A method and system for identifying and detecting automotive faults

Abstract

This application provides a method and system for identifying automotive faults. The method includes: acquiring operational status data of an automotive seat through a multimodal sensor network; wherein the multimodal sensor network includes a current sensor, a temperature sensor, a triaxial accelerometer, and a displacement sensor, and the operational status data includes at least current signals, vibration signals, temperature signals, and displacement signals; generating a seat status mapping matrix based on the multimodal data; inputting the seat status mapping matrix into a preset hybrid diagnostic model to perform fault identification; and, based on the fault identification results output by the hybrid diagnostic model, providing a fault warning and generating a fault risk quantification report. The method provided in this application can effectively identify faults occurring in automotive seats.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fault detection, in particular to an automobile fault identification method and detection system. BACKGROUND

[0002] In the field of automobile aftermarket, the demand for automatic detection and repair of seat faults is increasingly urgent. Moreover, with the continuous improvement of vehicle intelligence and comfort requirements, the functions of automobile seats are becoming increasingly complex.

[0003] Traditional seat fault detection methods mainly rely on manual experience, which is not only inefficient but also lacks accuracy. Especially when faced with complex fault patterns, traditional methods often fail to provide timely and accurate diagnostic results, leading to increased repair costs and decreased user experience.

[0004] Therefore, the present application provides an automobile fault identification method and detection system to solve one of the above technical problems. SUMMARY

[0005] The purpose of the present application is to provide an automobile fault identification method and detection system that can solve at least one of the above technical problems. The specific scheme is as follows:

[0006] According to the specific embodiment of the present application, in a first aspect, the present application provides an automobile fault identification method, comprising: obtaining running state data of an automobile seat through a multi-modal sensor network; wherein the multi-modal sensor network includes a current sensor, a temperature sensor, a three-axis accelerometer, and a displacement sensor, and the running state data includes at least current signals, vibration signals, temperature signals, and displacement signals; generating a seat state mapping matrix based on the multi-modal data; inputting the seat state mapping matrix into a pre-set hybrid diagnostic model to perform fault identification; and performing fault warning and generating a fault risk quantification report based on the fault identification result output by the hybrid diagnostic model.

[0007] In one embodiment, the seat state mapping matrix is generated based on the multi-modal data, including: performing feature extraction on the running state data in time series to generate a multi-dimensional feature vector; and performing dynamic mapping on the multi-dimensional feature vector based on a pre-set threshold range to generate the seat state mapping matrix.

[0008] In one embodiment, the state mapping matrix contains the deviation proportion between the real-time value of each sensor signal and the pre-set threshold value, which is represented by the following formula: ; wherein, i is the time point index, j is the sensor index, represents the i time point, jThe deviation ratio between the real-time value of each sensor signal and a preset threshold. Indicates the first i The first time point j Each sensor signal value, Indicates the first j The preset threshold upper limit for each sensor signal value.

[0009] In one embodiment, the hybrid diagnostic model performs fault identification as follows: in response to any deviation ratio in the state mapping matrix exceeding a ratio threshold, the characterization term of the deviation ratio is identified as an abnormal mode; for the abnormal mode, trajectory prediction is performed using a Bi-LSTM neural network with the following formula to determine the offset of the seat adjustment path: ;in, k For sensor type labeling, n Number of sensor types This represents the total offset of the seat adjustment path. For the first k Weights of sensor-like signals, For the first k The deviation value of the sensor signal; the generation of a fault analysis report based on the fault identification result output by the hybrid diagnostic model, including: generating a fault risk quantification report in response to the total offset of the seat adjustment path exceeding a preset critical value.

[0010] In one embodiment, the Bi-LSTM neural network is configured as follows: the input of the Bi-LSTM neural network includes historical data from the multimodal sensor network, and the input data format is a time-series tensor with a dimension of [T×]. n ], where T is the number of time steps, n The number of sensor types is denoted by ; the Bi-LSTM neural network comprises two LSTM units and one fully connected layer, with each LSTM unit containing 64 nodes; the loss function is the weighted mean square error (WMSE) function; the initial weights of various sensor signals are allocated as follows: current signal weight 0.4; vibration signal weight 0.3; temperature signal weight 0.2; displacement signal weight 0.1; the output of the Bi-LSTM neural network is the total offset of the seat adjustment path.

[0011] In an embodiment, the method further comprises: determining a current risk level according to the offset distribution in the failure risk quantification report, in combination with historical maintenance data, using a fuzzy logic algorithm; in response to the current risk level being a first risk level, triggering an emergency shutdown and prompting immediate replacement of the seat assembly; in response to the current risk level being a second risk level, recommending replacement of the seat motor brush; and in response to the current risk level being a third risk level, prompting the user to regularly check the lubrication of the seat slide rail.

[0012] In an embodiment, the state mapping matrix is configured with a dynamic updating mechanism, comprising: performing wavelet transform on the current signal to extract fundamental frequency components and harmonics thereof; performing short-time Fourier transform on the vibration signal to extract energy distribution in a specified frequency band; performing moving average filtering on the temperature signal; dynamically configuring a temperature threshold range according to the seat usage environment; dynamically configuring a current threshold range according to the seat usage time; combining historical results of failure early warning and historical maintenance records, analyzing the occurrence probability of different failure modes through a clustering algorithm, and dynamically updating the weight distribution of the state mapping matrix according to the occurrence probability.

