Windscreen wiper service life prediction and maintenance system and method

Abstract

The invention relates to a windscreen wiper service life prediction and maintenance system and method, and belongs to the technical field of intelligent networked automobiles. The system comprises a sensor network, a processing unit, an abrasion detection module, a service life prediction module, a user reminding module and the like. The system collects current of a wiper motor, working pressure of a wiper blade, visual images after wiping and operation audio in real time. The health index of the wiper blade is output by integrating the data through a multi-mode fusion model based on LSTM; predicting the remaining service life by adopting a GRU-based time sequence prediction model in combination with environment, use habits and adhesive tape material data; when the service life is about to reach or the health index is too low, graded early warning is sent to a user through the vehicle-mounted terminal and the mobile application, and the vehicle speed is adaptively limited to ensure driving safety; the system supports continuous optimization of an algorithm model and system parameters through OTA upgrading. The problem that a traditional windscreen wiper system lacks real-time abrasion monitoring and predictive maintenance capacity is solved, and the driving safety and the system intelligence level are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicle technology, and in particular to a windshield wiper life prediction and maintenance system and method. Background Technology

[0002] Traditional windshield wiper systems lack real-time monitoring and predictive maintenance capabilities for wiper blade wear, leading to sudden wear or failure that directly impacts driving safety. Current technology cannot accurately assess the actual lifespan of wiper blades, making timely replacement difficult and posing a serious safety hazard. Furthermore, the algorithms of traditional systems are fixed and cannot be optimized and updated based on actual usage and environmental changes, lacking continuous improvement capabilities and failing to adapt to different user habits and diverse driving environments.

[0003] The invention patent application with publication number CN120724679A discloses a method for predicting the lifespan of windshield wipers, an aging reminder method and device. The solution has the following shortcomings: (1) It only focuses on the mechanical and material parameters such as the hardness, friction and wear thickness of the wiper blade rubber strip, without integrating multimodal data such as audio and visual cleaning effects. The simulation of the actual working state of the windshield wipers is not comprehensive enough and the data dimension is relatively simple; (2) The solution relies on manually derived mathematical formulas and dynamic calibration logic, which makes it difficult to capture the complex nonlinear relationship in the aging process of windshield wipers. The prediction accuracy is easily affected when facing multi-factor coupling scenarios.

[0004] The invention patent application with publication number CN114347952A discloses a method, device and system for predicting the remaining life of a windshield wiper. The solution has the following shortcomings: (1) Although it involves images, audio, environment and operating characteristics, it uses simple weighting to calculate the remaining life, and does not fully explore the intrinsic relationship between multimodal data. Its data fusion depth is insufficient and the prediction accuracy is still insufficient; (2) It only relies on the similarity comparison of the preset reference sequence and the linear weighting model, and lacks the ability to model the long-term dependence relationship of time series data, making it difficult to adapt to the dynamic process of the gradual aging of windshield wipers.

[0005] Application publication number CN120764066A discloses a method, device, vehicle and program product for detecting the life of a vehicle windshield wiper. The solution has the following shortcomings: (1) It uses an end-to-end deep neural network to predict the remaining life, but it does not establish a physical mapping relationship between the hardness of the rubber strip and the coefficient of friction, and lacks a detailed model of the aging mechanism of the rubber strip material; (2) The acquisition of environmental parameter characteristics of this solution depends on an external meteorological data interface. When the vehicle is in an area with weak network signal or no network coverage, the real-time performance and accuracy of environmental parameter acquisition will be affected; (3) Although this solution proposes multimodal state data fusion, it does not involve a dynamic calibration mechanism for the coefficient of friction based on motor current parameters, and cannot perceive in real time the sudden change in friction characteristics caused by instantaneous load changes such as foreign objects and ice crystals on the glass surface. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a wiper blade life prediction and maintenance system and method, which can monitor the wear status of wiper blades in real time, accurately predict the remaining life and have continuous optimization capabilities.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: a windshield wiper life prediction and maintenance system, including a sensor network, an on-board ECU, a cloud server, a wear detection module, a life prediction module, a user reminder module, and an OTA upgrade module respectively deployed on the on-board ECU; The sensor network includes a current sensor, a pressure sensor, a vision sensor, and an audio sensor, used to collect real-time data on the working current of the wiper motor, the working pressure of the wiper blades on the windshield, the visual image data of the windshield after wiping, and the audio data of the wiper during operation. The wear detection module adopts a multimodal fusion model based on a long short-term memory network. It takes the current feature vector, pressure feature vector, visual cleanliness feature vector and audio feature vector collected and preprocessed by the sensor network as input, and outputs the wiper blade health index HI(t). The life prediction module adopts a time series prediction model based on recurrent neural networks. It inputs the historical health index sequence output by the wear detection module, and integrates environmental factor vectors, usage habit vectors and rubber strip material feature vectors to output the remaining life of the wiper blade RUL(t). The user reminder module monitors the health index HI(t) and remaining service life RUL(t) in real time. When the preset warning conditions are met, it issues a graded warning to the user through the vehicle terminal and / or mobile application. The OTA upgrade module communicates with the vehicle ECU and the cloud server respectively, and is used to receive algorithm model update packages, system parameter configuration packages and firmware update packages sent by the cloud server, so as to realize remote optimization and upgrade of the system.

[0008] A further technical solution of the present invention is that the multimodal fusion model based on long short-term memory network includes multiple independent LSTM encoders and a feature fusion fully connected network. Each LSTM encoder is a multi-layer stacked structure, used to process current feature vectors, pressure feature vectors, visual cleanliness feature vectors, and audio feature vectors respectively. After extracting high-level features, the health index HI(t) is output through nonlinear mapping by the fully connected network. The health index HI(t) ranges from 0 to 1, where 0 represents complete failure and 1 represents a brand new state. The environmental factor vector E includes historical average rainfall intensity, ultraviolet radiation intensity, temperature fluctuation, pollutant type, and rainfall acidity / alkalinity. The usage habit vector U includes average daily usage time, the proportion of high-speed wiping mode, and the aggressiveness of pressure dynamic adjustment. The adhesive strip material feature vector M includes adhesive strip material type, material hardness benchmark value, and material aging sensitivity coefficient.

[0009] A further technical solution of the present invention is that the graded early warning includes: Level 1 alert: When RUL(t) < 30 days or HI(t) < 0.6, a gentle maintenance suggestion text message will be pushed; Level 2 Reminder: When RUL(t) < 7 days or HI(t) < 0.4, a prominent icon warning will be displayed along with a prompt sound, providing information on nearby service outlets and appointment functions; Level 3 Reminder: When RUL(t) = 0 days or HI(t) < 0.25, calculate the environmental risk index Risk_Env, send a speed limit command to the vehicle power system, and simultaneously display a red flashing warning icon, a steering wheel vibration reminder, and provide one-click navigation to service outlets.

[0010] A further technical solution of the present invention is that the current sensor is a Hall effect sensor with a sampling frequency of not less than 1kHz; the pressure sensor is a piezoelectric sensor or indirectly obtains pressure data through a motor torque model; the vision sensor is an in-vehicle camera with a resolution of not less than 1080p; the audio sensor is a directional microphone used to collect audio signals generated by the friction between the wipers and the windshield during operation; the upgrade content supported by the OTA upgrade module includes: wear detection model update, life prediction model update, current threshold adjustment, pressure threshold adjustment, reminder strategy optimization, system firmware repair, and addition of monitoring indicators or early warning functions.

