Vehicle fault prediction method and device, and storage medium

Abstract

This application proposes a vehicle fault prediction method, device, and storage medium. The method includes: acquiring sound information collected by a sound sensor array on the vehicle; acquiring the propagation path and abnormal features of the sound information along the vehicle structure based on the sound information; acquiring vehicle operating condition information, and fusing the vehicle operating condition information, propagation path, and abnormal features to obtain fused features. The vehicle operating condition information includes at least one of vehicle speed information, motor operating information, battery information, and ambient temperature information. The fused features are input into a fault prediction model to obtain a vehicle fault prediction result, thereby achieving accurate prediction of vehicle faults and avoiding the occurrence of potential faults.

Description

Technical Field

[0001] This application relates to the field of vehicle fault detection technology, and in particular to vehicle fault prediction methods, devices and storage media. Background Technology

[0002] Vehicle malfunctions can easily lead to traffic accidents. Prior art often involves detecting malfunctions after they have occurred to determine their specific cause. However, since the malfunction has already happened and damage has already been incurred, detecting the malfunction after the fact is merely a reactive measure. Summary of the Invention

[0003] This application provides a vehicle fault prediction method, apparatus, and storage medium, which can be used to predict vehicle faults and accurately predict the location of potential vehicle faults. The technical solution is as follows: Firstly, a vehicle fault prediction method is provided, the method comprising: Acquire sound information collected by the sound sensor array on the vehicle; Based on sound information, the propagation path and abnormal features of sound information along the vehicle structure are obtained; The vehicle operating condition information is obtained, and the vehicle operating condition information, propagation path and abnormal features are fused to obtain fused features. The vehicle operating condition information includes at least one of the following: vehicle speed information, motor operating information, battery information and ambient temperature information. The fused features are input into the fault prediction model to obtain the vehicle fault prediction results.

[0004] Secondly, a vehicle fault prediction device is provided, the device comprising: The sound acquisition module is used to acquire sound information collected by the sound sensor array on the vehicle; The first processing module is used to obtain the propagation path of sound information along the vehicle structure and abnormal features based on sound information; The fusion module is used to acquire vehicle operating condition information and fuse the vehicle operating condition information, propagation path and abnormal features to obtain fused features. The vehicle operating condition information includes at least one of the following: vehicle speed information, motor operating information, battery information and ambient temperature information. The fault prediction module is used to input fused features into the fault prediction model to obtain vehicle fault prediction results.

[0005] Thirdly, a computer program product is also provided, which includes computer instructions that, when executed by a processor, implement the steps of any of the vehicle fault prediction methods described above.

[0006] The technical solution provided in this application brings at least the following beneficial effects: This application acquires sound information collected by a sound sensor array on a vehicle; based on the sound information, it determines the propagation path and abnormal features of the sound information along the vehicle structure; it acquires vehicle operating condition information, and fuses the vehicle operating condition information, propagation path, and abnormal features to obtain fused features. These fused features are then input into a fault prediction model to obtain vehicle fault prediction results. This application determines the propagation path using sound information collected by a sound sensor array, and accurately locates abnormal sound sources based on the propagation path. Since the sound emitted by the abnormal sound source can characterize the vehicle's condition, that is, the propagation path and abnormal features can characterize the vehicle's condition, a multi-source fused feature is constructed by combining vehicle operating condition information, propagation path, and abnormal features. Based on this multi-source fused feature, vehicle fault prediction is performed, thereby accurately predicting possible faults. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is one of the schematic diagrams of an implementation environment provided in the embodiments of this application; Figure 2 This is a second schematic diagram of an implementation environment provided in the embodiments of this application; Figure 3 This is a third schematic diagram of an implementation environment provided in the embodiments of this application; Figure 4 This is one of the flowcharts of a vehicle fault prediction method provided in the embodiments of this application; Figure 5 This is a second flowchart of a vehicle fault prediction method provided in the embodiments of this application; Figure 6 This is the third flowchart of a vehicle fault prediction method provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of a vehicle fault prediction device provided in an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0010] Please refer to Figures 1-3 The diagram illustrates the implementation environment of the method provided in the embodiments of this application.

[0011] The implementation environment may include: at least one vehicle 11 and a cloud server 12, wherein the cloud server 12 is communicatively connected to the controller of at least one vehicle 11.

[0012] Vehicle 11 includes an array of sound sensors, a controller, and an ECU (Electronic Control Unit).

[0013] The sound sensor array is electrically connected to the controller. The sound sensor array is installed on the vehicle 11 and is used to collect sound information and send the collected sound information to the controller.

[0014] The sound sensor array comprises multiple sound sensors. The placement of these sensors on the vehicle is determined based on the characteristics of the sounds emitted by vehicle components and the vehicle's structure, ensuring coverage of all potential sources of abnormal noise. Types of sound sensors include external and internal sensors. External sensors are positioned on locations such as the fenders, bumper shock absorber towers, subframe, and / or engine compartment. Multiple external sensors can be symmetrically distributed on both sides of the vehicle to capture sound information caused by structural vibrations. Internal sensors can be placed in the four corners of the cabin, such as the four corners inside the headliner, to capture sound information caused by vehicle structural vibrations.

[0015] The controller is used to acquire sound information collected by the sound sensor array, analyze the sound information to determine if there are any anomalies, and, in the event of anomalies, obtain the propagation path of the sound information along the vehicle structure and the characteristics of the anomaly. For example, the controller is a CSC (Core Super Computer) module, which is a type of vehicle controller.

[0016] The controller connects to the ECU to acquire vehicle operating condition information sent by the ECU. Based on this information, the propagation path, and abnormal characteristics, it generates fused features, and then uses these fused features to predict vehicle faults. The controller also deploys a fault prediction model, which receives the fused features to obtain the predicted vehicle faults.

[0017] In some embodiments, the vehicle 11 further includes an on-board terminal for receiving and displaying vehicle fault prediction results so that users can view the vehicle fault prediction results.

[0018] Cloud server 12 communicates with controllers of multiple vehicles to obtain vehicle operating condition information and sound information sent by each controller, and trains a fault prediction model based on the vehicle operating condition information and sound information. The trained fault prediction model is then sent to each controller so that each controller can deploy the fault prediction model on the vehicle.

[0019] The implementation environment may also include: a user terminal. The user terminal is connected to the controller and is used to receive and display the vehicle fault prediction results sent by the controller.

[0020] Based on the above Figures 1-3 The implementation environment shown in this application provides a vehicle fault prediction method, such as... Figure 4 As shown, taking the application of this method to a controller as an example, the method includes steps 201-204.