[0013] In an embodiment, the method further comprises: obtaining structured data in the maintenance process, the structured data including failure type, maintenance time consumption, spare part model, and technician feedback text; screening target data in the structured data that meets a predetermined data proportion, and fusing the target data with historical maintenance records; in response to triggering a new failure early warning, and the cosine similarity between the failure risk quantification report and a historical failure risk quantification report being greater than a preset similarity threshold, pushing a historical solution corresponding to the historical failure risk quantification report.

[0014] According to the specific embodiment of the present application, in a second aspect, the present application provides an automobile fault detection system, comprising:

[0015] A multi-modal sensor network module is used to collect the running state data of the car seat, and the running state data at least includes current signal, vibration signal, temperature signal and displacement signal; the multi-modal sensor network includes: a current sensor used to collect the current signal when the seat motor is running; a temperature sensor used to collect the temperature signal of the seat surface and the inside; a three-axis accelerometer used to collect the vibration signal of the seat slide rail; a displacement sensor used to collect the displacement signal in the seat adjustment process; a data processing module used to perform feature extraction and dynamic mapping on the running state data to generate a seat state mapping matrix; the data processing module includes: a feature extraction submodule used to perform feature extraction on the time series data to generate a multi-dimensional feature vector; a state mapping submodule used to perform dynamic mapping on the multi-dimensional feature vector based on a preset threshold range to generate a seat state mapping matrix, wherein the state mapping matrix contains the deviation proportion between the real-time value of each sensor signal and the preset threshold, and the formula is: ; wherein, i is a time point label, j is a sensor label, represents the deviation proportion between the real-time value of the first i sensor signal at the first j time point and the preset threshold, represents the first i sensor signal value at the first j time point, represents the upper limit of the preset threshold of the first j sensor signal value; a hybrid diagnosis model module used to perform fault identification on the seat state mapping matrix, including: an abnormal pattern recognition submodule used to detect whether any deviation proportion in the state mapping matrix exceeds the proportion threshold, and if so, it is determined as an abnormal pattern; a trajectory prediction submodule used to calculate the total deviation of the seat adjustment path through a Bi-LSTM neural network, and the formula is: ; wherein, k is a sensor type label, n is the number of sensor types, is the total deviation of the seat adjustment path, is the weight of the first k sensor signal, is the weight of the first kThe bias value of the sensor signal; a fault early warning and risk quantification module for generating a fault risk quantification report according to the output of the hybrid diagnostic model; the fault early warning and risk quantification module comprises: a risk level determination submodule for determining the current risk level by combining the offset distribution in the fault risk quantification report and the historical maintenance data, and using a fuzzy logic algorithm; a maintenance suggestion submodule for pushing maintenance suggestions according to the risk level, including: if the risk level is a first risk level, triggering an emergency shutdown and prompting to replace the seat component immediately; if the risk level is a second risk level, suggesting to replace the seat motor brush; if the risk level is a third risk level, prompting the user to check the lubrication degree of the seat slide rail regularly; a dynamic updating module for dynamically optimizing the state mapping matrix and the hybrid diagnostic model, including: a sensor data preprocessing submodule for performing wavelet transform on the current signal to extract the fundamental frequency and its harmonics, performing short-time Fourier transform on the vibration signal to extract the energy distribution of the specified frequency band, and performing moving average filtering on the temperature signal; a threshold self-adaptive submodule for dynamically adjusting the temperature threshold range according to the seat use environment and adjusting the current threshold range according to the service life of the seat; a weight updating submodule for dynamically updating the weight distribution of the state mapping matrix by analyzing the occurrence probability of historical fault modes through a clustering algorithm; a closed-loop feedback module for synchronizing the fault early warning result and the maintenance record through a cloud database, and matching the new fault with the historical cases based on a cosine similarity algorithm, and if the similarity is greater than a preset threshold, the corresponding historical solution is pushed.

[0016] Compared with the prior art, the above-mentioned scheme of the embodiments of the present application has at least the following beneficial effects:

[0017] The present application provides an automobile fault identification method, which can comprehensively cover the running state of the key functional modules of the seat, such as mechanical adjustment, electronic control and environmental adaptation, by deploying a multi-modal sensor network to collect the running state data of the seat, and inputting the multi-source data into a hybrid diagnostic model, which not only retains the rapid response ability of the rule base to known fault modes, but also improves the identification ability of complex fault modes through a machine learning model, and finally generates a fault risk quantification report to provide a clear basis for maintenance decision, effectively solving the problems of low diagnostic efficiency and insufficient accuracy caused by traditional reliance on artificial experience. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of an automobile fault identification method is shown;

[0019] Figure 2 A flowchart of a method for generating a seat state mapping matrix based on multi-modal data is shown;

[0020] Figure 3 A flowchart of a method for executing fault identification using a hybrid diagnostic model is shown;

[0021] Figure 4 A method flowchart showing further analysis from the failure risk quantification report is shown;

[0022] Figure 5 A method flowchart showing outputting a solution is shown;

[0023] Figure 6 A unit block diagram of an automobile fault detection system according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0025] The terms used in the embodiments of the present application are merely for the purpose of describing particular embodiments and are not intended to limit the present application. The singular forms "a," "an," and "the" used in the embodiments of the present application and the appended claims are intended to include the plural forms as well, unless the context clearly indicates otherwise. "Plural" generally includes at least two.

[0026] It should be understood that the term "and / or" used herein only describes an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, the character " / " herein generally means that the front and rear associated objects are in an "or" relationship.

