub-frequency bands; and combine the time-domain statistical features, the spectral features, and the time-frequency domain wavelet packet features to obtain multi-dimensional features.

[0083] In some implementations, the anomaly detection module 203 is used to: apply the wear assessment branch network in the anomaly detection model to process the multidimensional features and the strain voltage signal to obtain the wear features of the wiper blade; apply the contamination assessment branch network in the anomaly detection model to process the multidimensional features and the strain voltage signal to obtain the contamination features of the glass; and apply the feature cross-attention module in the anomaly detection model to share information between the wear features and the contamination features to obtain the wiper blade wear detection results and the glass contamination detection results of the vehicle.

[0084] In some implementations, the anomaly detection module 203 is further configured to: use the wear feature as a query to retrieve a first relevant information from the pollution feature, and fuse the first relevant information with the wear feature to obtain the wear condition of the vehicle's wiper blades, wherein the first relevant information is a context vector generated from feature information associated with the wear feature in the pollution feature; use the pollution feature as a query to retrieve a second relevant information from the wear feature, and fuse the second relevant information with the pollution feature to obtain the glass pollution condition of the vehicle, wherein the second relevant information is a context vector generated from feature information associated with the pollution feature in the wear feature.

[0085] In some implementations, the anomaly detection module 203 is further configured to: adjust the control parameters of the vehicle by applying a model predictive control method based on the anomaly detection result; and / or, predict the remaining lifespan of the vehicle based on the anomaly detection result, obtain the remaining lifespan of the vehicle, and issue graded warnings to the vehicle based on the anomaly detection result or the remaining lifespan.

[0086] In some implementations, the anomaly detection module 203 is further configured to: obtain candidate control parameters of the vehicle based on the anomaly detection result; apply a model predictive control method to obtain the predicted cleanliness of the glass based on the anomaly detection result and the candidate control parameters; determine an objective function based on the predicted cleanliness and the anomaly detection result; and select the control parameters of the vehicle that minimize the objective function from the candidate control parameters.

[0087] In some implementations, the anomaly detection result includes a wiper blade health index; the anomaly detection module 203 is further configured to: if, based on the anomaly detection result, it is determined that the frame used to install the wiper blade has a risk of contacting the glass, output a first prompt message, the first prompt message being used to suggest stopping the use of the current wiper blade; if the wiper blade health index is less than a first health threshold, or if unilateral wear of the wiper blade is detected; output a second prompt message each time the wiper is used, the second prompt message being used to indicate that the current wiper blade is severely worn; if the wiper blade health index is less than a second health threshold and greater than or equal to the first health threshold, or the remaining lifespan is less than a first lifespan; output a third prompt message when the wiper blade is activated, the third prompt message being used to suggest replacing the current wiper blade; if the wiper blade health index is less than a third health threshold and greater than or equal to the second health threshold, or the remaining lifespan is less than a second lifespan and the second lifespan is greater than the first lifespan; output a fourth prompt message, the fourth prompt message being used to indicate that the current wiper blade is performing well.

[0088] In some implementations, the anomaly detection results include a wiper blade health index, a glass cleanliness index, and a contamination type; the anomaly detection module 203 is further configured to: if the wiper blade health index is greater than a fourth health threshold, or the glass cleanliness index is less than or equal to a first cleaning threshold, then determine that the candidate control parameters for the vehicle include at least one of: reducing the wiper blade brush pressure, delaying the water spray time, and reducing the wiper blade return speed; if the wiper blade health index is less than or equal to the fourth health threshold and greater than a fifth health threshold, or the glass cleanliness index is greater than a second cleaning threshold and less than or equal to a third cleaning threshold, then determine that the vehicle's... Candidate control parameters include at least one of the following: increasing water spray volume, adding cleaning agent, increasing the swiping pressure of the wiper blade, and increasing the number of swiping strokes; if the glass cleanliness index is greater than the third cleaning threshold and the contamination type is spot stains, then the candidate control parameters for the vehicle are determined to include: point spraying and / or multiple short-stroke swiping strokes based on the location of the spot stains, wherein the short-stroke swiping strokes refer to swiping strokes that include the location of the spot stains; if the wiper blade health index is less than or equal to the sixth health threshold, then the candidate control parameters for the vehicle are determined to include: water lubrication during each swiping stroke and / or limiting the maximum swiping speed.