[0011] Another technical solution of the present invention is a method for predicting and maintaining windshield wiper life, employing the aforementioned windshield wiper life prediction and maintenance system, the method comprising the following steps: S1. Data Acquisition: Real-time acquisition of wiper motor operating current data, wiper blade operating pressure data, post-wiping visual image data, and wiper operating audio data through current sensor, pressure sensor, vision sensor, and audio sensor; S2. Data preprocessing and feature extraction: The collected raw data is filtered, time-synchronized and aligned, and image-processed to extract the current feature vector V_I, pressure feature vector V_P, visual cleanliness feature vector V_S, and audio feature vector V_A, respectively. S3. Health status detection: Input the four feature vectors into the LSTM-based multimodal fusion model and output the wiper blade health index HI(t); S4. Remaining life prediction: Input the historical health index sequence, environmental factor vector E, usage habit vector U, and adhesive strip material feature vector M into the GRU-based time series prediction model, and output the remaining life RUL(t); S5. Tiered Early Warning: Based on the monitoring results of HI(t) and RUL(t), trigger the corresponding Level 1, Level 2 or Level 3 early warning, send reminder information to the user and implement corresponding safety control measures; S6. System Upgrade: Receives update packages from the cloud via the OTA upgrade module to remotely update and optimize algorithm models, system parameters, and firmware.

[0012] A further technical solution of the present invention is that step S1 includes: S11. Motor current data acquisition: The working current of the wiper motor is monitored in real time by a current sensor, with a sampling frequency of not less than 1kHz. The current waveform of each wiping cycle is recorded. The wiping cycle includes the start, constant speed and stop phases. S12. Wiper blade pressure data acquisition: The pressure value of the wiper blade on the windshield is obtained in real time through a pressure sensor integrated on the wiper arm or indirectly calculated based on the motor torque model. The sampling frequency is synchronized with the current to ensure data timing alignment. S13. Visual data acquisition of wiping effect: The windshield image is captured by a visual sensor after each wiping, and the image covers the main wiping area; S14. Wiping audio data acquisition: Acquire audio signals generated by the friction between the wipers and the windshield during operation using a directional microphone, with a sampling frequency of not less than 44.1kHz; Step S2 includes: S21. Current signal preprocessing: Apply a low-pass filter to the original current waveform to eliminate high-frequency noise and power supply interference; extract the following features from each scraping cycle: average current I_average during the uniform scraping phase, peak current I_peak during startup and commutation, current fluctuation amplitude I_wave during the uniform scraping phase, and total energy consumption Energy_consumption during this scraping cycle, forming a feature vector V_I. S22. Pressure signal preprocessing: Synchronize and align pressure data with current data using timestamps; extract the following features: the target pressure value P_target of the system command, the actual measured average pressure P_actual_average, and the fluctuation variance P_variance during the pressure control process, to form a feature vector V_P; S23. Visual Preprocessing of Scraping Effect: Image processing algorithms are used to analyze camera images and calculate cleanliness scores. By comparing the pixel differences between the scraped area and the ideal clean area, a comprehensive cleanliness percentage score S_score is obtained. Watermark lines are identified using edge detection algorithms. The ratio of the residual water film or stain area to the total field of view area is calculated to form a feature vector V_S=[S_score, Streak_count, Residue_area_ratio], where S_score is the comprehensive cleanliness percentage score; Streak_count is the number of strip-shaped watermarks or stains identified in the image; and Residue_area_ratio is the proportion of the residual water film or stain area to the total field of view area. S24. Audio signal preprocessing: Apply wavelet threshold denoising algorithm to the original audio signal to eliminate environmental noise, and convert the audio signal to the frequency domain through Fourier transform to extract the following features: abnormal noise frequency band energy ratio A_abnormal_ratio, short-time energy entropy A_entropy, and zero crossing rate A_zero_crossing, forming a feature vector V_A=[A_abnormal_ratio, A_entropy, A_zero_crossing], where the abnormal noise frequency band is 5kHz-7kHz.

[0013] A further technical solution of the present invention is that step S3 includes: inputting the current feature vector V_I, pressure feature vector V_P, visual cleanliness feature vector V_S, and audio feature vector V_A into the LSTM-based multimodal fusion model, and performing health status detection according to the following process: S31: Determine whether the current I(t) is trending upward. If not, return to step S1 and start again; if yes, proceed to step S32. S32: Determine whether the pressure P(t) increases synchronously. If not, proceed to step S36; if yes, proceed to step S33. S33: Determine whether the visual cleanliness S(t) has decreased. If not, proceed to step S35; if yes, proceed to step S34. S34: Strong wear characteristics confirmed, indicating that the windshield wipers are in a state of strong wear, proceed to step S37; S35: The cause is determined to be a minor performance change, possibly due to environmental changes or slight degradation from scratching. S36: It is determined that it may be a single interference that is ignored or the weight is reduced, which is a temporary interference situation; S37: Input the results of steps S34, S35, and S36 into the multimodal fusion model to calculate the health index HI(t); S38: Update the health index HI(t) to quantify the current health status of the windshield wipers; S39: Determine whether HI(t) is less than the threshold. If not, return to step S1; if yes, proceed to step S5. Step S37 includes: S371. Fusion and Inference Process: The feature vectors V_I, V_P, V_S, and V_A are input into the corresponding LSTM encoders to extract high-level feature representations H_I, H_P, H_S, and H_A; H_I, H_P, H_S, and H_A are concatenated to form a 256-dimensional fusion feature vector; a nonlinear mapping is performed through a fully connected network to finally output the health index HI(t); a forward propagation is performed after each scraping, and HI(t) is updated in real time; S372. Offline Training: Train the model on a cloud server using a historical dataset, which includes wiper blade data ranging from brand new to malfunctioning. Optimize the loss function and train for at least 1000 rounds to ensure model convergence. S373. Online Adaptation: When the user performs manual wiping or replaces the wiper blades, the model is fine-tuned using new data, and the weights of the LSTM and fully connected network are updated; the fine-tuning data is uploaded to the cloud via OTA for model optimization.

[0014] A further technical solution of the present invention is that step S4 includes: S41. Data Collection and Preprocessing: S411. Collect historical health index sequence: Collect the historical health index sequence [HI(t), HI(t-1), ..., HI(tn)] output by the wear detection subunit, where n is at least 50 scrapes; S412. Collect environmental data E: Obtain historical average rainfall intensity, UV intensity, temperature fluctuation, pollutant type, and rainfall acidity / alkalinity through vehicle-mounted sensors or external APIs, and normalize them into vectors; S413. Collect usage habit data U: Extract average daily usage time, high-speed wiping mode percentage, pressure dynamic adjustment frequency, etc. from the vehicle CAN bus, and normalize them into a vector; S414. Collect rubber strip material data M: Obtain the material type, material hardness baseline value and material aging sensitivity coefficient of the wiper rubber strip, and normalize them into a vector; S42.RNN prediction model predicts: The GRU network learns sequence dependencies using the past n health index sequences, environmental vector E, usage habit vector U, and adhesive strip material vector M, outputting the hidden state. A fully connected layer maps the hidden state to RUL(t). Mathematically, RUL(t) = NN_Model(HI(t), HI(t-1), ..., HI(tn)|E,U,M), where NN_Model is the trained GRU model. If RUL(t) < 30 days, a user reminder is triggered. S43. Model Update: The initial training of the model is completed in the cloud, using historical failure data; the model parameters are updated regularly via OTA to adapt to new environmental patterns and usage habits, thereby improving prediction accuracy.

