Vehicle maintenance suggestion obtaining method and system, terminal equipment and storage medium

Abstract

The invention belongs to the technical field of automobile diagnosis, and discloses a vehicle maintenance suggestion obtaining method and system, terminal equipment and a storage medium, and the method comprises the steps: carrying out the collection and processing of target data of a to-be-detected vehicle, and obtaining the processing data; mapping the training data to a corresponding fault category by using a basic neural network architecture so as to train and obtain a deep learning model; performing feature extraction on the processed data by using a deep learning model to obtain actual feature data; comparing the actual feature data with training data corresponding to the fault category by using a deep learning model, and searching to obtain a target fault sub-category in the fault category; and obtaining a maintenance suggestion of the target fault sub-category according to a preset fault mapping table in the knowledge base. After the training data is mapped to the corresponding fault category, the actual feature data obtained through feature extraction is compared with the training data corresponding to the fault category, so that the target fault sub-category in the fault category is found out, and the diagnosis efficiency and the diagnosis precision are effectively improved.

Description

Technical Field

[0001] This application relates to the technical field of vehicle diagnosis, and particularly to a method, a system, a terminal device, and a storage medium for obtaining vehicle maintenance suggestions. Background Art

[0002] Automobiles have become an indispensable part of people's daily lives. People rely on automobiles in all aspects of their lives, work, and entertainment. However, automobiles often malfunction during use, which affects the driving experience of vehicle owners. Nevertheless, traditional vehicle maintenance methods involve the processes of inspecting, troubleshooting, and repairing the vehicle, which are time-consuming and laborious.

[0003] With the continuous development and popularization of artificial intelligence technology, more and more traditional industries have started to use intelligent technologies to improve efficiency and accuracy. The vehicle maintenance industry is no exception. Currently, some vehicles are equipped with self-diagnosis systems and on-board diagnostic instruments, which can help vehicle owners and technical maintenance personnel quickly check and troubleshoot some common faults, but the inspection accuracy is poor and the efficiency is low.

[0004] Therefore, the existing technology still needs to be improved. Summary of the Invention

[0005] This application provides a method, a system, a terminal device, and a storage medium for obtaining vehicle maintenance suggestions, aiming to solve the problems of low diagnostic efficiency and poor accuracy in vehicle fault diagnosis in the existing technology.

[0006] In a first aspect, an embodiment of this application provides a method for identifying vehicle faults, including: Collect and process the target data of the vehicle to be detected to obtain processed data; Use a basic neural network architecture to map training data to corresponding fault categories to train a deep learning model; Use the deep learning model to extract actual feature data from the processed data; Use the deep learning model to compare the actual feature data with the training data corresponding to the fault category, and search for the target fault subcategory in the fault category; Obtain maintenance suggestions for the target fault subcategory according to a preset fault mapping table in the knowledge base.

[0007] In some embodiments, the process of obtaining the training data includes: Obtain historical case data in the historical fault data, and obtain simulation data of faulty components in the vehicle when simulating vehicle faults; Merge the simulation data, the historical case data, and a preset proportion of the target data obtained to obtain training data; Among them, the historical fault data includes: historical case data, and general characteristic data of the target sensor under historical fault phenomena; the historical case data includes: the historical fault phenomena, historical fault codes, and maintenance records; the simulation data includes: simulated fault phenomenon data and simulated image data.

[0008] In some embodiments, the vehicle fault identification method further includes: Select or adjust the structure of the deep learning model according to a preset proportion of the actual characteristic data, detect the performance change of the deep learning model according to user feedback or internal system evaluation content, and trigger the update of the deep learning model; Alternatively, after defining the action set to be pre-taken by the deep learning model, evaluate the execution effect of the action set using a reward function, and optimize the action execution strategy using a reinforcement learning algorithm to optimize the execution effect of the action set; Among them, the reinforcement learning algorithm includes: Q-learning and policy gradient methods; the action set includes: replacing components and adjusting parameters.

[0009] In some embodiments, mapping the training data to the corresponding fault category using the basic neural network architecture includes: Using the fully connected layer in the deep learning model to map the feature vector of the training data to the target feature map; Using the activation function layer in the deep learning model to convert the target feature map into the probability distribution result of the fault category; Obtain the fault category corresponding to the maximum probability in the probability distribution result; Among them, the deep learning model includes: convolutional neural network, recurrent neural network, fully connected neural network, autoencoder, or reinforcement learning model.