[0027] It should be understood that although the terms first, second, third, etc. can be used in the embodiments of the present application to describe, these descriptions should not be limited to these terms. These terms are only used to distinguish the description. For example, without departing from the scope of the embodiments of the present application, the first can also be called the second, and similarly, the second can also be called the first.

[0028] Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if it is determined" or "if (a stated condition or event) is detected" can be interpreted as "when it is determined" or "in response to determining" or "when (a stated condition or event) is detected" or "in response to detecting (a stated condition or event)".

[0029] It is also important to note that the terms "comprises" and / or "comprising", or "includes" and / or "including" when used in this specification, specify the presence of stated features, integers, steps, operations, elements, or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or groups thereof.

[0030] In particular, it is to be noted that symbols and / or numbers present in the description, if not marked in the description of the figures, are not figure references.

[0031] The optional embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0032] The embodiments provided by the present application are embodiments of a vehicle fault identification method.

[0033] The embodiments provided by the present application are embodiments of a vehicle fault identification method. Figure 1 The embodiments provided by the present application are embodiments of a vehicle fault identification method.

[0034] Figure 1 A flowchart of a vehicle fault identification method is shown, as shown in Figure 1 The flowchart includes the following steps.

[0035] Step S101, obtaining running state data of a vehicle seat through a multi-modal sensor network.

[0036] The multi-modal sensor network includes a current sensor, a temperature sensor, a three-axis accelerometer, and a displacement sensor, and the running state data includes at least a current signal, a vibration signal, a temperature signal, and a displacement signal.

[0037] Step S102, generating a seat state mapping matrix based on multi-modal data.

[0038] Step S103, inputting the seat state mapping matrix into a preset hybrid diagnosis model to perform fault identification.

[0039] Step S104, according to the fault identification result output by the hybrid diagnosis model, performing fault warning and generating a fault risk quantification report.

[0040] The application provides a vehicle fault identification method. The running state data of a vehicle seat is collected through deployment of a multi-modal sensor network, which can comprehensively cover the running states of key functional modules such as mechanical adjustment, electronic control and environmental adaptation of the seat. Multi-source data is input into a hybrid diagnosis model, which not only retains the rapid response capability of the rule base to known fault modes, but also improves the identification capability of complex fault modes through a machine learning model. The finally generated fault risk quantization report provides a clear basis for maintenance decision-making, effectively solving the problems of low diagnosis efficiency and insufficient accuracy caused by traditional reliance on artificial experience.

[0041] In the embodiment of the application, the seat state mapping matrix generated based on multi-modal data is a multi-dimensional dynamic structured data table. Its core function is to compare the multi-source sensor signals collected during the running of the vehicle seat with the preset threshold value, and quantify the real-time state deviation degree of each sensor signal. Specifically, the matrix takes time points as rows and sensor types as columns. The deviation proportion of the real-time value of a certain type of sensor signal at a corresponding time point and the preset threshold value is stored in each cell. For example, if the current signal value at a certain time point is 8A and the upper limit of the preset threshold value is 10A, the deviation proportion is 20%. If the frequency energy distribution of the vibration signal exceeds the normal range, the value of the corresponding cell will increase significantly. The matrix reflects the running state of each functional module of the seat in real time through dynamic mapping, provides quantitative input for the hybrid diagnosis model, and thus realizes accurate identification of abnormal modes and fault warning.

[0042] Figure 2 A method flowchart for generating a seat state mapping matrix based on multi-modal data is shown, as shown in Figure 2 The method comprises the following steps.

[0043] In step S201, the running state data is subjected to feature extraction according to time sequence, and a multi-dimensional feature vector is generated.

[0044] In step S202, the multi-dimensional feature vector is subjected to dynamic mapping based on a preset threshold range, and a seat state mapping matrix is generated.

[0045] For example, the current threshold is preset to 5A-10A, the vibration frequency threshold is preset to 0-20Hz, and the temperature threshold is preset to 20℃-50℃.

[0046] In the embodiment of the application, the seat state mapping matrix is generated through time sequence feature extraction and dynamic mapping, which can convert multi-modal data into quantifiable state indicators. This mapping mechanism not only improves the real-time performance of fault detection, but also adapts to the running characteristics of the seat under different working conditions through dynamic threshold adjustment. For example, the current threshold range is automatically relaxed as the service life of the seat increases, avoiding false judgments caused by aging, thereby prolonging the effective monitoring period of the system and reducing maintenance costs.

[0047] For example, the state mapping matrix contains the deviation ratio between the real-time value of each sensor signal and a preset threshold, expressed by the following formula:

[0048] .

[0049] in, i To label the time points, j For sensor labeling, Indicates the first i The first time point j The deviation ratio between the real-time value of each sensor signal and a preset threshold. Indicates the first i The first time point j Each sensor signal value, Indicates the first j The preset threshold upper limit for each sensor signal value.

[0050] In this embodiment, a deviation ratio formula is used to quantify the degree of deviation between the sensor signal and a preset threshold, which can intuitively reflect the abnormal state of each functional module of the seat. For example, when the deviation ratio of the vibration signal exceeds the threshold, the system can prioritize checking for slide rail wear problems. An abnormal temperature signal may indicate a malfunction in the heating or ventilation system. This mathematically-based quantification method enhances the accuracy of fault location, reduces subjective judgment errors by maintenance personnel, and thus shortens fault diagnosis time.