[0015] A further technical solution of the present invention is that step S5 includes: S51. Threshold Monitoring Real-time monitoring of RUL(t) and HI(t): If RUL(t) < 30 days or HI(t) < 0.6, trigger a Level 1 alert; If RUL(t) < 7 days or HI(t) < 0.4, a level 2 alert is triggered; If RUL(t) = 0 days or HI(t) < 0.25, a Level 3 alert is triggered, and an environmental risk assessment and speed limit mechanism are activated. S52. Reminder Generation and Sending The first-level reminder is "Maintenance Recommendation," which is displayed as a text message on the in-vehicle display screen and pushed to a mobile application. The second-level alert is "urgently need replacement," displayed as a prominent icon and warning sound on the dashboard; a push notification is sent via the mobile application, embedding a map of nearby service centers and an appointment link; and wear and tear analysis is provided. The Level 3 alert is for "emergency replacement." A prominent flashing red warning icon is displayed on the instrument panel, the current environmental risk index is calculated, the vehicle speed is appropriately limited, and a "Wipe Blade Emergency Replacement - Speed ​​Limited" message is displayed on the central control screen. The warning is amplified through steering wheel vibration and accompanied by an audible alert. The system sends a speed limit command to the vehicle's powertrain via the CAN bus, and the electronic stability system works in conjunction to ensure smooth deceleration. Simultaneously, it automatically recommends the nearest service center and provides one-click navigation, allowing for simultaneous appointment of the replacement service. S53: User Interaction Users confirm the reminder through the in-vehicle interface or mobile app. The system records the feedback data and uploads it to the cloud for subsequent model optimization and reminder strategy adjustment.

[0016] A further technical solution of the present invention is that step S61 includes: S61: Upgrade Detection and Download The vehicle's ECU periodically connects to the cloud server to check for available updates; it downloads update packages via a secure HTTP protocol and verifies digital signatures to ensure security. S62: Upgrade Execution Model update: Replace deep learning models for wear detection or lifespan prediction; Parameter update: Adjust the current threshold, pressure threshold, or alert threshold to adapt to different regional environmental conditions; Firmware update: Update system firmware to fix bugs or add features; Personalized optimization: Push customized algorithm packages based on vehicle usage history; S63: Data Feedback and Optimization The system anonymously collects operational data and uploads it to the cloud via an encrypted channel; the cloud uses this data to retrain the model, enabling continuous optimization and adaptive improvement of the system algorithm.

[0017] By adopting the above technical solution, the wiper life prediction and maintenance system and method of the present invention have the following advantages compared with the prior art: 1. Real-time wear monitoring: This invention includes a sensor network, an on-board ECU, a cloud server, a wear detection module, a life prediction module, a user reminder module, and an OTA upgrade module, all deployed on the on-board ECU. Through multi-sensor data fusion and deep learning algorithms, it can keenly capture early and subtle signs of wear, solving the problem that traditional systems cannot monitor wear status in real time.

[0018] 2. Improved Lifespan Prediction Accuracy: This invention integrates four core data categories: current, pressure, visual cleanliness, and audio. It extracts the temporal features of each modality through an LSTM encoder-fusion architecture, accurately capturing the complex correlations in the aging process of windshield wipers, and providing a more comprehensive quantification of the health index HI(t). It adopts an LSTM+GRU dual-model architecture, which uses LSTM to model the long-term dependencies of multimodal data and GRU to combine environmental, usage habits, and rubber strip material data to predict the remaining lifespan RUL(t). This adapts to the dynamic characteristics of progressive aging and provides higher prediction accuracy compared to traditional methods, offering users an accurate reference for maintenance time.

[0019] 3. Ensuring driving safety: This invention sets up three levels of warning based on HI(t) and RUL(t). In particular, the speed limit function of the three-level warning extends "predictive maintenance" to "active safety protection", effectively avoiding driving risks caused by wiper blade failure.

[0020] 4. Support for continuous optimization: This invention enables remote updates of algorithm models and system parameters through an OTA upgrade module, solving the defects of traditional system algorithm fixation, enabling the system to continuously adapt to different usage scenarios and environmental changes, and improving the long-term reliability and intelligence level of the system.

[0021] 5. Material property modeling integration: This invention introduces the feature vector of rubber strip material to perform personalized modeling for the differences in aging rate of different materials (silicone rubber, fluororubber, etc.), so that the life prediction is more in line with the actual rubber strip characteristics, and realizes universal and accurate prediction for wipers of different brands and materials.

[0022] The technical features of the wiper life prediction and maintenance system and method of the present invention will be further described below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0023] Figure 1 Example 1: A schematic diagram of the overall architecture of the wiper life prediction and maintenance system; Figure 2 Flowchart of the wiper life prediction and maintenance method described in Example 2; Figure 3 Flowchart of step S3 as described in Example 2; Figure 4 : Windshield wiper health index decay curve over time; exist Figure 4 In the diagram, JK represents the healthy status zone, WH represents the zone recommended for maintenance, and GH represents the zone urgently needing replacement. Detailed Implementation Example 1

[0024] A windshield wiper life prediction and maintenance system includes a sensor network, an on-board ECU, a cloud server, a wear detection module, a life prediction module, a user reminder module, and an OTA upgrade module, respectively deployed on the on-board ECU; wherein: The sensor network includes current sensors, pressure sensors, vision sensors, and audio sensors, used to collect real-time data on the wiper motor's operating current, the pressure of the wiper blades on the windshield, visual images of the windshield after wiping, and audio data from the wipers during operation. The current sensor is a Hall effect sensor with a sampling frequency of at least 1kHz, collecting the wiper motor's operating current waveform in real-time, including current data during start-up, constant speed, and stop phases. The pressure sensor is a piezoelectric sensor integrated into the wiper arm, with a sampling frequency synchronized with the current sensor (1kHz), acquiring real-time pressure data from the wiper blades on the windshield; or indirectly calculating the pressure value through a motor torque model to ensure data timing alignment. The vision sensor is a high-resolution in-vehicle camera with a resolution of at least 1080p, installed in a suitable location inside the vehicle to ensure that images of the main wiped area of ​​the windshield are captured after each wipe. The audio sensor is a directional microphone installed near the wipers to collect audio signals generated by the friction between the wipers and the windshield during operation, with a sampling frequency of at least 44.1kHz.

[0025] The onboard ECU is used for real-time data processing and model inference, possessing sufficient computing power to support real-time inference of deep learning models. It integrates a 4G / 5G communication module to achieve bidirectional communication with the cloud server. The cloud server adopts a high-performance computing cluster for model training, optimization, and upgrade package management, possessing large-scale data storage and model training capabilities, and supporting the generation, management, and distribution of OTA upgrade packages.

[0026] The wear detection module is deployed in the vehicle ECU and adopts a multimodal fusion model based on a long short-term memory network. It inputs the current feature vector, pressure feature vector, visual cleanliness feature vector, and audio feature vector collected and preprocessed by the sensor network, and outputs the wiper blade health index HI(t). The wear detection module fuses the aforementioned current, pressure, visual cleanliness, and audio signals and performs comprehensive analysis through the multimodal fusion model based on a long short-term memory network. This multimodal fusion model includes four independent LSTM encoders and a feature fusion fully connected network. Each LSTM encoder has a multi-layer stacked structure. After extracting high-level features, the health index HI(t) is output through nonlinear mapping by the fully connected network. The health index HI(t) ranges from 0 to 1, where 0 represents complete failure and 1 represents a brand-new state. Each LSTM encoder consists of two stacked LSTM layers, each with a hidden state dimension of 64, used to process current feature vectors, pressure feature vectors, visual cleanliness feature vectors, and audio feature vectors respectively. The feature fusion fully connected network includes two hidden layers (containing 64 and 32 neurons respectively) and one output layer. This wear detection module employs a multi-sensor data fusion strategy, comprehensively evaluating wear status by integrating motor current changes, dynamic wiping effect assessment, wiper blade working pressure, and operating audio, significantly improving the accuracy and reliability of diagnosis.