[0010] In some embodiments, using the deep learning model to extract actual characteristic data from the processed data includes: Using the convolutional layer and pooling layer in the deep learning model to extract image features and spatial features from the processed data to obtain the actual characteristic data; Among them, the image features include: edges, textures, and shapes; the spatial features include: layouts and positions; the actual characteristic data includes at least one of edges, textures, shapes, measurement data of the target sensor, color changes and brightness differences of the faulty components, and contrast of local regions; among them, the measurement data of the target sensor includes at least one of temperature, pressure, and vibration data; the vibration data includes at least one of acceleration, speed, and displacement.

[0011] In some embodiments, comparing the actual feature data with the training data corresponding to the fault category and finding the target fault sub-category in the fault category includes: Obtaining the generalized feature data and actual feature data of the target sensors corresponding to each fault sub-category in the fault category; Calculating the similarity between the generalized feature data and the actual feature data corresponding to each of the target sensors respectively; Taking the fault sub-category corresponding to the maximum similarity as the target fault sub-category; Wherein, the generalized feature data is the common feature of the faults of the corresponding category.

[0012] In some embodiments, post-processing the target data of the vehicle to be detected to obtain processed data includes: Taking pictures of the target position and target components of the vehicle to be detected, and obtaining the operating parameters of the target components to obtain the target data; Performing pre-processing on the target data to obtain the processed data; Wherein, the target position includes: the outside of the vehicle body and the inside of the vehicle body; the target components include: the engine, the braking system, and the tires.

[0013] In a second aspect, an embodiment of the present application provides a vehicle fault identification system, including: A processing module, configured to post-process the target data of the vehicle to be detected to obtain processed data; A mapping module, configured to map the training data to the corresponding fault category by using a basic neural network architecture to train a deep learning model; An extraction module, configured to extract actual feature data from the processed data by using the deep learning model; A comparison module, configured to compare the actual feature data with the training data corresponding to the fault category by using the deep learning model to find the target fault sub-category in the fault category; An acquisition module, configured to obtain maintenance suggestions for the target fault sub-category according to a preset fault mapping table in a knowledge base.

[0014] In a third aspect, an embodiment of the present application provides a terminal device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the computer program is executed by the processor, the steps of the vehicle fault identification method described above are implemented.

[0015] Fourthly, an embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the vehicle fault identification method described above are implemented.

[0016] Compared with the prior art, a method, a system, a terminal device and a storage medium for obtaining vehicle maintenance suggestions provided by the present application map training data to corresponding fault categories, and then use the trained deep learning model to compare the actually extracted feature data with the training data corresponding to the fault categories, so as to find the target fault sub-category in the fault category, realizing fast and accurate fault judgment and giving maintenance and repair suggestions, thus greatly improving the efficiency and accuracy of maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of a method for obtaining vehicle maintenance suggestions provided by the present application; Figure 2 It is a flowchart of obtaining processed data in the method for obtaining vehicle maintenance suggestions provided by the present application; Figure 3 It is a flowchart of using the method for obtaining vehicle maintenance suggestions provided by the present application to obtain training data; Figure 4 It is a flowchart of mapping fault categories in the method for obtaining vehicle maintenance suggestions provided by the present application; Figure 5 It is a flowchart of finding the target fault sub-category in the method for obtaining vehicle maintenance suggestions provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments.

[0020] The components of the embodiments of the present application that are generally described and illustrated in the accompanying drawings herein can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0021] Hereinafter, the terms "including", "having" and their cognates that can be used in various embodiments of the present application are only intended to represent specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be construed as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or the possibility of adding one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items. In addition, the terms "first", "second", "third", etc. are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0022] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which various embodiments of the present application belong. The terms (such as those defined in a general-use dictionary) will be construed to have the same meaning as the contextual meaning in the relevant technical field and will not be construed to have an idealized meaning or an overly formal meaning unless clearly defined in various embodiments of the present application.

[0023] The present application provides a method, system, terminal device and storage medium for obtaining vehicle maintenance suggestions. After mapping the training data to the corresponding fault categories, the vehicle maintenance suggestion obtaining method uses the trained deep learning model to compare the actually extracted feature data with the training data corresponding to the fault categories, and finds the target fault sub-category in the fault category, realizing fast and accurate fault judgment and giving maintenance inspections, thus greatly improving the efficiency and accuracy of maintenance.

[0024] The design scheme of the vehicle maintenance suggestion obtaining method will be described below through some specific embodiments.

[0025] Please refer to Figure 1 , the embodiments of the present application provide a vehicle fault identification method, including S100-S500: S100. After collecting and processing the target data of the vehicle to be detected, obtain the processed data.