[0051] Figure 3 A flowchart illustrating a method for fault identification using a hybrid diagnostic model is shown, such as... Figure 3 As shown, it includes the following steps.

[0052] Step S301: In response to any deviation ratio in the state mapping matrix exceeding the ratio threshold, the characterization term of the deviation ratio is determined as an abnormal mode.

[0053] Step S302: For abnormal patterns, trajectory prediction is performed using a Bi-LSTM neural network to determine the offset of the seat adjustment path.

[0054] In some embodiments, the Bi-LSTM neural network uses the following formula to determine the offset of the seat adjustment path:

[0055] .

[0056] in, k For sensor type labeling, n Number of sensor types This represents the total offset of the seat adjustment path. For the first kWeights of sensor-like signals, For the first k The deviation value of the sensor signal.

[0057] Step S303: If the total offset of the seat adjustment path exceeds a preset threshold, a fault risk quantification report is generated.

[0058] In this embodiment, a Bi-LSTM neural network is used to predict the trajectory of abnormal patterns in the state mapping matrix, which can capture the offset trend of the seat adjustment path. For example, when both motor current and vibration signal are abnormal, the model can predict the path offset that may be caused by slide rail jamming and trigger an early warning before the offset exceeds a critical value. This deep learning-based prediction mechanism makes up for the shortcomings of traditional static threshold judgment, realizes the transformation from "passive maintenance" to "proactive prevention", and significantly reduces the risk of sudden failures.

[0059] As a specific embodiment, the Bi-LSTM neural network is configured as follows:

[0060] The input to the Bi-LSTM neural network includes historical data from a multimodal sensor network. The input data format is a time-series tensor with dimensions [T×]. n ], where T is the number of time steps, n This represents the number of sensor types.

[0061] The Bi-LSTM neural network consists of two LSTM units and one fully connected layer, with each LSTM unit containing 64 nodes.

[0062] The loss function is the weighted mean squared error (WMSE) function.

[0063] The initial weights of various sensor signals are allocated as follows:

[0064] The current signal weight is 0.4;

[0065] Vibration signal weight 0.3;

[0066] Temperature signal weight 0.2;

[0067] The displacement signal weight is 0.1.

[0068] The output of the Bi-LSTM neural network is the total offset of the seat adjustment path.

[0069] In this embodiment, the Bi-LSTM neural network adopts a two-layer LSTM unit + fully connected layer structure and uses weighted mean square error as the loss function, which can effectively balance the influence of different sensor signals on fault prediction. Specifically, the high weight of the current signal ensures that motor faults are identified first, while the low weight of the displacement signal avoids interference from minor deviations in the guide rail on the overall judgment. This hierarchical weight allocation mechanism improves the model's adaptability to multi-source heterogeneous data, and further enhances the stability of the prediction results by training and optimizing parameters using historical data.

[0070] Figure 4 A flowchart illustrating a method for further analysis based on a fault risk quantification report is shown, such as... Figure 4 As shown, it includes the following steps.

[0071] Step S401: Based on the offset distribution in the fault risk quantification report and combined with historical maintenance data, a fuzzy logic algorithm is used to determine the current risk level.

[0072] Step S402a: In response to the current risk level being the first risk level, an emergency stop is triggered and a prompt is made to immediately replace the seat components.

[0073] Step S402b: In response to the current risk level being Level 2, it is recommended to replace the seat motor brushes.

[0074] In step S402c, in response to the current risk level being the third risk level, the user is prompted to periodically check the lubrication of the seat slide rails.

[0075] In this embodiment, a fuzzy logic algorithm combining fault risk quantification reports and historical maintenance data can transform abstract offset distributions into specific maintenance recommendations. For example, when the slide rail offset is at a medium risk level, the system can recommend replacing the motor brushes instead of immediately shutting down the machine, thus avoiding resource waste caused by over-maintenance. This tiered risk assessment strategy not only improves the targeting of maintenance solutions but also reduces operating costs through differentiated responses, making it particularly suitable for centralized management scenarios of large fleets.

[0076] In some embodiments, the state mapping matrix is ​​configured with a dynamic update mechanism, specifically including the following update mechanisms:

[0077] Mechanism 1: Perform wavelet transform on the current signal to extract the fundamental frequency component and its harmonics.

[0078] Mechanism 2: Perform a short-time Fourier transform on the vibration signal to extract the energy distribution of a specified frequency band.

[0079] Mechanism 3: Apply moving average filtering to the temperature signal.

[0080] Mechanism 4: Dynamically configure the temperature threshold range based on the seat's usage environment. For example, the upper limit of the threshold is adjusted to 60℃ in summer and the lower limit is adjusted to 0℃ in winter.

[0081] Mechanism 5: Dynamically configure the current threshold range based on the age of the seat. For example, configure the current threshold for seats used for 0-3 years, 3-5 years, and more than 5 years. Taking more than 5 years as an example, the upper limit of the current threshold for seats used for more than 5 years can be adjusted to 12A.

[0082] Mechanism 6: Combining historical fault warning results with historical maintenance records, clustering algorithms are used to analyze the probability of occurrence of different fault modes, and the weight allocation of the state mapping matrix is ​​dynamically updated according to the probability of occurrence.