[0027] The lifespan prediction module is deployed in the vehicle's ECU. It employs a time-series prediction model based on a recurrent neural network, inputting the historical health index sequence output by the wear detection module. It then integrates environmental factor vectors, usage habit vectors, and rubber strip material feature vectors to output the remaining lifespan RUL(t) of the wiper blades. The environmental factor vector E includes historical average rainfall intensity, UV radiation intensity, temperature fluctuations, pollutant types, and rainfall pH. The usage habit vector U includes average daily usage time, the proportion of high-speed wiping modes, and the aggressiveness of pressure dynamic adjustment. The rubber strip material feature vector M includes the rubber strip material type, material hardness baseline value, and material aging sensitivity coefficient. This lifespan prediction module considers multiple influencing factors such as usage frequency, environmental conditions, wiping modes, and rubber strip material, resulting in higher prediction accuracy.

[0028] The user alert module is deployed in the vehicle ECU to monitor the health index HI(t) and remaining service life RUL(t) in real time. When the preset warning conditions are met, it issues a graded warning to the user through the vehicle terminal and / or mobile application.

[0029] The OTA upgrade module communicates with both the vehicle ECU and the cloud server to receive algorithm model update packages, system parameter configuration packages, and firmware update packages from the cloud server, enabling remote system optimization and upgrades. The OTA upgrade module supports upgrades including: wear detection model updates, lifespan prediction model updates, current threshold adjustments, pressure threshold adjustments, alert strategy optimization, system firmware repair, and the addition of new monitoring indicators or warning functions.

[0030] In this system, the LSTM network for the wear detection module is chosen based on its inherent advantages in processing time-series data. Its internal gating mechanism includes an input gate, a forget gate, and an output gate. The input gate determines the strength of the current information Xt, the forget gate determines the strength of the past memory Ct-1, and the two together constitute the current memory Ct. The output gate determines the degree of influence of the current memory Ct on the current hidden state Ht. The LSTM network can selectively remember or forget historical information, effectively solving the gradient vanishing or exploding problem in traditional recurrent neural networks (RNNs). This mechanism enables the model to effectively identify long-term patterns spanning dozens of time steps, such as "the current reference value monotonically increases in 50 consecutive wipes," while also sensing short-term events such as "an abnormal current spike at the moment of a single wipe," thus achieving accurate modeling of the wear process. It is particularly suitable for modeling the slowly changing degradation trend during wiper wear.

[0031] The lifespan prediction module employs a GRU network, comprising one input layer, two GRU layers (32 units each), and one fully connected output layer. Its working principle is based on historical health status data output by the wear detection module, combined with data on environment, usage habits, and rubber strip material. A time-series prediction model based on a recurrent neural network (RNN) is used to accurately predict the remaining lifespan of the wiper blades. This time-series prediction model considers multiple influencing factors such as usage frequency, environmental conditions, wiping patterns, and rubber strip material, resulting in higher prediction accuracy.

[0032] The core of a time series forecasting model is a neural network function, whose mathematical expression is as follows: RUL(t) = NN_Model( HI(t), HI(t-1), ..., HI(tn) | E, U, M ) Where: RUL(t): the remaining service life predicted at time t (unit: days or wipe cycles); NN_Model(...): Represents a trained recurrent neural network model (such as LSTM or GRU); [HI(t), HI(t-1), ..., HI(tn)]: This is a sequence of health indices over the past n time points, which serves as the main input to the RNN to capture the trend of health degradation. E: Environmental factor vector, including historical average rainfall intensity, ultraviolet radiation intensity, temperature fluctuation, pollutant type (acid rain, dust, etc.) and rainfall acidity / alkalinity, etc. U: Usage habit vector, including average daily usage time, percentage of high-speed wiping mode, and aggressiveness of dynamic pressure adjustment; M: Rubber strip material feature vector, including rubber strip material type (silicone rubber, fluororubber, etc.), material hardness benchmark value and material aging sensitivity coefficient. Different materials of rubber strips have different aging rates. For example, under the same environment, 1 hour of operation of silicone rubber wiper blades is equivalent to 0.7 hours of operation of ordinary fluororubber wiper blades.

[0033] This neural network model, trained on a large amount of historical wiper failure data, learns a complex nonlinear mapping between the degradation trajectory of the health index HI(t) and eventual failure under various complex environments and usage patterns. The model dynamically updates its predictions, providing a personalized, continuously revised estimate of the remaining life for each wiper blade.

[0034] When the system detects that the wiper blades are nearing the end of their lifespan, the user reminder module will send maintenance reminders to the user via the in-vehicle display and a mobile application, including suggested replacement times and information on nearby service centers. This module serves as the user interface. When the lifespan prediction module calculates that the remaining lifespan RUL(t) is lower than a preset threshold, or the wear detection subunit determines in real-time that the health index HI(t) is lower than a safety threshold, the user reminder subunit will immediately activate a tiered early warning mechanism.

[0035] The tiered early warning system includes: Level 1 Reminder (Maintenance Recommendation): When RUL(t) < 30 days or HI(t) < 0.6, a gentle text message will be pushed through the in-vehicle display and mobile application, such as "Wiper blade life is about 30 days remaining. We recommend that you plan your maintenance."

[0036] Level 2 Reminder (Urgent Replacement): When RUL(t) < 7 days or HI(t) < 0.4, a prominent icon warning will be displayed on the dashboard, accompanied by a prompt sound; at the same time, a push notification will be sent via the mobile application, and the functions of nearby car service outlets and appointment for replacement service will be provided.

[0037] Level 3 Reminder (Emergency Replacement): When RUL(t) = 0 days or HI(t) < 0.25, the vehicle obtains the current precipitation level through at least two rain sensors, combines ambient light sensors and a high-resolution camera to detect surrounding environmental information in real time, calculates the current environmental risk index Risk_Env, quantifies the current driving visibility, and calculates the safe speed limit based on the current vehicle speed and road type. It then sends a progressive speed limit command to the engine control unit (ECU) via the vehicle network to avoid safety hazards caused by sudden deceleration. A prominent red flashing warning icon is displayed on the instrument panel, and the central control screen displays a "Wipe Blade Emergency Replacement - Speed ​​Limited" prompt. The warning is amplified by steering wheel vibration and accompanied by an audible alert. The system sends a speed limit command to the vehicle's powertrain via the CAN bus, and the Electronic Stability Program (ESP) works in conjunction to ensure smooth deceleration. Simultaneously, the system automatically recommends the nearest service center and provides one-click navigation, allowing for simultaneous appointment of the replacement service.

[0038] The conditions for lifting the Level 3 warning are: (1) the wiper blade replacement is confirmed by system reset or diagnostic equipment at the repair shop; (2) when the environmental risk coefficient Risk_Env drops to 0 and lasts for 5 minutes, the system gradually lifts the speed limit; (3) the driver can temporarily lift the speed limit by pressing and holding the confirmation button, but the system will record this operation and continue to display the warning.