[0021] Step 201: Acquire the sound information collected by the sound sensor array on the vehicle.

[0022] The sound sensor array includes multiple sound sensors, and the placement of these sensors on the vehicle can be determined based on the characteristics of the sounds emitted by the vehicle's components and the vehicle's structure to ensure coverage of the locations of most potential sound sources.

[0023] Step 202: If there is an anomaly in the sound information, then based on the sound information, obtain the propagation path of the sound information along the vehicle structure and the anomaly characteristics.

[0024] In this embodiment of the application, it can be determined whether there is an abnormality in the sound information by comparing the sound information with normal sound information.

[0025] Specifically, time-domain analysis, frequency-domain analysis, and time-frequency analysis are performed on the sound information to obtain its sound features. After obtaining the sound features, these features are compared with reference sound features of normal sound information to determine the similarity between the sound features and the reference sound features. This determines whether there are any anomalies in the sound features of the sound information. If the similarity is less than a similarity threshold, an anomaly is found, and the sound information is considered abnormal. The similarity threshold is set based on experience.

[0026] It's understandable that a vehicle doesn't suddenly malfunction; rather, it gradually evolves from a normal state to an abnormal state, and then from the abnormal state to a malfunction. For example, with brake pads, as long as the braking distance they can support doesn't exceed safety standards, brake pads that have been used for a period of time are not considered to be in a malfunctioning state. However, generally speaking, the performance of brake pads that have been used for a period of time is not as good as that of unused brake pads. For instance, the sound emitted by brake pads that have been used for a period of time when braking is different from the sound emitted by unused brake pads. Therefore, by setting a reasonable similarity threshold, it is possible to identify whether the sound information is abnormal.

[0027] The embodiments of this application can determine whether the sound information is abnormal, which is essentially to determine whether the vehicle is in an abnormal state, so as to predict possible future faults even if no fault has occurred.

[0028] The process involves performing time-domain, frequency-domain, and time-frequency analysis on sound information. This includes: performing time-domain analysis to obtain the time-domain characteristics of the sound information, including peak value, mean, variance, and kurtosis; converting the signal to the frequency domain using Fast Fourier Transform (FFT) and performing frequency-domain analysis to obtain frequency-domain characteristics, including the frequency components and power spectral density of the sound information in the frequency domain. Frequency components refer to the set of different frequency components contained in the sound information and their amplitude and phase characteristics; and using algorithms such as STFT and Wigner-Ville distribution to analyze the time-frequency characteristics of the signal and capture unstable / time-varying transient abnormal sounds.

[0029] refer to Figure 5 In some possible embodiments, based on sound information, the propagation path of sound information along the vehicle structure and abnormal features are obtained, including: steps 301-303.

[0030] Step 301: Obtain the vehicle's dynamic stiffness dataset. The dynamic stiffness dataset includes the dynamic stiffness curve of the reference propagation path. The dynamic stiffness curve is used to characterize the change of sound under the action of the dynamic stiffness of the reference propagation path when the sound is transmitted along the reference propagation path.

[0031] The reference propagation path is a preset sound propagation route, used to characterize the vehicle structure through which sound passes as it propagates along the vehicle structure corresponding to the preset sound propagation route. For example, if the reference propagation path is the path from the motor to sound sensor A, then the sound emitted by the motor will be transmitted to sound sensor A and collected by sound sensor A. Similarly, if the reference propagation path is the path from the chassis area to sound sensor B, then the sound emitted by the chassis will be transmitted to sound sensor B and collected by sound sensor B.

[0032] Dynamic stiffness is the ability of a structure or material to resist deformation under dynamic loads.

[0033] Dynamic stiffness curves characterize how sound changes as it propagates along a reference propagation path, influenced by the dynamic stiffness of that path. In other words, the dynamic stiffness dataset represents the attenuation and phase delay of sound as it propagates along the reference propagation path, influenced by the dynamic stiffness of the vehicle structure along that path.

[0034] The dynamic stiffness dataset includes dynamic stiffness curves for at least one reference propagation path.

[0035] In this embodiment, a dynamic stiffness dataset can be obtained based on a vehicle structural dynamics model. Since the vehicle structural dynamics model is a simulation of the vehicle structure, it reflects the variation of sound information under dynamic stiffness as the sound propagates along different reference propagation paths to various sound sensors when each structure of the vehicle acts as a sound source. Therefore, a dynamic stiffness dataset can be obtained from the vehicle structural dynamics model. For example, when a motor operates, it emits sound. If the sound source is the motor and the reference propagation path is from the motor to sound sensor A, the vehicle structural dynamics model can be used to obtain the variation of the sound emitted by the motor under dynamic stiffness as the sound propagates along the reference propagation path.

[0036] In this embodiment, a 3D digital model of the vehicle can be constructed based on finite element analysis software (such as ANSYS, Abagus, etc.), thereby constructing a vehicle structural model. The vehicle structural model includes the material properties (such as material, density, and elastic modulus), fixing conditions (such as weld stiffness, bolt preload, etc.), and boundary conditions (such as suspension constraints, intersecting bushing stiffness, etc.) of each component on the vehicle. Based on the vehicle structural model, the natural frequencies of each component and the mode shape distribution under specified excitation are calculated. Simultaneously, modal experiments are conducted, applying excitation to measurement points (such as motor mounting bases, chassis crossbeams, and body pillars) on the test vehicle in a laboratory test bench, and acquiring vibration response signals collected by the sound sensor array on the test vehicle. By comparing the differences between the vibration response signals and the natural frequencies and mode shapes of the vehicle structural model, the vehicle structural model is adjusted to obtain the vehicle structural dynamic model.

[0037] Step 302: Correct the sound information based on the dynamic stiffness dataset to obtain corrected sound information.

[0038] Due to the influence of the vehicle's structural dynamic stiffness, the sound information collected by the sound sensor array differs from the sound emitted by the sound source. Therefore, the sound information can be corrected using the dynamic stiffness dataset to obtain corrected sound information. By correcting the sound information, the location of vehicle faults can be predicted more accurately.

[0039] In some possible embodiments, the sound information is corrected based on a dynamic stiffness dataset to obtain corrected sound information, including: Preprocessing is performed on the audio information to obtain the target audio information; The target sound information is corrected based on the dynamic stiffness dataset to obtain the corrected sound information.

[0040] Preprocessing operations include: performing signal synchronization and noise reduction on the audio information to obtain the target audio information.