[0083] In the above embodiments, the dynamic update mechanism extracts the fundamental frequency component of the current signal using wavelet transform, analyzes the energy distribution of the vibration signal frequency band using short-time Fourier transform, and processes the temperature signal using moving average filtering. This effectively separates noise and enhances the extraction accuracy of key features. For example, automatically adjusting the upper limit of the temperature threshold in high-temperature summer environments can avoid false alarms caused by environmental fluctuations. Furthermore, dynamically calibrating the current threshold based on the seat's service life extends the system's applicability during long-term operation. This adaptive adjustment capability significantly improves the environmental robustness of fault detection.

[0084] In this embodiment of the application, the hybrid diagnostic model includes a rule base and a machine learning model, wherein the machine learning model preferably uses a Bi-LSTM neural network.

[0085] In the above embodiments, each step is completed by a rule base or a machine learning model, respectively. The following will describe in detail the specific participation methods and application scenarios of the rule base and the machine learning model.

[0086] In some embodiments, the rule base in the hybrid diagnostic model mainly performs threshold determination, abnormal pattern recognition, and risk level classification, specifically including the following:

[0087] Threshold determination: The rule base is used to define preset threshold ranges and compare the deviation ratios of multimodal sensor signals in real time. For example, in high-temperature environments during summer, the rule base dynamically adjusts the upper limit of the temperature threshold to 60℃ to adapt to environmental changes.

[0088] Abnormal pattern recognition: If the deviation ratio exceeds 10%, the rule base directly triggers abnormal pattern judgment. For example, when the motor current fluctuation exceeds 10% and lasts for more than 5 seconds, the rule base immediately judges it as "motor stall" and recommends replacing the motor brushes.

[0089] Risk Level Classification: Based on preset logic, the offset distribution is classified into low / medium / high risk levels, and maintenance suggestions are pushed accordingly. For example, if the slide rail offset is at the medium risk level, the system recommends replacing the seat motor brushes instead of immediately shutting down the machine, thereby avoiding resource waste caused by excessive maintenance.

[0090] As a feasible implementation, the machine learning model is responsible for trajectory prediction and dynamic weight adjustment in the hybrid diagnostic model, specifically including the following:

[0091] Trajectory prediction: The machine learning model, trained on historical data, is used to calculate the total offset of the seat adjustment path. For example, by analyzing historical data from the past 100 seat adjustments, the model predicts the offset of the current adjustment path and corrects for the error between the actual and predicted values.

[0092] Dynamic weight adjustment: The machine learning model updates the weight allocation based on new fault data. For example, if a certain type of fault occurs frequently, the model automatically increases the weight of vibration signals to improve its sensitivity to that type of fault.

[0093] As a specific implementation, the rule base and the machine learning model each play an important role in the dynamic threshold adjustment and closed-loop feedback mechanism.

[0094] The rule base participates in static threshold definition and environment adaptation logic, as detailed below:

[0095] Static threshold definition: The rule base has a preset basic threshold range. For example, in low-temperature winter environments, the rule base automatically relaxes the current threshold to 12A to accommodate the low-temperature starting characteristics of motors.

[0096] Environmental adaptation logic: The rule base adjusts thresholds based on the season or the age of the seats. For example, in high-temperature summer environments, the rule base dynamically adjusts the upper limit of the temperature threshold to 60℃ to avoid false alarms caused by temporary failures of the air conditioning system.

[0097] The machine learning model participates in threshold adaptive optimization and closed-loop feedback, as follows:

[0098] Threshold adaptive optimization: Machine learning models analyze high-frequency failure modes through clustering algorithms and dynamically adjust the weight distribution of the state mapping matrix. For example, statistics show that 15% of seat rail jamming occurs in a certain car model, mainly in vehicles that have been used for more than 3 years. Companies can then make targeted improvements to the rail materials or add self-lubricating coating processes.

[0099] Closed-loop feedback driven: The machine learning model updates the parameters of the hybrid diagnostic model based on maintenance records. For example, by aggregating matrix data of millions of seats through a cloud database, the model updates its parameters regularly, continuously improving the system's adaptability.

[0100] In addition, rule bases and machine learning models each play different roles in the implementation of trajectory prediction algorithms.

[0101] The rule base participates in input data validation and output result verification, as detailed below:

[0102] Input data validation: The rule base ensures the integrity of the input data. For example, before each seat adjustment, the rule base checks whether the signals from each sensor meet the preset accuracy requirements to avoid affecting the prediction results due to data loss or noise interference.

[0103] Output verification: The rule base verifies the error of the trajectory prediction algorithm. For example, if the error between the predicted value ΔD and the actual value exceeds 5%, the rule base triggers an alarm, prompting the sensor to be recalibrated or the model parameters to be updated.

[0104] The mechanisms by which machine learning models participate in Bi-LSTM model construction and retraining are as follows:

[0105] Bi-LSTM model construction: The machine learning model is responsible for implementing the trajectory prediction algorithm, specifically including: the input data is historical data of multimodal sensor signals, the model structure is 2 layers of LSTM units and 1 layer of fully connected layer, the loss function is weighted mean square error, and the weight allocation is consistent with the weights defined in claim 11.

[0106] Model retraining mechanism: If the prediction error exceeds 5%, the machine learning model triggers a retraining process to update the Bi-LSTM parameters. For example, by periodically uploading user feedback data to the cloud, the model automatically updates its parameters, ensuring long-term operational stability.

[0107] In some embodiments, the seat state mapping matrix can be applied to production line quality control.