[0039] The formula for calculating the current environmental risk index Risk_Env is as follows:

[0040] In the formula, W_p = 0.50 is the precipitation weight, which is the core factor that directly affects the wiping effect and field of vision; W_v = 0.30 is the visibility weight, which directly determines the field of vision; W_l = 0.20 is the illumination weight, which affects visual perception and fatigue. In the formula, s is a shape parameter used to control the compensation effect between factors. Let s = 2, which makes the formula approximate the weighted Euclidean distance, that is, a high risk of any factor will significantly increase the total risk, which is more in line with the principle of safety and conservatism; In the formula, C_syn is the synergistic enhancement factor. When multiple adverse conditions occur at the same time, the risks are not simply added together, but synergistically amplified to ensure that Risk_Env can reach its maximum value in the most dangerous situation such as "nighttime heavy rain and extremely low visibility". In the formula, R_rain is the risk value of the precipitation measured by the rain sensor (unit: mm / h) mapped to [0, 1]. No rain (< 0.1 mm / h): R_rain = 0 Light rain (0.1 - 2.5 mm / h): R_rain = 0.2 Moderate rain (2.5 - 10 mm / h): R_rain = 0.5 Heavy rain (10 - 50 mm / h): R_rain = 0.8 Heavy rainfall (> 50 mm / h): R_rain = 1.0 In the formula, R_vis is the estimated visibility distance (unit: meters). The calculation formula is as follows: R_vis = exp( -visibility distance / V_scale ) Where V_scale is the scale parameter, set to 500. This means: Visibility ≥ 2000 meters: R_vis ≈ 0 Visibility = 500 meters: R_vis ≈ 0.37 Visibility = 200 meters: R_vis ≈ 0.67 Visibility ≤ 100 meters: R_vis ≥ 0.82 In the formula, R_light is based on ambient light intensity (unit: lux) and day / night information.

[0041] Daytime (light intensity > 100 lux): R_light = 0 Dusk / Dawn (10 - 100 lux): R_light = 0.3 Nighttime (< 10 lux): R_light = 0.6 Nighttime + severe weather (nighttime and R_rain > 0.5): R_light = 0.8 Example calculations are shown below: Assume it is currently nighttime with heavy rain and visibility is 200 meters.

[0042] R_rain = 0.8 (Heavy rain) R_vis = exp(-200 / 500) ≈ 0.67 R_light = 0.8 (Nighttime + Severe Weather) Risk_Env=[(0.5*0.8)²+(0.3*0.67)²+(0.2*0.8)² ]^(1 / 2) +(1 / 2)*(0.8*0.67*0.8)^(1 / 2) = [ 0.16 + 0.040 + 0.026 ]^0.5 + 0.327 = [0.226]^0.5 + 0.327 ≈ 0.475 + 0.327 ≈ 0.802 The risk level is high, and the system should enforce a maximum speed limit of 50 km / h.

[0043] Information details: The reminder message not only includes time estimates, but may also include a simple analysis of the cause of wear, such as "wear has accelerated due to recent frequent use in dusty environments," to help users better understand the vehicle's condition.

[0044] The early warning function not only provides more timely reminders for equipment replacement, but more importantly, through deep integration with the environmental perception system, it achieves proactive safety control based on actual risks. When wiper blade failure may seriously affect driving safety, it ensures driving safety through intelligent speed limiting, demonstrating the system's leap from "predictive maintenance" to "proactive safety assurance".

[0045] The OTA upgrade management module of this system is responsible for the remote upgrade and optimization of the system, including algorithm model updates: receiving improved wiper wear detection and life prediction models from the cloud; updating system parameters such as current threshold, pressure threshold, and reminder strategies; completing remote firmware updates and bug fixes; adding new monitoring indicators or early warning functions through software updates; pushing personalized algorithm optimization packages based on the vehicle's usage environment and wiper blade wear status; and collecting system operation data and feeding it back to the cloud for subsequent optimization.

[0046] The specific principle of this embodiment is as follows: The system collects the operating current curve of the wiper motor in real time. For a healthy wiper blade, under constant pressure and without extreme environmental resistance, the current value during its wiping cycle will remain within a stable baseline range. As the wiper blade rubber ages and wears, its coefficient of friction with the windshield increases, leading to increased wiping resistance. The system will monitor an increase in the steady-state value and peak current of the motor. The system obtains the real-time pressure of the wiper blade on the glass through an integrated pressure sensor or indirectly through a motor torque model. During adaptive control, the system records the pressure value P(t) required to achieve a specific wiping effect. When the wiper blade wears down, the system needs to apply greater pressure to maintain the same level of cleanliness. Therefore, the "trend of increasing pressure required to maintain a specific level of cleanliness" is itself a strong indicator of wear. As final verification, the onboard camera analyzes the windshield after each wipe, quantitatively assessing the residual water film area, the number of streaks, and the amount of dirt residue, generating an objective "cleanliness score" S(t). Even under higher current and pressure, the S(t) score of a worn wiper blade will show a downward trend. The audio sensor captures the sound characteristics generated by the friction between the wiper blade and the glass during operation. As the rubber strip hardens, it will produce high-frequency abnormal noises. By analyzing the energy ratio of the abnormal noise frequency band, the degree of aging of the rubber strip can be determined.

[0047] This invention analyzes historical usage data and current wear status of windshield wipers, and uses a time-series prediction algorithm to estimate the remaining lifespan of the wiper blades. The algorithm considers various influencing factors such as usage frequency, environmental conditions, wiping patterns, and rubber strip material, resulting in higher prediction accuracy. When the system detects that the wiper blades are nearing the end of their lifespan, it will send a maintenance reminder to the user via the in-vehicle display and a mobile application, including suggested replacement time and information on nearby service centers. Example 2

[0048] A method for predicting and maintaining windshield wiper lifespan, employing the windshield wiper lifespan prediction and maintenance system described in Example 1, includes the following steps: S1. Data Acquisition: Real-time acquisition of wiper motor operating current data, wiper blade operating pressure data, post-wiping visual image data, and wiper operating audio data through current sensor, pressure sensor, vision sensor, and audio sensor; S2. Data preprocessing and feature extraction: The collected raw data is filtered, time-synchronized and aligned, and image-processed to extract the current feature vector V_I, pressure feature vector V_P, visual cleanliness feature vector V_S, and audio feature vector V_A, respectively. S3. Health status detection: Input the four feature vectors into the LSTM-based multimodal fusion model and output the wiper blade health index HI(t); S4. Remaining life prediction: Input the historical health index sequence, environmental factor vector E, usage habit vector U, and adhesive strip material feature vector M into the GRU-based time series prediction model, and output the remaining life RUL(t); S5. Tiered Early Warning: Based on the monitoring results of HI(t) and RUL(t), trigger the corresponding Level 1, Level 2 or Level 3 early warning, send reminder information to the user and implement corresponding safety control measures; S6. System Upgrade: Receives update packages from the cloud via the OTA upgrade module to remotely update and optimize algorithm models, system parameters, and firmware.

[0049] Step S1 includes: S11. Motor current data acquisition: The working current of the wiper motor is monitored in real time by a current sensor, with a sampling frequency of not less than 1kHz. The current waveform of each wiping cycle is recorded. The wiping cycle includes the start, constant speed and stop phases. S12. Wiper blade pressure data acquisition: The pressure value of the wiper blade on the windshield is obtained in real time through a pressure sensor integrated on the wiper arm or indirectly calculated based on the motor torque model. The sampling frequency is synchronized with the current to ensure data timing alignment. S13. Visual data acquisition of wiping effect: The windshield image is captured by a visual sensor after each wiping, and the image covers the main wiping area; S14. Wiping Audio Data Acquisition: Acquire audio signals generated by the friction between the wiper and the windshield during operation using a directional microphone, with a sampling frequency of not less than 44.1kHz, focusing on capturing the acoustic characteristics of abnormal noise frequency bands in the 5kHz-7kHz range.