[0041] Signal synchronization is used to ensure that the first sound information collected by different sound sensors is synchronized in time for accurate analysis.

[0042] Noise reduction processing utilizes Gaussian filters and digital filtering algorithms (such as Kalman filtering and wavelet denoising) to remove background noise and interference signals from audio signals. Anomalies in audio information often occur within a preset frequency range. Noise reduction processing can eliminate sounds outside this range and remove background noise and interference signals within it. Therefore, removing background noise and interference signals from audio information is highly significant. For example, the frictional impact of motor bearing wear is essentially the contact frequency between the rolling elements and the inner and outer rings. Based on the bearing's structural parameters (number of balls, diameter), the abnormal wear of the motor bearing is determined to occur in the 200-500Hz range. Thus, the preset frequency range for abnormal motor bearing wear is 200Hz-500Hz. During noise reduction processing, signals within this range will be retained, while sounds outside this range will be automatically removed, along with background noise and interference signals within the 200Hz-500Hz range.

[0043] The preset frequency range is determined based on historical data. Specifically, it is determined through experimental / engineering experience. For example, if historical data determines that before the motor bearing failed, the signal amplitude of the sound emitted by the motor bearing was high in a certain frequency range A, then frequency range A can be determined as the preset frequency range. If, before the motor bearing failed, there was abnormal noise from the motor bearing, and the amplitude of the abnormal noise in the 200-500Hz range was more than three times that of other ranges, then that frequency range is the preset frequency range.

[0044] Specifically, when performing signal synchronization and noise reduction on sound information, time synchronization can be performed first to obtain synchronized sound information, and then noise reduction can be used to retain the signals corresponding to each first preset frequency in the synchronized sound information to obtain the target sound information.

[0045] Correcting sound information includes correcting the amplitude and phase of the sound information.

[0046] For amplitude correction of sound information, for example, if the sound source is a motor, the reference propagation path is from the motor to sound sensor A, and the amplitude of the sound information collected by sound sensor A is 5dB, and the information attenuation of the sound information collected by sound sensor A in the reference propagation path is determined to be 3dB according to the dynamic stiffness dataset, then the amplitude of the information emitted by the sound source is 8dB (5dB+3dB=8dB). Similarly, if the sound source is a motor, the reference propagation path is from the motor to sound sensor B, and the amplitude of the sound information collected by sound sensor B is 4dB, and the information attenuation of the sound information collected by sound sensor B in the reference propagation path is determined to be 2dB according to the dynamic stiffness dataset, then the amplitude of the information emitted by the sound source is 6dB (4dB+2dB=6dB).

[0047] For phase correction of sound information, the sound source is a motor. One reference propagation path is from the motor to sound sensor A, and the other reference propagation path is from the motor to sound sensor B. Theoretically, sound sensor A and sound sensor B receive the information simultaneously. However, due to the delay caused by the dynamic stiffness of the reference propagation path, the time when sound sensor B collects the sound information is 0.001 seconds later than the time when sound sensor A collects the sound information. The dynamic stiffness dataset determines that this delay is caused by "transmission" rather than "distance". Therefore, the sound information is corrected by adjusting the phase of the sound information collected by sound sensor B forward by 0.001 seconds, thereby restoring the true time when the sound information arrives at the two measuring points.

[0048] By correcting the sound information, distortions caused by the propagation path are avoided from being mistaken for sound source location, thus preventing inaccurate sound source localization.

[0049] It is understandable that the sound information includes the first sound information collected by each sound sensor, and the corrected sound information is obtained after correcting the sound information. Therefore, correcting the sound information means correcting each first sound information, and the corrected sound information includes the second sound information obtained after correcting the first sound information.

[0050] Step 303: Based on the corrected sound information, generate the propagation path and abnormal features.

[0051] Time-domain analysis, frequency-domain analysis, and time-frequency analysis are performed on the corrected audio information to obtain its audio features. Based on the correlation between the elements in the audio features of the corrected audio information and the fault, elements with high correlation with the fault are extracted from the audio features of the corrected audio information to generate abnormal features.

[0052] The process of obtaining the sound features of the corrected sound information can refer to the process of obtaining the sound features of the sound information described above, and will not be repeated here.

[0053] The correlation between elements in the sound characteristics of corrected sound information and faults is determined by analyzing historical data.

[0054] For example, historical data revealed that in 90% of motor bearing failure cases, the amplitude of the 200-500Hz frequency band and the temporal kurtosis of the corrected sound information were significantly increased, while the peak value of the corrected sound information only changed in 60% of motor bearing failure cases. Thus, it was determined that the abnormal features with high amplitude correlation in motor bearing failure included amplitude and temporal kurtosis.

[0055] The correlation between elements and faults in the sound features of corrected sound information includes correlation and mutual information. Correlation focuses on linear relationships, while mutual information can capture any type of dependency. Preferably, mutual information focuses on nonlinear relationships. For example, the frequency shift of sound features and the abnormal noise fault of reducer gears are not linearly related, but based on mutual information, it can be determined that there is a correlation between the frequency shift of sound features and the abnormal noise fault of reducer gears.

[0056] In this embodiment, a correlation threshold is set. If the correlation between an element of the sound feature and the fault is greater than the correlation threshold, the element is extracted to generate an abnormal feature.

[0057] In this embodiment of the application, redundant elements are also excluded when generating abnormal features. For example, if two elements in the sound features are highly correlated with the fault and are essentially the same, one of the elements will be selected as the basis for generating the abnormal feature to avoid repeated calculations and ultimately form a highly correlated abnormal feature. For example, for the elements "amplitude in the 200Hz-500Hz frequency band" and "energy proportion in the 200Hz-500Hz frequency band", the abnormal feature = {amplitude in the 200Hz-500Hz frequency band, time domain kurtosis} will be generated based on "amplitude in the 200Hz-500Hz frequency band".

[0058] The correlation threshold can be adjusted as needed. In some possible embodiments, the correlation threshold can be continuously optimized based on vehicle usage. For example, if it is found that "under low-temperature conditions, abnormal noise from the motor bearing is accompanied by a shift in the {phase} element in the sound characteristics," the correlation threshold corresponding to the correlation between the {phase} element and the fault will be adjusted to include the {phase} element in the abnormal features, making the diagnosis more accurate. In some possible embodiments, the correlation threshold may differ between different vehicles. For example, motors of the same model but from different batches may have slight differences. Therefore, the controller can correct the correlation threshold, such as a 30% increase in kurtosis for vehicle A being considered highly correlated, and a 25% increase for vehicle B.