[0026] Exemplarily, first enter the data acquisition stage: collect the target data of the vehicle to be detected through means such as cameras, sensors, data lines, and OBD (On-Board Diagnostics) interfaces. For example, collect image data by taking pictures of the body of the vehicle to be detected, or obtain the operating parameters of a certain component on the vehicle (such as tires and chassis, etc.).

[0027] Among them, the OBD interface is an interface that connects to the vehicle's ECU (Electronic Control Unit) vehicle computer and is used to detect and diagnose vehicle faults.

[0028] Then, perform preprocessing (such as denoising, normalization, and standardization, etc.) on the target data to obtain processed data.

[0029] It can be understood that in this application, by collecting target data and then processing it, the processed data can be effectively learned by the model.

[0030] In one implementation method, please refer to Figure 2 , S100: After collecting and processing the target data of the vehicle to be detected, obtain processed data, including: S110: Take pictures of the target position and target components of the vehicle to be detected, and obtain the operating parameters of the target components to obtain target data; S120: Perform preprocessing on the target data to obtain processed data.

[0031] Among them, the target position includes: the outside of the body and the inside of the body, etc.; the target components include: the engine, braking system, tires, etc.

[0032] Among them, the preprocessing includes at least one of removing noise, missing value processing, numerical normalization, image standardization, image enhancement, and time series enhancement, etc.

[0033] Exemplarily, the process of obtaining processed data is as follows: Use a camera to take pictures of the target position and target components of the vehicle to be detected. For example, take pictures of the entire body outside the vehicle, the interior of the vehicle, the engine, tires, or braking system, etc., and use sensors to detect, or directly connect to the vehicle to be detected with a data line or OBD interface to obtain the parameters of the target components (such as the engine, tires, or braking system) during operation. For example, use a crankshaft position sensor or a camshaft position sensor to detect the engine speed, etc., to achieve the target data in aspects such as inside the vehicle, outside the vehicle, and the engine.

[0034] Then, preprocess the target data. For example, filter out outliers or signal drifts from sensor data (such as temperature, pressure, or rotational speed, etc.) to remove noise; use interpolation method to fill in the missing operating parameters; normalize the sensor data to a specific range (such as [0, 1] or [-1, 1]) to improve the convergence speed of model training, etc.

[0035] S200. Use the basic neural network architecture to map the training data to the corresponding fault categories to train a deep learning model.

[0036] Among them, the fault categories include: engine faults, braking system faults, tire faults, etc.; the deep learning models include: convolutional neural network, recurrent neural network, fully connected neural network, autoencoder, or reinforcement learning model, etc. The fault categories represent a class of vehicle problems with similar fault phenomena and causes.

[0037] Exemplarily, in the model training stage: train the basic neural network architecture, for example, train a general image classification model or a time series analysis model to obtain a deep learning model. The training process is mainly a process of mapping the training data to the corresponding fault categories (such as engine faults, braking system faults, or tire faults, etc.).

[0038] It can be understood that in the model training stage, by using the training data to train the basic neural network architecture to obtain a deep learning model, that is, the deep learning technology is adopted to realize the rapid identification of fault categories.

[0039] In one implementation method, please refer to Figure 3 , the process of obtaining training data includes: S198. Obtain the historical case data in the historical fault data, and obtain the simulation data of the faulty components in the vehicle when simulating the vehicle's failure; S199. Combine the simulation data, historical case data, and the obtained preset proportion of target data to obtain the training data.

[0040] Among them, the historical fault data includes: historical case data, and the general feature data of the target sensors under historical fault phenomena; the historical case data includes: historical fault phenomena, historical fault codes, and repair records; the simulation data includes: simulated fault phenomenon data and simulated image data, and the simulated fault phenomenon data includes the general feature data of the target sensors under simulated fault phenomena. Among them, the preset proportion is not specifically limited in this application and is set according to actual needs, such as 30%. In this application, some representative target data can also be selected as training data.

[0041] Exemplarily, the training data in this application mainly includes three parts: simulation data, historical case data, and part of the target data.

[0042] Then, when simulating data through a simulated vehicle failure, data of faulty components in the vehicle is obtained. For example, various failure conditions of the engine are simulated in the laboratory, and corresponding sensor data and image data are recorded, namely, simulated failure phenomenon data and simulated image data. The simulated data can cover some rare failures or extreme situations, enabling the model to accurately identify similar situations when encountered, so as to increase the generalization ability and robustness of the deep learning model.