[0050] Step S2 includes: S21. Current signal preprocessing: Apply a low-pass filter to the original current waveform to eliminate high-frequency noise and power supply interference; extract the following features from each scraping cycle: average current I_average during the uniform scraping phase, peak current I_peak during startup and commutation, current fluctuation amplitude I_wave during the uniform scraping phase, and total energy consumption Energy_consumption during this scraping cycle, forming a feature vector V_I. S22. Pressure signal preprocessing: Synchronize and align pressure data with current data using timestamps; extract the following features: the target pressure value P_target of the system command, the actual measured average pressure P_actual_average, and the fluctuation variance P_variance during the pressure control process, to form a feature vector V_P; S23. Visual Preprocessing of Scraping Effect: Image processing algorithms are used to analyze camera images and calculate cleanliness scores. By comparing the pixel differences between the scraped area and the ideal clean area, a comprehensive cleanliness percentage score S_score is obtained. Watermark lines are identified using edge detection algorithms. The ratio of the residual water film or stain area to the total field of view area is calculated to form a feature vector V_S=[S_score, Streak_count, Residue_area_ratio], where S_score is the comprehensive cleanliness percentage score; Streak_count is the number of strip-shaped watermarks or stains identified in the image; and Residue_area_ratio is the proportion of the residual water film or stain area to the total field of view area. S24. Audio signal preprocessing: Apply wavelet threshold denoising algorithm to the original audio signal to eliminate environmental noise (such as wind noise, engine noise, etc.), and convert the audio signal to the frequency domain through Fourier transform to extract the following features: abnormal noise frequency band energy ratio A_abnormal_ratio, short-time energy entropy A_entropy, and zero crossing rate A_zero_crossing, forming a feature vector V_A=[A_abnormal_ratio, A_entropy, A_zero_crossing], where the abnormal noise frequency band is 5kHz-7kHz. When the abnormal noise frequency band energy ratio exceeds 30%, it can be determined that the wiper blade rubber strip shows signs of aging.

[0051] Step S3 includes: inputting the current feature vector V_I, pressure feature vector V_P, visual cleanliness feature vector V_S, and audio feature vector V_A into the LSTM-based multimodal fusion model, and performing health status detection according to the following process: S31: Determine whether the current I(t) is trending upward. If not, return to step S1 and start again; if yes, proceed to step S32. S32: Determine whether the pressure P(t) increases synchronously. If not, proceed to step S36; if yes, proceed to step S33. S33: Determine whether the visual cleanliness S(t) has decreased. If not, proceed to step S35; if yes, proceed to step S34. S34: Strong wear characteristics confirmed, indicating that the windshield wipers are in a state of strong wear, proceed to step S37; S35: The cause is determined to be a minor performance change, possibly due to environmental changes or slight degradation from scratching. S36: It is determined that it may be a single interference that is ignored or the weight is reduced, which is a temporary interference situation; S37: Input the results of steps S34, S35, and S36 into the multimodal fusion model to calculate the health index HI(t); S38: Update the health index HI(t) to quantify the current health status of the windshield wipers; S39: Determine whether HI(t) is less than the threshold. If not, return to step S1; if yes, proceed to step S5. Step S37 includes: S371. Fusion and Inference Process: The feature vectors V_I, V_P, V_S, and V_A are input into the corresponding LSTM encoders to extract high-level feature representations H_I, H_P, H_S, and H_A; H_I, H_P, H_S, and H_A are concatenated to form a 256-dimensional fusion feature vector; a nonlinear mapping is performed through a fully connected network to finally output the health index HI(t); a forward propagation is performed after each scraping, and HI(t) is updated in real time; S372. Offline Training: Train the model on a cloud server using a historical dataset, which includes wiper blade data ranging from brand new to malfunctioning. Optimize the loss function and train for at least 1000 rounds to ensure model convergence. S373. Online Adaptation: When the user performs manual wiping or replaces the wiper blades, the model is fine-tuned using new data, and the weights of the LSTM and fully connected network are updated; the fine-tuning data is uploaded to the cloud via OTA for model optimization.

[0052] Step S4 includes: S41. Data Collection and Preprocessing: S411. Collect historical health index sequence: Collect the historical health index sequence [HI(t), HI(t-1), ..., HI(tn)] output by the wear detection subunit, where n is at least 50 scrapes; S412. Collect environmental data E: Obtain historical average rainfall intensity, UV intensity, temperature fluctuation, pollutant type, and rainfall acidity / alkalinity through vehicle-mounted sensors or external APIs, and normalize them into vectors; S413. Collect usage habit data U: Extract average daily usage time, high-speed wiping mode percentage, pressure dynamic adjustment frequency, etc. from the vehicle CAN bus, and normalize them into a vector; S414. Collect rubber strip material data M: Obtain the material type (silicone rubber, fluororubber, etc.), material hardness benchmark value and material aging sensitivity coefficient of the wiper rubber strip, and normalize them into a vector; the aging rate of rubber strips of different materials is different. For example, under the same environment, 1 hour of operation of silicone rubber wiper rubber strip is equivalent to 0.7 hours of operation of ordinary fluororubber wiper rubber strip. S42.RNN prediction model predicts: The GRU network learns sequence dependencies using the past n health index sequences, environmental vector E, usage habit vector U, and adhesive strip material vector M, outputting the hidden state. A fully connected layer maps the hidden state to RUL(t). Mathematically, RUL(t) = NN_Model(HI(t), HI(t-1), ..., HI(tn)|E,U,M), where NN_Model is the trained GRU model. If RUL(t) < 30 days, a user reminder is triggered. S43. Model Update: The initial training of the model is completed in the cloud, using historical failure data; the model parameters are updated regularly via OTA to adapt to new environmental patterns and usage habits, thereby improving prediction accuracy.

[0053] Step S5 includes: S51. Threshold Monitoring Real-time monitoring of RUL(t) and HI(t): If RUL(t) < 30 days or HI(t) < 0.6, trigger a Level 1 alert; If RUL(t) < 7 days or HI(t) < 0.4, a level 2 alert is triggered; If RUL(t) = 0 days or HI(t) < 0.25, a Level 3 alert is triggered, and an environmental risk assessment and speed limit mechanism are activated. S52. Reminder Generation and Sending The first-level reminder is "Maintenance Recommendation," which is displayed as a text message on the in-vehicle display screen and pushed to a mobile application. The secondary warning is "urgently need to be replaced," which is displayed with a prominent icon and warning sound on the dashboard; a notification is pushed through the mobile application, embedding a map of nearby service outlets and an appointment link; wear cause analysis is provided, such as "wear has accelerated due to frequent use in dusty environments recently" or "high-frequency abnormal noise detected, rubber strip has hardened"; The Level 3 alert is for "emergency replacement." A prominent flashing red warning icon is displayed on the instrument panel. The system calculates the current environmental risk index (Risk_Env), appropriately limits the vehicle speed, and displays a "Wipe Blade Emergency Replacement - Speed ​​Limited" message on the central control screen. The warning is amplified through steering wheel vibration and accompanied by an audible alert. The system sends a speed limit command to the vehicle's powertrain via the CAN bus, and the electronic stability system works in conjunction to ensure a smooth deceleration. Simultaneously, it automatically recommends the nearest service center and provides one-click navigation, allowing for simultaneous appointment of the replacement service. S53. User Interaction Users confirm the reminder through the in-vehicle interface or mobile app. The system records the feedback data and uploads it to the cloud for subsequent model optimization and reminder strategy adjustment.