[0059] refer to Figure 6 In some possible embodiments, the propagation path is obtained through the following steps 401-403: Step 401: Based on the energy changes of the corrected sound information, determine the candidate propagation path.

[0060] Due to the positional relationship of the sound sensors, there are energy variations between the first sound information collected by each sensor. Corrected sound information is obtained after correcting the original sound information; therefore, there are also energy variations between the second sound information within the corrected sound information. These energy variations can reflect the propagation path of the sound information. In other words, the energy variation of the corrected sound information is the same as the energy variation between the second sound information. If the energy attenuation of sound information along the propagation path corresponding to certain second sound information is small, it indicates that the sound emitted by the sound source is transmitted more smoothly along that propagation path, and the vibration energy transmission efficiency is higher. This allows for the identification of several high-weight propagation paths, resulting in candidate propagation paths. Obtaining candidate propagation paths is equivalent to initially determining the location of the sound source. That is, if the vibration energy transmission efficiency of a certain propagation path is high, then that propagation path is a candidate propagation path, and the starting point of the candidate propagation path is the location of the sound source.

[0061] For example, if the energy transfer efficiency of the propagation paths “motor → bracket (sound sensor B at the bracket) → sound sensor A” and “motor → housing (sound sensor D at the bracket) → sound sensor C” is higher than that of other propagation paths, based on the energy changes of each second sound information, then the candidate propagation paths are “motor → bracket → sound sensor A” and “motor → housing → sound sensor C”.

[0062] Step 402: Input the corrected sound information into the sound source localization model to determine the candidate location information of the sound source.

[0063] The corrected sound information is input into the TDOA / beamforming algorithm. The TDOA algorithm calculates the candidate location information of the sound source based on the actual time difference between the arrival of the corrected sound information at the sound sensor. The beamforming algorithm focuses the sound source direction based on the energy distribution of each corrected sound information, thereby calculating the candidate location information of the sound source. Since the signal has been corrected, the TDOA / beamforming algorithm can directly calculate the spatial coordinate range of the candidate sound source location, i.e., the candidate sound source location information. This candidate sound source location information is used to characterize the candidate sound source location. For example, the candidate sound source location information characterizes the candidate sound source location as being at the left bearing position of the motor, with an error range of ±10cm.

[0064] Step 403: Generate a propagation path based on the comparison results between the candidate propagation path and the candidate location information of the sound source.

[0065] Candidate propagation paths were determined by correcting the energy changes of sound information under dynamic stiffness. Candidate sound source locations were calculated using a sound source localization model. By comparing the candidate propagation paths with the candidate sound source locations, it was determined whether the candidate sound source locations were the starting point of the candidate propagation paths. If they were, the candidate propagation paths were adopted as the propagation paths, and the starting point of the propagation paths was designated as the sound source locations.

[0066] Step 203: Obtain vehicle operating condition information, and fuse the vehicle operating condition information, propagation path and abnormal features to obtain fused features. The vehicle operating condition information includes at least one of the following: vehicle speed information, motor operating information, battery information and ambient temperature information.

[0067] The time for collecting vehicle operating condition information and the time for collecting audio information can be the same or different. It can collect audio information and vehicle operating condition information from a vehicle in motion, or from a vehicle at rest, or vice versa. During audio information collection, vehicle operating condition information can be collected simultaneously, or audio information can be collected first, followed by vehicle operating condition information, or vice versa, or the time difference between collecting audio information and collecting vehicle operating condition information is less than a time threshold.

[0068] In one possible embodiment, the vehicle operating information includes at least one of the following: vehicle speed information, motor operating information, battery information, ambient temperature information, operation information, and acceleration information.

[0069] In one possible embodiment, the vehicle operating condition information, propagation path, and abnormal features are weighted and fused to obtain a fused feature. For example, the weight of the propagation path is 0.4, the weight of the abnormal features is 0.3, and the weight of the vehicle operating condition information is 0.3.

[0070] Vehicle operating condition information can interfere with fault prediction. Therefore, this application fuses vehicle operating condition information, propagation paths, and abnormal features to obtain fused features, thereby improving the accuracy of fault prediction.

[0071] Vehicle speed information reflects the operating status of components on the vehicle that affect its speed. This information influences the propagation of sound information in terms of frequency, amplitude, and transmission efficiency. For example, regarding frequency: the frequency of sound emitted by rotating components such as tires, wheel bearings, drive shafts, and reducer gears is positively correlated with vehicle speed; regarding amplitude: background noise such as tire noise and wind noise increases with the square of vehicle speed (e.g., background noise at 100 km / h is about 10 dB higher than at 60 km / h), potentially masking low-amplitude abnormal noises (e.g., slight bolt loosening) or amplifying airflow-related abnormal noises (e.g., wind noise from poor door sealing); regarding transmission efficiency: increased vehicle speed activates body vibration modes (e.g., body resonance at 120 km / h), improving the transmission efficiency of specific abnormal noises (e.g., chassis suspension noise), resulting in "no abnormalities at low speeds, but significant abnormal noises at high speeds." For instance, at speeds of 50-80 km / h, it is easier to detect abnormalities such as loose body / chassis components or interior trim clips making noise.

[0072] At the same time, as vehicle speed increases, the probability of malfunctions also increases. For example, if the vehicle speed increases, the wheel hub bearing speed increases, the rolling element load increases, and the wheel hub bearing impact noise intensifies. Also, if the vehicle speed increases, the reducer gear speed increases, the reducer gear tooth surface friction / impact frequency increases, and the probability of reducer gear tooth failure increases sharply.

[0073] Similar to vehicle speed, motor operating information and battery information reflect the operating status of components on the vehicle that affect vehicle speed. Motor operating information and battery information can also interfere with fault prediction.

[0074] Ambient temperature affects the physical properties and operating status of components, thereby altering the probability, amplitude, and frequency of abnormal noises.

[0075] The influence of ambient temperature causes changes in the physical properties of materials in a vehicle. For example, changes in fit clearances can occur. At temperatures as low as -10°C, rubber parts (such as suspension bushings and door seals) harden and shrink, increasing the fit clearances and making them more prone to squeaking or clicking sounds. Similarly, compared to normal temperatures, metal parts (such as body welds and chassis bolts) shrink at low temperatures (below -10°C), causing loosening of connections and increased vibration and noise. Furthermore, compared to normal temperatures, metal parts (such as engine blocks, chassis beams, and braking mechanisms) expand at high temperatures (above 35°C), reducing fit clearances (e.g., the gap between brake discs and brake pads), making them more prone to noise. Finally, the influence of ambient temperature can cause plastic parts (such as interior trim panels and battery pack casings) to deform, leading to loosening of clips and increased probability, amplitude, and frequency of resonance noises.