[0089] Specific limitations regarding the vehicle anomaly detection device can be found in the limitations regarding the vehicle anomaly detection method described above, and will not be repeated here. Each module in the aforementioned vehicle anomaly detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0090] Regarding the apparatus in the above embodiments, the specific manner in which each unit performs its operation has been described in detail in the embodiments concerning the vehicle anomaly detection method, and will not be elaborated upon here.

[0091] Figure 3 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application.

[0092] For example, such as Figure 3 As shown, the vehicle includes a memory 301 and a processor 302. The memory 301 stores executable program code 3011, and the processor 302 is used to call and execute the executable program code 3011 to perform an anomaly detection method for the vehicle.

[0093] This embodiment can divide the vehicle into functional modules according to the above method embodiment. For example, each function can be assigned to a separate module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0094] When each function is divided into its own modules, the vehicle may include: an information acquisition module, a feature extraction module, and an anomaly detection module, etc.

[0095] It should be noted that all relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0096] The vehicle provided in this embodiment is used to execute the above-described vehicle anomaly detection method, and thus can achieve the same effect as the above implementation method.

[0097] When using integrated units, the vehicle may include a processing module and a storage module. The processing module is used to control and manage the vehicle's actions. The storage module supports the vehicle in executing program code and data.

[0098] The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits as disclosed in this application. The processor may also be a combination of computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and microprocessors, etc., and the storage module may be a memory.

[0099] An embodiment of the present invention also provides a vehicle, including a vehicle body and a controller, the controller being used to perform the steps of the method described above.

[0100] This embodiment also provides a computer-readable storage medium (including but not limited to disk storage, CD-ROM, optical storage, etc.) storing computer program code. When the computer program code is run on a computer, the computer executes the above-mentioned related method steps to implement the vehicle anomaly detection method provided in the above embodiment. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, Digital Video Discs (DVDs), Compact Disc Read-Only Memory (CD-ROMs), microdrives, and magneto-optical disks, read-only memory (ROMs), random access memory (RAMs), erasable programmable read-only memory (EPROMs), electrically erasable programmable read-only memory (EEPROMs), dynamic random access memory (DRAMs), video random access memory (VRAMs), flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0101] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to implement the vehicle anomaly detection method provided in the above embodiment.

[0102] The beneficial effects of the above embodiments can be referred to the beneficial effects of the corresponding methods provided above, and will not be repeated here.

[0103] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0104] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0105] In the description of this disclosure, it should be understood that if the terms "upper", "lower", "front", "rear", "left" and "right" are used to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the position or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this disclosure.

[0106] It should be noted that, in the embodiments of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0107] The above are merely embodiments of this disclosure and are not intended to limit the scope of this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of the claims of this disclosure.

Claims

1. A method for detecting anomalies in a vehicle, characterized in that, The method includes: Collect strain voltage signals to measure the magnitude of the torque experienced by the vehicle's windshield wipers during their wiping motion; The strain voltage signal is subjected to feature processing to extract multidimensional features; An anomaly detection model is applied to evaluate the wear of the wiper blades and the contamination of the glass opposite the wiper blades to obtain the anomaly detection results of the vehicle. The anomaly detection model adopts a dual-branch network structure to simultaneously perform wear detection on the wiper blades and contamination detection on the glass.

2. The method according to claim 1, characterized in that, The feature processing of the strain voltage signal to extract multidimensional features includes: The strain voltage signal is processed in the time domain to extract time-domain statistical features; The strain voltage signal is subjected to a fast Fourier transform, and its spectral features are extracted. The strain voltage signal is decomposed into multiple wavelet packets to obtain multiple sub-bands, and time-frequency domain wavelet packet features are extracted based on the multiple sub-bands. By combining the time-domain statistical features, the spectral features, and the time-frequency domain wavelet packet features, multidimensional features are obtained.

3. The method according to claim 1, characterized in that, The application uses a pre-defined anomaly detection model to assess the wiper blade wear and the contamination of the glass opposite the wiper, based on the multi-dimensional features, to obtain anomaly detection results for the vehicle, including: The wear assessment branch network in the anomaly detection model is used to process the multidimensional features and the strain voltage signal to obtain the wear characteristics of the scraper. The pollution assessment branch network in the anomaly detection model is applied to process the multidimensional features and the strain voltage signal to obtain the pollution characteristics of the glass; The feature cross-attention module in the anomaly detection model is used to share information between the wear features and the contamination features to obtain the wiper wear detection results and glass contamination detection results of the vehicle.