[0054] Step S6 includes: S61. Upgrade Detection and Download The vehicle's ECU periodically connects to the cloud server to check for available updates; it downloads update packages via a secure HTTPS protocol and verifies digital signatures to ensure security. S62. Upgrade Execution Model update: Replace deep learning models for wear detection or lifespan prediction; Parameter update: Adjust the current threshold, pressure threshold, or alert threshold to adapt to different regional environmental conditions; Firmware update: Update system firmware to fix bugs or add features; Personalized optimization: Push customized algorithm packages based on vehicle usage history; S63. Data Feedback and Optimization The system anonymously collects operational data and uploads it to the cloud via an encrypted channel; the cloud uses this data to retrain the model, enabling continuous optimization and adaptive improvement of the system algorithm.

[0055] This invention utilizes a deep learning-based multi-signal fusion method, enabling the wear detection subunit to keenly capture early, subtle signs of wear that are undetectable by the human eye or simple threshold judgment, thus achieving a precise, robust, and forward-looking assessment of the health status of wiper blades.

Claims

1. A windshield wiper life prediction and maintenance system, characterized in that, It includes a sensor network, an on-board ECU, a cloud server, a wear detection module, a life prediction module, a user reminder module, and an OTA upgrade module, which are respectively deployed on the on-board ECU; The sensor network includes a current sensor, a pressure sensor, a vision sensor, and an audio sensor, used to collect real-time data on the working current of the wiper motor, the working pressure of the wiper blades on the windshield, the visual image data of the windshield after wiping, and the audio data of the wiper during operation. The wear detection module adopts a multimodal fusion model based on a long short-term memory network. It takes the current feature vector, pressure feature vector, visual cleanliness feature vector and audio feature vector collected and preprocessed by the sensor network as input, and outputs the wiper blade health index HI(t). The life prediction module adopts a time series prediction model based on recurrent neural networks. It inputs the historical health index sequence output by the wear detection module, and integrates environmental factor vectors, usage habit vectors and rubber strip material feature vectors to output the remaining life of the wiper blade RUL(t). The user reminder module monitors the health index HI(t) and remaining service life RUL(t) in real time. When the preset warning conditions are met, it issues a graded warning to the user through the vehicle terminal and / or mobile application. The OTA upgrade module communicates with the vehicle ECU and the cloud server respectively, and is used to receive algorithm model update packages, system parameter configuration packages and firmware update packages sent by the cloud server, so as to realize remote optimization and upgrade of the system.

2. The wiper life prediction and maintenance system according to claim 1, characterized in that, The multimodal fusion model based on long short-term memory networks includes multiple independent LSTM encoders and a feature fusion fully connected network. Each LSTM encoder is a multi-layer stacked structure, used to process current feature vectors, pressure feature vectors, visual cleanliness feature vectors, and audio feature vectors respectively. After extracting high-level features, the health index HI(t) is output through nonlinear mapping by the fully connected network. The health index HI(t) ranges from 0 to 1, where 0 represents complete failure and 1 represents a brand new state. The environmental factor vector E includes historical average rainfall intensity, ultraviolet radiation intensity, temperature fluctuation, pollutant type, and rainfall acidity / alkalinity. The usage habit vector U includes average daily usage time, the proportion of high-speed wiping mode, and the aggressiveness of dynamic pressure adjustment. The adhesive strip material feature vector M includes adhesive strip material type, material hardness benchmark value, and material aging sensitivity coefficient.

3. The wiper life prediction and maintenance system according to claim 1, characterized in that, The tiered early warning system includes: Level 1 alert: When RUL(t) < 30 days or HI(t) < 0.6, a gentle maintenance suggestion text message will be pushed; Level 2 Reminder: When RUL(t) < 7 days or HI(t) < 0.4, a prominent icon warning will be displayed along with a prompt sound, providing information on nearby service outlets and appointment functions; Level 3 Reminder: When RUL(t) = 0 days or HI(t) < 0.25, calculate the environmental risk index Risk_Env, send a speed limit command to the vehicle power system, and simultaneously display a red flashing warning icon, a steering wheel vibration reminder, and provide one-click navigation to service outlets.

4. The wiper life prediction and maintenance system according to claim 1, characterized in that, The current sensor is a Hall effect sensor with a sampling frequency of not less than 1kHz; the pressure sensor is a piezoelectric sensor or indirectly obtains pressure data through a motor torque model; the vision sensor is an in-vehicle camera with a resolution of not less than 1080p; the audio sensor is a directional microphone used to collect audio signals generated by the friction between the windshield wipers and the windshield during operation. The OTA upgrade module supports upgrades including: wear detection model updates, life prediction model updates, current threshold adjustments, pressure threshold adjustments, reminder strategy optimization, system firmware repair, and the addition of monitoring indicators or early warning functions.

5. A method for predicting and maintaining the lifespan of windshield wipers, characterized in that, The wiper life prediction and maintenance system according to any one of claims 1-4, the method includes the following steps: S1. Data Acquisition: Real-time acquisition of wiper motor operating current data, wiper blade operating pressure data, post-wiping visual image data, and wiper operating audio data through current sensor, pressure sensor, vision sensor, and audio sensor; S2. Data preprocessing and feature extraction: The collected raw data is filtered, time-synchronized and aligned, and image-processed to extract the current feature vector V_I, pressure feature vector V_P, visual cleanliness feature vector V_S, and audio feature vector V_A, respectively. S3. Health status detection: Input the four feature vectors into the LSTM-based multimodal fusion model and output the wiper blade health index HI(t); S4. Remaining life prediction: Input the historical health index sequence, environmental factor vector E, usage habit vector U, and adhesive strip material feature vector M into the GRU-based time series prediction model, and output the remaining life RUL(t); S5. Tiered Early Warning: Based on the monitoring results of HI(t) and RUL(t), trigger the corresponding Level 1, Level 2 or Level 3 early warning, send reminder information to the user and implement corresponding safety control measures; S6. System Upgrade: Receives update packages from the cloud via the OTA upgrade module to remotely update and optimize algorithm models, system parameters, and firmware.