[0076] For example, when performance parameters change, such as at low temperatures, the viscosity of the grease in bearings and gearboxes increases, the grease flow becomes worse, resulting in insufficient lubrication, increased friction between components, and an increased probability and amplitude of sound. At high temperatures, the grease ages and its viscosity decreases, leading to lubrication failure, increased friction between components, an increased probability of sound, and a wider frequency range and a sharp increase in amplitude. Similarly, at low temperatures, the battery pack cooling system starts frequently, increasing fan speed and the probability of cooling fan failure, making the sounds emitted by the cooling fan (such as frequency and amplitude) more easily perceived. At high temperatures, the battery pack thermal management system is overloaded, causing pipe vibration and a higher probability of failure, easily producing a "humming" sound.

[0077] It is evident that vehicle operating condition information affects the time domain (such as amplitude, pulse interval, etc.) and time frequency (such as characteristic frequency shift) of abnormal features. In the embodiments of this application, in addition to analyzing the propagation path and abnormal features, vehicle operating condition information is also introduced when predicting vehicle faults, thereby accurately predicting vehicle faults.

[0078] Step 204: Input the fused features into the fault prediction model to obtain the vehicle fault prediction results.

[0079] The fault prediction model takes fused features as input and aims to learn the mapping relationship between the fused features and possible future faults. Based on the mapping relationship learned by the model and the fused features before the fault occurs, it outputs possible future faults to obtain vehicle fault prediction results.

[0080] The vehicle fault prediction results include the fault type, location, severity / time, probability of occurrence, and fault level of possible future faults.

[0081] For example, if the vehicle operating condition information represents an average brake pad wear of 8.2 mm in 1 hour, and the propagation path and abnormal features represent abnormal sounds corresponding to the brake pads, then the fused features of the vehicle operating condition information, propagation path, and abnormal features are input into the fault prediction model, and the structured vehicle fault prediction result is output as follows: {Predicted fault type: brake pad wear, location: left rear brake pad, degree / time: the brake pad will cause the braking distance to exceed the safe distance after working for another 24 hours, probability of fault occurrence: the probability of the brake pad causing the braking distance to exceed the safe distance after working for another 24 hours is 90%, fault level: high}.

[0082] In some possible embodiments, the vehicle fault prediction results also include a confidence level, which reflects the reliability and accuracy of the vehicle fault prediction results.

[0083] For example, the structured vehicle fault prediction result is {Predicted fault type: cooling fan failure of battery pack, location: cooling fan next to battery pack, severity / time: abnormal noise of cooling fan is 45% higher than normal, probability of failure: probability of cooling fan failure is 90%, fault level: medium, confidence level: 95%}.

[0084] In some possible embodiments, the fused features are input into the fault prediction model to obtain vehicle fault prediction results, including: Acquire historical fusion features of the vehicle; By inputting historical fusion features and fusion features into the fault prediction model, vehicle fault prediction results are obtained.

[0085] Historical fusion characteristics are the fusion features of a vehicle's past performance. Both historical fusion characteristics and fusion features reflect the developmental patterns of a vehicle's condition, allowing for the prediction of vehicle faults based on these patterns. For example, if a developmental pattern indicates a slow increase in the amplitude and frequency shift of sound and vibration signals, although no fault has yet formed, a trend towards fault formation is emerging. Therefore, potential faults can be predicted based on historical fusion characteristics and fusion features.

[0086] In some possible embodiments, the fused features are input into the fault prediction model to obtain vehicle fault prediction results, including: Acquire historical fusion characteristics and historical usage characteristics of vehicles; By inputting the vehicle's historical usage characteristics, historical fusion characteristics, and fusion characteristics into the fault prediction model, the vehicle fault prediction results are obtained.

[0087] Vehicle historical usage characteristics characterize the usage of vehicle components throughout their life cycle. These characteristics include component aging data, which includes mileage, motor operating time, and battery charge cycle count.

[0088] Based on historical integration characteristics and integration characteristics, further incorporating vehicle historical usage characteristics can provide a more accurate understanding of vehicle condition development patterns and more accurate prediction of vehicle malfunctions.

[0089] Historical fusion characteristics and vehicle historical usage characteristics are obtained based on historical data.

[0090] In some possible embodiments, the method further includes performing uncertainty quantification on the features input to the fault prediction model before inputting the fused features into the fault prediction model.

[0091] Specifically, a confidence level is assigned to each input feature. For example, when the sensor signal is affected by noise, the confidence level is set to (0.7-0.9). The 3σ criterion is used to remove outliers of the input features and to perform interval processing on fuzzy features. For example, low-amplitude outliers are fuzzed to determine low-amplitude and high-amplitude labels. If the amplitude of the outlier is ∈ [5-10dB], then the amplitude of the outlier is fuzzed to a low-amplitude label.

[0092] In one possible embodiment, the fault prediction model takes historical fusion features and fusion features as inputs and aims to learn the mapping relationship between the inputs and possible future faults.

[0093] In another possible embodiment, the fault prediction model takes vehicle historical usage characteristics, historical fusion characteristics, and fusion characteristics as inputs, and learns the mapping relationship between the inputs and possible future faults as its learning objective.

[0094] In this embodiment, by using the fused features prior to the occurrence of a fault as input to the fault prediction model and quantifying the uncertainty of the features input to the fault prediction model, high-quality trend data is provided to the fault prediction model, thereby accurately predicting possible vehicle faults.

[0095] The fault prediction model is obtained by training the initial fault prediction model with fault sample data and sample fusion features. The initial fault prediction model includes prior knowledge.

[0096] Prior knowledge is derived from automotive engineering experience and characterizes the probability of future failures of vehicle components under different operating conditions. For example, prior knowledge characterizes the probability of bearing wear when the motor speed is lower than a preset speed, and the probability of bearing wear failure when the motor speed is higher than 10,000 rpm is 30% higher than the probability of bearing wear when the motor speed is lower than the preset speed.

[0097] The fault sample data is determined based on fault data observed on the vehicle. This fault data includes: motor bearing wear fault data, chassis bolt loosening fault data, battery pack cooling fan abnormal noise fault data, and reducer gear meshing abnormality fault data.

[0098] The fault prediction model is a Bayesian network (BN) or a hidden Markov model.