[0043] Historical case data can be collected from the after-sales service centers of automobile manufacturers for actual failure case data. These data contain detailed information about vehicle failures, such as failure phenomena, failure codes, and repair records, etc. Among them, historical failure data is an important part of the training data, reflecting various failures and problems that occur during the actual use of the vehicle, and helping the model learn the characteristics and patterns of failures.

[0044] The target data with a preset ratio is part of the real-time collected data. That is, during the data collection stage, data (operating parameters and taken photos) of the interior, exterior, engine, braking system, tires, and chassis of the vehicle to be detected is collected. The real-time collected data can reflect various states and failure conditions of the vehicle during actual operation, and is more authentic and targeted. At the same time, it can also avoid problems such as excessive data volume and uneven data quality that may be brought about by using all collected data as training data.

[0045] In this application, by obtaining three types of specially constructed training data, namely simulated data, historical case data, and partial target data, the diversity of the training data is effectively enriched to expand the learning scope and identification ability of the model.

[0046] In one implementation method, please refer to Figure 4 , and use the basic neural network architecture to map the training data to the corresponding failure categories, including: S201: Use the fully connected layer in the deep learning model to map the feature vector of the training data to the target feature map; S202: Use the activation function layer in the deep learning model to convert the target feature map into the probability distribution result of the failure category; S203: Obtain the failure category corresponding to the maximum probability in the probability distribution result.

[0047] Exemplarily, before mapping the training data to the fault categories, similarly, the basic neural network architecture first extracts features from the training data to obtain feature vectors and then performs classification. Taking vehicle fault pictures as an example, these features may include color changes of faulty components, brightness differences, or contrast in local areas. In addition, other sensor data such as temperature, pressure, or vibration can be combined, and these multi-modal data are fused as the basis for classification.

[0048] Then, after classifying the training data and mapping it to the fault categories, that is, inputting the feature vector into the fully connected layer of the CNN (Convolutional Neural Network). The fully connected layer will perform weighted summation on the feature vector and perform a non-linear transformation through an activation function to obtain an intermediate representation (which belongs to a type of feature map). Then, the softmax layer (i.e., the activation function layer) in the CNN converts the intermediate representation into a probability distribution or score for each fault category. The deep learning model calculates a probability distribution based on the feature vector of the training data, indicating the likelihood of the training data belonging to each fault category. Finally, the fault category corresponding to the maximum probability in the probability distribution result is used as the finally mapped fault category. For example, assuming there are three fault categories: engine fault, braking system fault, and tire fault, the softmax layer will output the probabilities of the faulty vehicle picture belonging to these three fault categories, such as [0.7, 0.2, 0.1], then it indicates that the faulty vehicle picture is most likely to belong to an engine fault.

[0049] S300. Use the deep learning model to extract features from the processed data to obtain actual feature data.

[0050] Among them, the actual feature data includes at least one of the following: edges, textures, shapes, measurement data of target sensors, color changes and brightness differences of faulty components, and contrast in local areas; among them, the measurement data of target sensors includes at least one of the following: temperature, pressure, and vibration data, etc.; the vibration data includes at least one of the following: acceleration, speed, and displacement, etc.

[0051] Exemplarily, after inputting the processed data into the deep learning model, the deep learning model will automatically extract features from the processed data, that is, automatically obtain features such as edges, textures, shapes, color changes and brightness differences of faulty components, and contrast in local areas in the image data of the processed data to obtain actual feature data, so as to provide a basis for subsequent fault diagnosis suggestions. For example, in an engine fault picture, the worn edge features of some components may imply the lifespan problem of the components; in a tire fault picture, the change in texture may reflect the wear degree of the tire or whether there is a foreign object embedded; in a chassis fault picture, the abnormal shape may indicate deformation or damage of the components.

[0052] It can be understood that in this application, feature extraction is performed on the input data to provide a basis for fault diagnosis, thereby improving the diagnosis accuracy.

[0053] In one implementation method, in S300, a deep learning model is used to extract actual feature data from the processed data, including: In S310, the convolutional layer and pooling layer in the deep learning model are used to extract image features and spatial features from the processed data to obtain actual feature data.

[0054] Among them, the image features include: edges, textures, shapes, etc.; the spatial features include: layouts, positions, etc.

[0055] Exemplarily, when performing feature extraction, the convolutional layer and pooling layer in the deep learning model are mainly used to extract image features (including edges, textures, shapes, etc.) and spatial features (including layouts, positions, etc.) from the processed data to obtain actual feature data.