4. The method according to claim 3, characterized in that, The application of the feature cross-attention module in the anomaly detection model to share information on the wear features and the contamination features, in order to obtain the vehicle's wiper wear detection results and glass contamination detection results, includes: Using the wear feature as the query, a first relevant information is retrieved from the pollution feature, and the first relevant information is fused with the wear feature to obtain the wiper blade wear detection result of the vehicle. The first relevant information is a context vector generated from the feature information associated with the wear feature in the pollution feature. Using the pollution feature as the query, a second relevant information is retrieved from the wear feature, and the second relevant information is fused with the pollution feature to obtain the glass pollution detection result of the vehicle. The second relevant information is a context vector generated from the feature information associated with the pollution feature in the wear feature.

5. The method according to claim 1, characterized in that, The method further includes: Adjust the vehicle's control parameters based on the anomaly detection results; and / or, The remaining lifespan is predicted based on the anomaly detection results to obtain the remaining lifespan of the vehicle, and the vehicle is given a graded warning based on the anomaly detection results or the remaining lifespan.

6. The method according to claim 5, characterized in that, The step of adjusting the vehicle's control parameters based on the anomaly detection results includes: Candidate control parameters for the vehicle are obtained based on the anomaly detection results; Based on the anomaly detection results and the candidate control parameters, a model predictive control method is applied to obtain the predicted cleanliness of the glass. The objective function is determined based on the predicted cleanliness and the anomaly detection results, and the control parameters of the vehicle that minimize the objective function are selected from the candidate control parameters.

7. The method according to claim 5, characterized in that, The abnormal detection results include the scraper health index; the step of issuing graded warnings to the vehicle based on the abnormal detection results or the remaining lifespan includes: If the abnormality detection results determine that the frame used to install the wiper blade is at risk of contacting the glass, a first prompt message is output, which is used to suggest stopping the use of the current wiper blade. If the wiper blade health index is less than the first health threshold, or if unilateral wear of the wiper blade is detected, a second prompt message is output each time the wiper blade is used. The second prompt message is used to indicate that the current wiper blade is severely worn. If the wiper blade health index is less than the second health threshold and greater than or equal to the first health threshold, or the remaining lifespan is less than the first lifespan, then when the wiper is activated, a third prompt message is output, which is used to suggest replacing the current wiper blade. If the scraper health index is less than the third health threshold and greater than or equal to the second health threshold, or if the remaining lifespan is less than the second lifespan and the second lifespan is greater than the first lifespan, then a fourth prompt message is output, which is used to indicate that the current scraper performance is good.

8. The method according to claim 6, characterized in that, The abnormal detection results include the wiper blade health index, glass cleanliness index, and pollution type; obtaining the candidate control parameters of the vehicle based on the abnormal detection results includes: If the wiper blade health index is greater than the fourth health threshold, or the glass cleanliness index is less than or equal to the first cleaning threshold, then the candidate control parameters for the vehicle are determined to include at least one of the following: reducing the wiper blade wiping pressure, delaying the water spraying time, and reducing the wiper blade return speed. If the wiper blade health index is less than or equal to the fourth health threshold and greater than the fifth health threshold, or the glass cleanliness index is greater than the second cleaning threshold and less than or equal to the third cleaning threshold, then the candidate control parameters for the vehicle are determined to include at least one of the following: increasing the water spray volume, adding cleaning agent, increasing the wiper blade brushing pressure, and increasing the number of brushing strokes. If the glass cleanliness index is greater than the third cleaning threshold and the contamination type is a point stain, then the candidate control parameters for the vehicle are determined to include: point spraying and / or multiple short-stroke wiping based on the location of the point stain, wherein the short-stroke wiping refers to wiping the location of the point stain in the wiping stroke. If the wiper blade health index is less than or equal to the sixth health threshold, then the candidate control parameters for the vehicle are determined to include: water lubrication during each wipe and / or limiting the maximum wipe speed.

9. A vehicle anomaly detection device, characterized in that, The vehicle anomaly detection device includes: The information acquisition module is used to collect strain voltage signals that measure the torque experienced by the vehicle's windshield wipers during their wiping motion. The feature extraction module is used to perform feature processing on the strain voltage signal and extract multidimensional features; An anomaly detection module is used to apply a preset anomaly detection model to evaluate the wear of the wiper blades and the contamination of the glass opposite the wiper blades, in order to obtain the anomaly detection results of the vehicle. The anomaly detection model adopts a dual-branch network structure to simultaneously perform wear detection on the wiper blades and contamination detection on the glass.

10. A vehicle, comprising a vehicle body, characterized in that, It also includes a controller for performing the steps of the method as described in any one of claims 1 to 8.

Images