[0108] For example, in the automotive manufacturing process, a seat status mapping matrix can be embedded in the assembly line to monitor the assembly accuracy of each seat in real time. For instance, high-precision sensors, such as magnetostrictive displacement sensors and triaxial accelerometers, can be deployed at assembly stations to collect data on the straightness error of the seat slide rails, motor current fluctuations, and temperature distribution. Real-time data is compared with preset thresholds to generate a deviation ratio matrix. If the straightness deviation ratio of the slide rails in a batch of seats exceeds 5%, the matrix will immediately trigger an alarm, prompting the assembly robot to adjust the slide rail installation angle or replace worn parts. Matrix data is linked to production batches, forming digital quality files. If subsequent users report abnormal seat noises, the problematic batch can be quickly located using historical matrix data, reducing recall costs.

[0109] As a feasible implementation, the seat state mapping matrix can be applied to after-sales maintenance scenarios.

[0110] For example, in 4S stores or third-party repair stations, the seat status mapping matrix can help technicians quickly diagnose complex faults. When a user reports that the power seat cannot be adjusted, the system automatically retrieves the current signal, vibration signal, and temperature data of the current seat, generates a matrix, and calculates the deviation ratio of each sensor. Rule base priority: If the current deviation ratio exceeds 30% and the duration is >5 seconds, the rule base directly determines it as "motor stall," recommending replacement of the motor brushes; Machine learning model supplement: If the deviation ratio is below the threshold but non-linear characteristics exist, the Bi-LSTM neural network will predict the path offset caused by foreign objects in the slide rail, recommending cleaning the slide rail and checking the lubrication. The system overlays the fault location onto the seat's 3D model, guiding technicians to use specific tools to clean the foreign objects and providing grease type recommendations, reducing trial and error time.

[0111] As a specific implementation, the seat state mapping matrix can be applied to real-time early warning during vehicle operation.

[0112] Specifically, after the vehicle is delivered to the user, the seat status mapping matrix enables proactive health management. In high-temperature summer environments, the matrix dynamically adjusts the upper limit of the temperature threshold to avoid false alarms caused by temporary malfunctions of the air conditioning system; in low-temperature winter environments, it automatically relaxes the current threshold to adapt to the low-temperature start-up characteristics of the motor. When the seat adjustment path deviates, the Bi-LSTM neural network combines historical data to predict the wear trend of the slide rail. If the deviation is expected to exceed the critical value within 100 adjustments, the system will push a "slide rail lubrication insufficient, check recommended" reminder to the vehicle's central control screen, guiding the user to schedule maintenance. If the matrix detects abnormal motor winding resistance and a sudden temperature rise to 80°C, the system triggers an emergency stop and locks the seat adjustment function, while simultaneously uploading a fault code to the cloud via the vehicle communication module for remote diagnosis.

[0113] In addition, the seat condition mapping matrix can also be applied to long-term data analysis and preventive maintenance.

[0114] For example, through a closed-loop feedback mechanism, the seat state mapping matrix can drive companies to optimize product design and maintenance strategies. A cloud-based database aggregates matrix data from millions of seats, and clustering algorithms identify high-frequency failure patterns. For instance, statistics show that seat rail jamming accounts for 15% of a certain car model's seats, primarily occurring in vehicles older than three years. Companies can then specifically improve rail materials or add self-lubricating coatings. Based on maintenance record analysis, if the weight contribution of vibration signals in rail failure cases increases to 40%, the system automatically adjusts the input weights of the Bi-LSTM model to enhance sensitivity to vibration anomalies. Through correlation analysis between matrix data and driving habits, the system can push personalized suggestions to users, extending seat lifespan.

[0115] Figure 5 A flowchart illustrating a method for outputting a solution is shown, such as... Figure 5As shown, it includes the following steps.

[0116] Step S501: Obtain structured data during the maintenance process.

[0117] The structured data includes fault type, repair time, spare parts model, and technician feedback text.

[0118] Step S502: Filter target data that meets the predetermined data ratio in the structured data, and merge the target data with historical maintenance records.

[0119] Step S503: In response to triggering a new fault warning, and if the cosine similarity between the fault risk quantification report and the historical fault risk quantification report is greater than a preset similarity threshold, push the historical solution corresponding to the historical fault risk quantification report.

[0120] In this embodiment, by fusing structured data with historical records and introducing a cosine similarity algorithm to match new and old fault cases, validated repair solutions can be quickly reused. For example, when the similarity between a new fault and a historical case exceeds 80%, the system can directly push the corresponding solution, reducing the time technicians spend on repetitive analysis. This closed-loop feedback mechanism not only accelerates the fault repair process but also improves the system's adaptability to new fault modes through data-driven continuous optimization, providing reliable support for the iteration of long-term maintenance strategies.

[0121] This application also provides system embodiments that follow the above embodiments, for implementing the method steps of the above embodiments. The interpretation of the same names is the same as that of the above embodiments, and they have the same technical effects as those of the above embodiments, so they will not be repeated here.

[0122] like Figure 6 As shown, this application provides an automotive fault detection system 600, comprising:

[0123] The multimodal sensor network module 601 is used to collect operating status data of the car seat. The operating status data includes at least current signals, vibration signals, temperature signals and displacement signals.

[0124] The multimodal sensor network 601 includes:

[0125] A current sensor is used to collect the current signal when the seat motor is running.

[0126] Temperature sensors are used to collect temperature signals from the surface and interior of the seat.

[0127] A three-axis accelerometer is used to collect vibration signals from the seat rails.

[0128] Displacement sensors are used to collect displacement signals during seat adjustment.