[0099] When the fault prediction model is a Bayesian network, the parent node of the fault prediction model is the fusion feature, and the child nodes are the types of faults that the vehicle may fail in the future and their probabilities. The Conditional Probability Table (CPT) is obtained by training with historical data.

[0100] When the fault prediction model is a Hidden Markov Model (HMM), the hidden state set Q of the HMM represents the type, location, and severity / time of potential future vehicle faults. The hidden state set Q contains all possible hidden states q. i That is, Q = {q1, q2, ..., q} n The set of observation symbols V in a Hidden Markov Model is a fused feature, i.e., the set of observation symbols V = {v1, v2, ..., v...} m The elements in the observation symbol set V include time-domain features (such as peak value, kurtosis, RMS value, etc.), time-frequency features (such as wavelet coefficients, characteristic frequency range, etc.), vehicle operating condition information, and propagation path. For example, the observation symbol set V = {kurtosis ∈ [3, 5], kurtosis ∈ [5, 8], amplitude proportion of 200-500Hz ≥ 50%, peak value ≥ 10dB}; the state transition matrix A of the Hidden Markov Model is the fault development probability, generated based on historical fault data in the cloud. The observation probability matrix B of the Hidden Markov Model is an n×m matrix, and the elements of the observation probability matrix B are b. ij b ij =P(O j =v j |St=qi), representing the system in the hidden state q. i At that time, the observation value v is generated. j The probability of the observed probability matrix B is the cloud server's "fault state (q)" based on all vehicles of the same model / platform. i → Observational features (v) j The data is generated corresponding to ")"; the initial state distribution π of the Hidden Markov Model is generated based on prior knowledge.

[0101] In some possible embodiments, after obtaining the vehicle fault prediction result, the method further includes: Send the vehicle fault prediction results to the terminal; The vehicle fault prediction results are displayed on the terminal.

[0102] In this embodiment of the application, by actively pushing vehicle fault prediction results to the vehicle terminal / user terminal and displaying the vehicle fault prediction results on the vehicle terminal / user terminal, users can be made aware of possible vehicle faults in a timely manner.

[0103] In some other possible embodiments, after obtaining the vehicle fault prediction result, the method further includes: If the probability of a fault occurring is higher than the probability of a fault in the vehicle fault prediction results, a repair suggestion is generated. If the probability of a fault occurring is less than or equal to the probability of a fault in the vehicle fault prediction results, a prompt message is generated.

[0104] For example, if the probability of failure is 70%, a repair suggestion is generated when the probability of failure is greater than 70% to suggest that the user select the vehicle for repair. When the probability of failure is less than or equal to 70% (the probability of failure is 30%-60%), a prompt message is generated to indicate the existence of a potential failure, so as to continuously monitor based on the prompt message.

[0105] In some other possible embodiments, after obtaining the vehicle fault prediction result, the method further includes: Based on the vehicle fault prediction results, maintenance suggestions are generated; The terminal displays vehicle fault prediction results and maintenance suggestions.

[0106] In some possible embodiments, maintenance suggestions are generated based on vehicle fault prediction results, including: extracting relevant maintenance information from a preset knowledge base based on the vehicle fault prediction results, the maintenance information including: maintenance part model, maintenance part supplier information, maintenance steps, maintenance price, maintenance provider information and precautions, and generating maintenance suggestions based on the maintenance information.

[0107] In some possible embodiments, maintenance recommendations are generated based on maintenance information, including: The vehicle operating condition information, vehicle fault prediction results, and maintenance information are input into the evaluation model, and the evaluation results are output. The evaluation results include maintenance costs and the safety level corresponding to the vehicle fault prediction results. The evaluation results are used to evaluate the benefits of maintenance costs and safety levels. Based on maintenance information and assessment results, maintenance recommendations are generated.

[0108] Since this application embodiment predicts vehicle malfunctions, and the actual malfunction has not yet occurred, the essence of repairing based on the predicted malfunction is to reduce the probability of the malfunction occurring. Therefore, to balance the repair cost and the benefit of reducing the probability of malfunction, the vehicle malfunction prediction result and repair information are input into an evaluation model, which outputs an evaluation result and generates a repair suggestion based on the evaluation result. This allows users to assess the repair cost and safety level based on the repair suggestion. For example, the vehicle operating condition information and the vehicle malfunction prediction result are represented as "current mileage 30,000 km, battery cooling fan failure probability 20%", the generated evaluation result is "current safety level is high, repair cost will be high if repair is performed", and the generated repair suggestion is "it is recommended to drive another 5,000 km before repairing the battery cooling fan to avoid increased costs due to premature repair".

[0109] In some possible embodiments, the terminal displays vehicle fault prediction results and repair suggestions, including: Obtain vehicle maintenance records; Based on vehicle maintenance records, the target maintenance time is obtained, which is the next maintenance time. Based on the target maintenance schedule, the terminal displays vehicle fault prediction results and maintenance suggestions.

[0110] In some possible embodiments, the terminal displays vehicle fault prediction results and repair suggestions, including: Obtain vehicle start information; When the vehicle start-up information indicates that the vehicle has started, the terminal displays vehicle fault prediction results and maintenance suggestions.

[0111] In some possible embodiments, the terminal displays vehicle fault prediction results and repair suggestions, including: On the user terminal, an app displays vehicle fault prediction results and repair suggestions in a structured manner, along with corresponding operation controls for each suggestion, allowing users to select their preferred repair method via touch. In this embodiment, after the user selects a repair method, the vehicle fault prediction results and the user-confirmed repair method are sent to a cloud server. The cloud server then forwards these results to the repair service provider to reduce repair waiting time. The operation controls include a repair parts model selection control and a repair service provider selection control. The repair parts model selection control is used to select the type of repair parts, and the repair service provider selection control is used to select the repair service provider.

[0112] In some possible embodiments, the terminal displays vehicle fault prediction results and repair suggestions, including: A vehicle fault prediction report is generated based on the vehicle fault prediction results and maintenance suggestions. The terminal displays a vehicle fault prediction report.

[0113] In some possible embodiments, after obtaining the vehicle fault prediction result, the method further includes: If the fault type in the vehicle fault prediction result is the first fault type, the vehicle fault prediction result and the first maintenance suggestion are displayed. The first fault type is a fault that does not trigger the high-voltage system.

[0114] In this embodiment of the application, when the fault type of the vehicle fault prediction result is not to trigger the safety of the high-voltage system, the vehicle fault prediction result and the first maintenance suggestion are actively pushed to the vehicle terminal / user terminal, and the vehicle fault prediction result and the first maintenance suggestion are displayed on the vehicle terminal / user terminal. This is to guide the user to quickly understand the vehicle fault prediction result without triggering the high-voltage system related faults, and to avoid the safety risks caused by the user checking the vehicle fault prediction result on their own.