6. The method for predicting and maintaining wiper life according to claim 5, characterized in that, Step S1 includes: S11. Motor current data acquisition: The working current of the wiper motor is monitored in real time by a current sensor, with a sampling frequency of not less than 1kHz. The current waveform of each wiping cycle is recorded. The wiping cycle includes the start, constant speed and stop phases. S12. Wiper blade pressure data acquisition: The pressure value of the wiper blade on the windshield is obtained in real time through a pressure sensor integrated on the wiper arm or indirectly calculated based on the motor torque model. The sampling frequency is synchronized with the current to ensure data timing alignment. S13. Visual data acquisition of wiping effect: The windshield image is captured by a visual sensor after each wiping, and the image covers the main wiping area; S14. Wiping audio data acquisition: Acquire audio signals generated by the friction between the wipers and the windshield during operation using a directional microphone, with a sampling frequency of not less than 44.1kHz; Step S2 includes: S21. Current signal preprocessing: Apply a low-pass filter to the original current waveform to eliminate high-frequency noise and power supply interference; extract the following features from each scraping cycle: average current I_average during the uniform scraping phase, peak current I_peak during startup and commutation, current fluctuation amplitude I_wave during the uniform scraping phase, and total energy consumption Energy_consumption during this scraping cycle, forming a feature vector V_I. S22. Pressure signal preprocessing: Synchronize and align pressure data with current data using timestamps; extract the following features: the target pressure value P_target of the system command, the actual measured average pressure P_actual_average, and the fluctuation variance P_variance during the pressure control process, to form a feature vector V_P; S23. Visual Preprocessing of Scraping Effect: Image processing algorithms are used to analyze camera images and calculate cleanliness scores. By comparing the pixel differences between the scraped area and the ideal clean area, a comprehensive cleanliness percentage score S_score is obtained. Watermark lines are identified using edge detection algorithms. The ratio of the residual water film or stain area to the total field of view area is calculated to form a feature vector V_S=[S_score, Streak_count, Residue_area_ratio], where S_score is the comprehensive cleanliness percentage score; Streak_count is the number of strip-shaped watermarks or stains identified in the image; and Residue_area_ratio is the proportion of the residual water film or stain area to the total field of view area. S24. Audio signal preprocessing: Apply wavelet threshold denoising algorithm to the original audio signal to eliminate environmental noise, and convert the audio signal to the frequency domain through Fourier transform to extract the following features: abnormal noise frequency band energy ratio A_abnormal_ratio, short-time energy entropy A_entropy, and zero crossing rate A_zero_crossing, forming a feature vector V_A=[A_abnormal_ratio, A_entropy, A_zero_crossing], where the abnormal noise frequency band is 5kHz-7kHz.

7. The method for predicting and maintaining wiper life according to claim 6, characterized in that, Step S3 includes: inputting the current feature vector V_I, pressure feature vector V_P, visual cleanliness feature vector V_S, and audio feature vector V_A into the LSTM-based multimodal fusion model, and performing health status detection according to the following process: S31: Determine whether the current I(t) is trending upward. If not, return to step S1 and start again; if yes, proceed to step S32. S32: Determine whether the pressure P(t) increases synchronously. If not, proceed to step S36; if yes, proceed to step S33. S33: Determine whether the visual cleanliness S(t) has decreased. If not, proceed to step S35; if yes, proceed to step S34. S34: Strong wear characteristics confirmed, indicating that the windshield wipers are in a state of strong wear, proceed to step S37; S35: The cause is determined to be a minor performance change, possibly due to environmental changes or slight degradation from scratching. S36: It is determined that it may be a single interference that is ignored or the weight is reduced, which is a temporary interference situation; S37: Input the results of steps S34, S35, and S36 into the multimodal fusion model to calculate the health index HI(t); S38: Update the health index HI(t) to quantify the current health status of the windshield wipers; S39: Determine whether HI(t) is less than the threshold. If not, return to step S1; if yes, proceed to step S5. Step S37 includes: S371. Fusion and Inference Process: The feature vectors V_I, V_P, V_S, and V_A are input into the corresponding LSTM encoders to extract high-level feature representations H_I, H_P, H_S, and H_A; H_I, H_P, H_S, and H_A are concatenated to form a 256-dimensional fusion feature vector; a nonlinear mapping is performed through a fully connected network to finally output the health index HI(t); a forward propagation is performed after each scraping, and HI(t) is updated in real time; S372. Offline Training: Train the model on a cloud server using a historical dataset, which includes wiper blade data ranging from brand new to malfunctioning. Optimize the loss function and train for at least 1000 rounds to ensure model convergence. S373. Online Adaptation: When the user performs manual wiping or replaces the wiper blades, the model is fine-tuned using new data, and the weights of the LSTM and fully connected network are updated; the fine-tuning data is uploaded to the cloud via OTA for model optimization.

8. The method for predicting and maintaining wiper life according to claim 5, characterized in that, Step S4 includes: S41. Data Collection and Preprocessing: S411. Collect historical health index sequence: Collect the historical health index sequence [HI(t), HI(t-1), ..., HI(tn)] output by the wear detection subunit, where n is at least 50 scrapes; S412. Collect environmental data E: Obtain historical average rainfall intensity, UV intensity, temperature fluctuation, pollutant type, and rainfall acidity / alkalinity through vehicle-mounted sensors or external APIs, and normalize them into vectors; S413. Collect usage habit data U: Extract average daily usage time, high-speed wiping mode percentage, pressure dynamic adjustment frequency, etc. from the vehicle CAN bus, and normalize them into a vector; S414. Collect rubber strip material data M: Obtain the material type, material hardness baseline value and material aging sensitivity coefficient of the wiper rubber strip, and normalize them into a vector; S42.RNN prediction model predicts: The past n health index sequences, environmental vector E, usage habit vector U, and adhesive strip material vector M are used to learn sequence dependencies through a GRU network, outputting the hidden state. A fully connected layer maps the hidden state to RUL(t). Mathematically, RUL(t) = NN_Model(HI(t), HI(t-1), ..., HI(tn)|E,U,M), where NN_Model is the trained GRU model. If RUL(t) < 30 days, a user reminder is triggered. S43. Model Update: The initial training of the model is completed in the cloud, using historical failure data; the model parameters are updated regularly via OTA to adapt to new environmental patterns and usage habits, thereby improving prediction accuracy.

9. The method for predicting and maintaining wiper life according to claim 5, characterized in that, Step S5 includes: S51. Threshold Monitoring Real-time monitoring of RUL(t) and HI(t): If RUL(t) < 30 days or HI(t) < 0.6, trigger a Level 1 alert; If RUL(t) < 7 days or HI(t) < 0.4, a level 2 alert is triggered; If RUL(t) = 0 days or HI(t) < 0.25, a Level 3 alert is triggered, and an environmental risk assessment and speed limit mechanism are activated. S52. Reminder Generation and Sending The first-level reminder is "Maintenance Recommendation," which is displayed as a text message on the in-vehicle display screen and pushed to a mobile application. The second-level alert is "urgently need replacement," displayed as a prominent icon and warning sound on the dashboard; a push notification is sent via the mobile application, embedding a map of nearby service centers and an appointment link; and wear and tear analysis is provided. The Level 3 alert is for "emergency replacement." A prominent flashing red warning icon is displayed on the instrument panel, the current environmental risk index is calculated, the vehicle speed is appropriately limited, and a "Wipe Blade Emergency Replacement - Speed ​​Limited" message is displayed on the central control screen. The warning is amplified through steering wheel vibration and accompanied by an audible alert. The system sends a speed limit command to the vehicle's powertrain via the CAN bus, and the electronic stability system works in conjunction to ensure smooth deceleration. Simultaneously, it automatically recommends the nearest service center and provides one-click navigation, allowing for simultaneous appointment of the replacement service. S53: User Interaction Users confirm the reminder through the in-vehicle interface or mobile app. The system records the feedback data and uploads it to the cloud for subsequent model optimization and reminder strategy adjustment.

10. The method for predicting and maintaining the lifespan of windshield wipers according to claim 5, characterized in that, Step S6 includes: S61: Upgrade Detection and Download The vehicle's ECU periodically connects to the cloud server to check for available updates; it downloads update packages via a secure HTTP protocol and verifies digital signatures to ensure security. S62: Upgrade Execution Model update: Replace deep learning models for wear detection or lifespan prediction; Parameter update: Adjust the current threshold, pressure threshold, or alert threshold to adapt to different regional environmental conditions; Firmware update: Update system firmware to fix bugs or add features; Personalized optimization: Push customized algorithm packages based on vehicle usage history; S63: Data Feedback and Optimization The system anonymously collects operational data and uploads it to the cloud via an encrypted channel; the cloud uses this data to retrain the model, enabling continuous optimization and adaptive improvement of the system algorithm.

Images