[0129] The data processing module 602 is used to extract features and dynamically map the running status data to generate a seat status mapping matrix.

[0130] Data processing module 602 includes:

[0131] The feature extraction submodule is used to extract features from time series data and generate multidimensional feature vectors.

[0132] The state mapping submodule is used to dynamically map multi-dimensional feature vectors based on a preset threshold range, generating a seat state mapping matrix. This matrix contains the deviation ratio between the real-time value of each sensor signal and the preset threshold, expressed as:

[0133] .

[0134] in, i To label the time points, j For sensor labeling, Indicates the first i The first time point j The deviation ratio between the real-time value of each sensor signal and a preset threshold. Indicates the first i The first time point j Each sensor signal value, Indicates the first j The preset threshold upper limit for each sensor signal value.

[0135] The hybrid diagnostic model module 603 is used for fault identification based on the seat state mapping matrix, including:

[0136] The abnormal pattern recognition submodule is used to detect whether any deviation ratio in the state mapping matrix exceeds the ratio threshold. If it does, it is determined to be an abnormal pattern.

[0137] The trajectory prediction submodule is used to calculate the total offset of the seat adjustment path using a Bi-LSTM neural network, as shown in the formula:

[0138] .

[0139] in, k For sensor type labeling, n Number of sensor types This represents the total offset of the seat adjustment path. For the first k Weights of sensor-like signals, For the first k The deviation value of the sensor signal.

[0140] The fault warning and risk quantification module 604 is used to generate a fault risk quantification report based on the output of the hybrid diagnostic model.

[0141] The fault warning and risk quantification module 604 includes:

[0142] The risk level determination submodule is used to determine the current risk level by combining the offset distribution in the fault risk quantification report and historical maintenance data, and using a fuzzy logic algorithm.

[0143] The maintenance suggestion submodule is used to push maintenance suggestions based on risk level, including:

[0144] If the risk level is Level 1, an emergency stop will be triggered and a prompt will be made to replace the seat components immediately.

[0145] If the risk level is Level 2, it is recommended to replace the seat motor brushes.

[0146] If the risk level is level three, users are advised to regularly check the lubrication of the seat rails.

[0147] Dynamic update module 605 is used to dynamically optimize the state mapping matrix and the hybrid diagnostic model, including:

[0148] The sensor data preprocessing submodule is used to extract the fundamental frequency and its harmonics by performing wavelet transform on the current signal, extract the energy distribution of the specified frequency band by performing short-time Fourier transform on the vibration signal, and perform moving average filtering on the temperature signal.

[0149] The threshold adaptive submodule is used to dynamically adjust the temperature threshold range according to the seat's usage environment and the current threshold range according to the seat's age.

[0150] The weight update submodule is used to analyze the probability of occurrence of historical failure modes through clustering algorithms and dynamically update the weight allocation of the state mapping matrix.

[0151] The closed-loop feedback module 606 is used to synchronize fault warning results and maintenance records through the cloud database, and to match new faults with historical cases based on the cosine similarity algorithm. If the similarity is greater than a preset threshold, the corresponding historical solution will be pushed.

[0152] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0153] Although the operations are described in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all of the operations shown to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0154] The methods and systems of this application can be implemented using standard programming techniques, utilizing rule-based logic or other logic to implement various method steps. It should also be noted that the terms "system" and "module" as used herein and in the claims are intended to include implementations using one or more lines of software code and / or hardware implementations and / or devices for receiving input.

[0155] Any step, operation, or procedure described herein may be performed or implemented using one or more hardware or software modules, either alone or in combination with other devices. In one embodiment, the software module is implemented using a computer program product comprising a computer-readable medium containing computer program code, which is executable by a computer processor to perform any or all of the described steps, operations, or procedures.

[0156] The foregoing description of implementations of this application has been provided for illustrative and descriptive purposes. The foregoing description is not exhaustive and is not intended to limit this application to the exact forms disclosed. Various modifications and variations may exist in accordance with the foregoing teachings, or may arise from practice of this application. These embodiments were chosen and described to illustrate the principles of this application and its practical application, enabling those skilled in the art to utilize this application in various implementations and modifications to suit the specific purpose of the concept.

[0157] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0158] It can be further understood that, unless otherwise specified, "connection" includes both direct connections where no other components exist between the two parties and indirect connections where other components exist between them.

[0159] It is further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of this application, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all the operations shown to be performed to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0160] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the field of this application that are not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0161] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

[0162] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method of identifying a fault in a vehicle, characterized by, The method comprises the following steps: obtaining running state data of a car seat through a multi-modal sensor network; wherein the multi-modal sensor network comprises a current sensor, a temperature sensor, a three-axis accelerometer and a displacement sensor, and the running state data at least includes current signals, vibration signals, temperature signals and displacement signals; generating a seat state mapping matrix based on the running state data; inputting the seat state mapping matrix into a preset hybrid diagnostic model to perform fault identification; wherein the state mapping matrix contains the deviation proportion between the real-time value of each sensor signal and the preset threshold value; generating a fault risk quantification report according to the fault identification result output by the hybrid diagnostic model; wherein the hybrid diagnostic model performs fault identification in the following manner, comprising: in response to any deviation proportion in the state mapping matrix exceeding the proportion threshold value, determining the representation item of the deviation proportion as an abnormal mode; for the abnormal mode, determining the offset amount of the seat adjustment path through trajectory prediction by a Bi-LSTM neural network using the following formula: ； wherein, k is a sensor type label, n is a sensor type number, is a total offset of a seat adjustment path, is a first k class sensor signal weight, is a first k class sensor signal bias value; generating a fault analysis report according to the fault identification result output by the hybrid diagnostic model, comprising: in response to the total offset amount of the seat adjustment path exceeding a preset critical value, generating a fault risk quantification report.