[0115] In some possible implementations, vehicle fault prediction results are displayed using AR (Augmented Reality) technology.

[0116] Specifically, when the vehicle fault prediction result indicates the fault type as the first fault type, the user terminal's camera is used to capture real-time images via an app. If the real-time images meet display requirements, the vehicle fault prediction result and repair suggestions are overlaid onto them. For example, the image might overlay "In one month, the tire pressure is insufficient; it is recommended to inflate to 2.5 bar" onto the tires. Furthermore, considering the complexity of components on a vehicle, voice guidance can be used to quickly locate the fault prediction result for the user.

[0117] In some possible embodiments, vehicle 11 also includes a keyless entry system and a camera / radar sensor. The keyless entry system is an automotive electronic system that enables seamless entry and start of a vehicle via wireless communication technology, allowing the driver to operate the vehicle without taking a key out of their pocket or bag. The camera / radar sensor is connected to the controller to acquire image / radar data and transmit the image / radar data to the controller.

[0118] By fusing vehicle operating condition information, propagation paths, and anomaly characteristics, a fused feature is obtained, including: When the vehicle owner's identity is confirmed through the keyless entry system, the user's image is obtained; Based on user image recognition, the vehicle body area observed by the user is identified; By fusing vehicle body area, vehicle operating condition information, propagation path, and abnormal features, a fused feature is obtained.

[0119] Due to the complexity of vehicle structure, vehicle malfunctions can occur inside the vehicle structure, on the side closer to the passenger compartment, or on the side farther away from the passenger compartment. For example, interior door handles are located closer to the passenger compartment, while exterior door handles are located farther away from the passenger compartment. Since the interior and exterior door handles are relatively close in position, if the user-observed area of ​​the vehicle body is identified as an interior door handle based on the user image, then the interior door handle is more likely to malfunction. Conversely, if the user-observed area of ​​the vehicle body is identified as an exterior door handle based on the user image, then the exterior door handle is more likely to malfunction. Therefore, by acquiring user images, identifying the user-observed area of ​​the vehicle body, and determining the possible malfunctions based on this area, potential malfunctions can be accurately predicted.

[0120] This method utilizes in-vehicle cameras (cameras in 360° panoramic imaging systems, in-vehicle cameras, etc.) or delayed recordings from dashcams to acquire user images. Image recognition algorithms (such as YOLO object detection and behavioral trajectory analysis) are then used to determine whether the vehicle owner is observing the vehicle's area, such as walking around or stopping to observe, and the specific area observed by the user. This allows for the fusion of vehicle area data, vehicle operating condition information, propagation paths, and abnormal features to predict vehicle malfunctions. This application's embodiment links keyless start scenarios, user behavior, images, and malfunction prediction generation, thereby improving the accuracy of vehicle malfunction prediction.

[0121] In some possible embodiments, the method further includes: Based on the time information corresponding to the vehicle fault prediction results, generate vehicle fault prediction and detection instructions. The steps of acquiring sound information collected by the sound sensor array on the vehicle are executed according to the vehicle fault prediction and detection instructions.

[0122] Since the vehicle fault prediction results include fault severity / time, and fault severity / time can characterize the time when the fault may occur, for example, the above "severity / time: the braking distance will exceed the safe distance after the brake pads work for another 24 hours" characterizes the time when the fault may occur. Therefore, the vehicle fault prediction can be performed again based on the time information corresponding to the vehicle fault prediction results to verify the vehicle fault prediction results and obtain the verification results. In this way, the accuracy of fault prediction can be improved through multiple vehicle fault predictions.

[0123] In some possible embodiments, the method further includes: A warning message will be issued when the fault level in the vehicle fault prediction result is higher than the fault level threshold. The prompt message will be displayed on the terminal.

[0124] In some possible embodiments, the prompt information can also be sent to the cloud-based rescue platform so that the cloud-based rescue platform can prepare a rescue plan in advance, such as dispatching a suitable rescue vehicle or a rescue vehicle carrying motor repair tools to shorten the rescue time.

[0125] In some possible embodiments, the method further includes: One or more of the following data are sent to the cloud server: sound information, vehicle operating condition information, vehicle fault prediction results, propagation path, abnormal features, verification results, and historical data. The cloud server then updates the fault prediction model based on these data for each vehicle and sends the updated model to the controller. The controller then deploys the updated model to predict vehicle faults.

[0126] The vehicle fault prediction method provided in this application acquires sound information collected by a sound sensor array on the vehicle; based on the sound information, it determines the propagation path and abnormal features of the sound information along the vehicle structure; it acquires the vehicle's operating condition information, and fuses the vehicle operating condition information, propagation path, and abnormal features to obtain fused features. The fused features are then input into a fault prediction model to obtain the vehicle fault prediction result. This application determines the propagation path using sound information collected by the sound sensor array, and accurately locates the abnormal sound source based on the propagation path. Since the sound emitted by the abnormal sound source can characterize the vehicle's condition, that is, the propagation path and abnormal features can characterize the vehicle's condition, a multi-source fused feature is constructed by combining the vehicle operating condition information, propagation path, and abnormal features. Vehicle fault prediction is then performed based on the multi-source fused feature, thereby accurately predicting possible faults.

[0127] See Figure 7 This application provides a vehicle fault prediction device 50, which includes: The sound acquisition module 501 is used to acquire sound information collected by the sound sensor array on the vehicle. The first processing module 502 is used to obtain the propagation path of the sound information along the vehicle structure and the abnormal characteristics based on the sound information if there is an anomaly in the sound information. The fusion module 503 is used to acquire vehicle operating condition information and fuse the vehicle operating condition information, propagation path and abnormal features to obtain fused features. The vehicle operating condition information includes at least one of the following: vehicle speed information, motor operating information, battery information and ambient temperature information. The fault prediction module 504 is used to input the fused features into the fault prediction model to obtain the vehicle fault prediction result.

[0128] In some possible embodiments, the fault prediction module 504 includes: The first fault prediction submodule is used to obtain the historical fusion features of the vehicle; The second fault prediction submodule is used to input historical fusion features and fusion features into the fault prediction model to obtain vehicle fault prediction results.