2. The method of claim 1, wherein, The method further comprises: extracting features from the running state data in time series to generate a multi-dimensional feature vector; based on a preset threshold range, dynamically mapping the multi-dimensional feature vector to generate a seat state mapping matrix.

3. The method of claim 2, wherein, The deviation proportion is expressed by the following formula: ； in, i To label the time points, j For sensor labeling, Indicates the first i The first time point j The deviation ratio between the real-time value of each sensor signal and a preset threshold. Indicates the first i The first time point j Each sensor signal value, Indicates the first j The preset threshold upper limit for each sensor signal value.

4. The method of claim 1, wherein, The Bi-LSTM neural network is configured in the following manner: The input of the Bi-LSTM neural network includes historical data of the multi-modal sensor network, and the input data format is a time series tensor with dimensions of [T x n ], T is the number of time steps, n is the number of sensor types; The Bi-LSTM neural network contains 2 layers of LSTM units and 1 layer of fully connected layers, wherein each layer of LSTM units contains 64 nodes; the loss function is a weighted mean square error (WMSE) function; the initial weights of various sensor signals are allocated in the following manner: current signal weight 0.4; vibration signal weight 0.3; temperature signal weight 0.2; displacement signal weight 0.1; the output of the Bi-LSTM neural network is the total offset amount of the seat adjustment path.

5. The method of claim 1, wherein, The method further comprises: determining the current risk level using a fuzzy logic algorithm according to the offset amount distribution in the fault risk quantification report and combining historical maintenance data; in response to the current risk level being a first risk level, triggering an emergency shutdown and prompting immediate replacement of the seat components; in response to the current risk level being a second risk level, recommending replacement of the seat motor brush; in response to the current risk level being a third risk level, prompting the user to regularly check the lubrication of the seat slide rail.

6. The method of claim 1, wherein, The state mapping matrix is configured with a dynamic updating mechanism, comprising: wavelet transform is performed on the current signal to extract the fundamental frequency component and its harmonics; short-time Fourier transform is performed on the vibration signal to extract the energy distribution of the specified frequency band; moving average filtering is performed on the temperature signal; dynamically configuring the temperature threshold range according to the seat usage environment; dynamically configuring the current threshold range according to the seat usage time. The historical results of the fault early warning and the historical maintenance records are combined, a clustering algorithm is used to analyze the occurrence probability of different fault modes, and the weight distribution of the state mapping matrix is dynamically updated according to the occurrence probability.

7. The method according to claim 1 or 6, characterized in that, The method further comprises: acquiring structured data in a maintenance process, the structured data including fault types, maintenance time consumption, spare part models, and technician feedback texts; screening target data meeting a predetermined data proportion from the structured data, and fusing the target data with historical maintenance records; in response to triggering a new fault early warning, and a cosine similarity between the fault risk quantification report and a historical fault risk quantification report being greater than a preset similarity threshold, pushing a historical solution corresponding to the historical fault risk quantification report.

8. An automotive trouble detection system characterized by comprising: comprises: a multi-modal sensor network module configured to collect running state data of a vehicle seat, the running state data including at least current signals, vibration signals, temperature signals, and displacement signals; the multi-modal sensor network comprises: a current sensor configured to collect current signals when a seat motor is running; a temperature sensor configured to collect temperature signals of a seat surface and inside; a three-axis accelerometer configured to collect vibration signals of a seat slide rail; a displacement sensor configured to collect displacement signals in a seat adjustment process; a data processing module configured to perform feature extraction and dynamic mapping on the running state data to generate a seat state mapping matrix, wherein the state mapping matrix contains a deviation proportion of a real-time value of each sensor signal from a preset threshold value; a hybrid diagnostic model module configured to perform fault identification on the seat state mapping matrix, including: an abnormal pattern recognition submodule configured to detect whether any deviation proportion in the state mapping matrix exceeds a proportion threshold value, and determine an abnormal pattern if the deviation proportion exceeds the proportion threshold value; a trajectory prediction submodule configured to calculate a total offset of a seat adjustment path by a Bi-LSTM neural network, according to the following formula: ； wherein, k is a sensor type label, n is a sensor type number, is a total offset of a seat adjustment path, is a weight of a sensor signal of a first k class, is a bias value of a sensor signal of a first k class; a fault early warning and risk quantification module configured to generate a fault risk quantification report according to an output of the hybrid diagnostic model; wherein the hybrid diagnostic model performs fault identification in the following manner, including: in response to any deviation proportion in the state mapping matrix exceeding a proportion threshold value, determining a representation item of the deviation proportion as an abnormal pattern; for the abnormal pattern, performing trajectory prediction by a Bi-LSTM neural network according to the following formula to determine an offset of a seat adjustment path: ； wherein, k is a sensor type label, n is a sensor type number, is a total offset amount of a seat adjustment path, is a first k class sensor signal weight, is a first k class sensor signal bias value; the fault identification result output by the hybrid diagnostic model is used to generate a fault analysis report, including: in response to the total offset of the seat adjustment path exceeding a preset threshold value, generating a fault risk quantification report.

Images