[0129] In some possible embodiments, the first processing module includes: The first processing submodule is used to acquire the vehicle's dynamic stiffness dataset. The dynamic stiffness dataset includes the dynamic stiffness curve of the reference propagation path. The dynamic stiffness curve is used to characterize the change of sound under the action of the dynamic stiffness of the reference propagation path when the sound is transmitted along the reference propagation path. The second processing submodule is used to correct the sound information based on the dynamic stiffness dataset to obtain corrected sound information; The third processing submodule is used to generate propagation paths and anomaly features based on the corrected sound information.

[0130] In some possible embodiments, the third processing submodule is specifically used to determine candidate propagation paths based on the energy changes of the corrected sound information; input the corrected sound information into the sound source localization model to determine the candidate sound source location information; and generate a propagation path based on the comparison results between the candidate propagation paths and the candidate sound source location information.

[0131] In some possible embodiments, the vehicle fault prediction device 50 further includes a display module.

[0132] The display module is used to display the vehicle fault prediction result and the first maintenance suggestion when the fault type of the vehicle fault prediction result is the first fault type. Preferably, the vehicle fault prediction device 50 further includes: an image acquisition module, a vehicle body area recognition module, and a second processing module.

[0133] The image acquisition module is used to acquire user images; The vehicle body area recognition module is used to recognize the vehicle body area observed by the user based on the user image; The second processing module fuses the vehicle body area, vehicle operating condition information, propagation path, and abnormal features to obtain fused features.

[0134] This device acquires sound information collected by a sound sensor array on a vehicle; based on the sound information, it determines the propagation path and abnormal features of the sound information along the vehicle structure; it acquires the vehicle's operating condition information, and fuses the vehicle operating condition information, propagation path, and abnormal features to obtain fused features. These fused features are then input into a fault prediction model to obtain vehicle fault prediction results. This application determines the propagation path using sound information collected by a sound sensor array, and accurately locates abnormal sound sources based on the propagation path. Since the sound emitted by the abnormal sound source can characterize the vehicle's condition—that is, the propagation path and abnormal features can characterize the vehicle's condition—a multi-source fused feature is constructed by combining the vehicle operating condition information, propagation path, and abnormal features. Based on this multi-source fused feature, vehicle fault prediction is performed, thereby accurately predicting possible faults.

[0135] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0136] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one computer program that is loaded and executed by a processor of a computer device to enable the computer to implement any of the vehicle fault prediction methods described above.

[0137] In one possible implementation, the aforementioned computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0138] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the vehicle fault prediction methods described above.

[0139] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0140] It should be understood that "multiple" as used in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0141] It should be noted that the terms "first," "second," etc. (if applicable) in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0142] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A vehicle fault prediction method, characterized in that, The method includes: Acquire sound information collected by the sound sensor array on the vehicle; If the sound information is abnormal, then based on the sound information, the propagation path of the sound information along the vehicle structure and the abnormal characteristics are obtained; The vehicle operating condition information of the vehicle is obtained, and the vehicle operating condition information, the propagation path and the abnormal features are fused to obtain fused features. The vehicle operating condition information includes at least one of the following: vehicle speed information, motor operating information, battery information and ambient temperature information. The fused features are input into the fault prediction model to obtain the vehicle fault prediction result.

2. The method according to claim 1, characterized in that, The step of inputting the fused features into the fault prediction model to obtain the vehicle fault prediction result includes: Obtain the historical fusion features of the vehicle; The historical fusion features and the fusion features are input into the fault prediction model to obtain the vehicle fault prediction result.

3. The method according to claim 1, characterized in that, The step of obtaining the propagation path and abnormal features of the sound information along the vehicle structure based on the sound information includes: Obtain the dynamic stiffness dataset of the vehicle, which includes the dynamic stiffness curve of the reference propagation path. The dynamic stiffness curve is used to characterize the change of sound under the action of the dynamic stiffness of the reference propagation path when the sound is transmitted along the reference propagation path. The sound information is corrected based on the dynamic stiffness dataset to obtain corrected sound information; Based on the corrected sound information, the propagation path and the abnormal features are generated.

4. The method according to claim 3, characterized in that, The propagation path is obtained through the following steps: Based on the energy changes of the corrected sound information, candidate propagation paths are determined; The corrected sound information is input into the sound source localization model to determine the candidate location information of the sound source. The propagation path is generated based on the comparison results between the candidate propagation path and the candidate location information of the sound source.

5. The method according to claim 1, characterized in that, The method further includes: If the fault type in the vehicle fault prediction result is the first fault type, the vehicle fault prediction result and the first repair suggestion are displayed. Preferably, the method further includes: Get user image; The user-observed vehicle body area is identified based on the user image; The vehicle body area, the vehicle operating condition information, the propagation path, and the abnormal features are fused to obtain fused features.

6. A vehicle fault prediction device, characterized in that, The device includes: The sound acquisition module is used to acquire sound information collected by the sound sensor array on the vehicle; The first processing module is used to, if the sound information is abnormal, obtain the propagation path of the sound information along the vehicle structure and the abnormal features based on the sound information; The fusion module is used to acquire the vehicle operating condition information of the vehicle, and fuse the vehicle operating condition information, the propagation path and the abnormal features to obtain fused features. The vehicle operating condition information includes at least one of the following: vehicle speed information, motor operating information, battery information and ambient temperature information. The fault prediction module is used to input the fused features into the fault prediction model to obtain vehicle fault prediction results.

7. The apparatus according to claim 6, characterized in that, The fault prediction module includes: The first fault prediction submodule is used to obtain the historical fusion features of the vehicle; The second fault prediction submodule is used to input the historical fusion features and the fusion features into the fault prediction model to obtain the vehicle fault prediction result.

8. The apparatus according to claim 6, characterized in that, The first processing module includes: The first processing submodule is used to acquire the dynamic stiffness dataset of the vehicle. The dynamic stiffness dataset includes the dynamic stiffness curve of the reference propagation path. The dynamic stiffness curve is used to characterize the change law of the sound under the action of the dynamic stiffness of the reference propagation path when the sound is transmitted along the reference propagation path. The second processing submodule is used to correct the sound information based on the dynamic stiffness dataset to obtain corrected sound information; The third processing submodule is used to generate the propagation path and the abnormal features based on the corrected sound information.

9. The apparatus according to claim 8, characterized in that, The third processing submodule is specifically used to determine candidate propagation paths based on the energy changes of the corrected sound information; and to input the corrected sound information into the sound source localization model to determine the candidate location information of the sound source. The propagation path is generated based on the comparison results between the candidate propagation path and the candidate location information of the sound source.

10. A non-transitory computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is loaded and executed by a processor to implement the vehicle fault prediction method as described in any one of claims 1 to 5.

Images