[0056] Moreover, through multi-layer convolution and pooling operations, the CNN can learn the spatial relationship features between the faulty components and other normal components. For example, when detecting engine faults, the CNN can identify whether the relative position of a certain faulty component and the surrounding normal components is abnormal, thereby assisting in judging the fault type. In addition, the CNN can also combine other technologies, such as the attention mechanism, to further focus on the key features of the fault area to further improve the accuracy of fault diagnosis.

[0057] Among them, the convolutional layer extracts local features of the image data in the processed data through a sliding window operation, such as edge, texture, and color changes. Then, the pooling layer retains the most important spatial information through downsampling operations while reducing the computational amount. The pooling layer can capture spatial relationship features in a larger range, such as the overall shape of an object or the relative position between multiple components.

[0058] In S400, a deep learning model is used to compare the actual feature data with the training data corresponding to the fault categories to find the target fault sub-categories in the fault categories.

[0059] Exemplarily, after the training phase is completed, the deep learning model is deployed to the intelligent vehicle maintenance assistant, and then the intelligent vehicle maintenance assistant is used for fault detection, that is, entering the model prediction phase: The deep learning model compares the actually collected feature data with known problem patterns, that is, compares the actual feature data with the training data corresponding to the fault categories, and identifies the fault category with the closest features by calculating the similarity (such as Euclidean distance or cosine similarity, etc.). Among them, the known problem patterns refer to the typical manifestation forms of various faults and problems that occur during the use of the vehicle. For example, engine faults may be manifested as engine jitter, insufficient power, abnormal noise, etc.; braking system faults may be manifested as braking failure, too long braking distance, etc.

[0060] It can be understood that in this application, by comparing the actual feature data with the training data corresponding to the fault categories, the similarity between the two is calculated, and the actual feature data is quickly classified and mapped, so as to provide more targeted maintenance suggestions later, thereby improving the diagnostic accuracy and efficiency.

[0061] In one implementation method, please refer to Figure 5 , compare the actual feature data with the training data corresponding to the fault categories, and find the target fault subcategory in the fault category, including: S401. Obtain the general feature data and actual feature data of the target sensors corresponding to each fault subcategory in the fault category; S402. Calculate the similarity between the general feature data corresponding to each target sensor and the actual feature data respectively; S403. Take the fault subcategory corresponding to the maximum similarity as the target fault subcategory; Among them, the general feature data is the common feature of the corresponding category of faults, including the general feature data of the target sensors under historical fault phenomena, simulated fault phenomena or actual operation.

[0062] Among them, the general feature data belongs to the generalization and summary of the commonality of the corresponding category of faults, is the statistical law and typical manifestation in the training data, and is obtained through the analysis and induction of the features in a large amount of training data. For example, the engine overheat fault category may correspond to a series of specific sensor data features (such as temperature exceeding a certain threshold, abnormal pressure, etc.) and image features (such as color changes and brightness differences of some components).

[0063] Exemplarily, there are multiple fault subcategories in the fault category. For example, engine faults include engine overheat faults and engine cylinder misfire faults, etc., and each fault subcategory may include the general feature data of multiple sensors.

[0064] Then, the specific process of finding the target fault subcategory is as follows: First, obtain the general feature data of the target sensors corresponding to each fault sub-category in the fault category, as well as the actual feature data of the target sensors in the feature data. Calculate the similarity between the general feature data and the actual feature data of each same sensor. It can be calculating the Euclidean distance or cosine similarity between the two to calculate the similarity of all fault sub-categories in the fault category.

[0065] For example, assume that the sensor data vector (i.e., general feature data) in the engine overheating fault mode is [80, 120, 0.5] (representing temperature, pressure, and vibration respectively), and the sensor data vector (actual feature data) collected in real time is [85, 130, 0.6]. Then the Euclidean distance between the two vectors is: = ≈11.18; Finally, filter out the maximum similarity among the similarities of all fault sub-categories in the fault category, and use the fault sub-category corresponding to the maximum similarity as the target fault sub-category this time, that is, filter out the fault sub-category with the highest possible probability from the large category of fault categories.

[0066] S500. Obtain the maintenance suggestions for the target fault sub-category according to the preset fault mapping table in the knowledge base.

[0067] Exemplarily, there is a mapping relationship established in advance between each fault sub-category and the corresponding fault maintenance suggestions in the knowledge base or rule base, that is, the preset fault mapping table. For example, the corresponding maintenance suggestions for the engine overheating fault are stored in the preset fault mapping table, such as checking the cooling system or replacing the radiator, etc.

[0068] Then, after identifying the target fault sub-category, according to the mapping relationship between the fault sub-category and the maintenance suggestions, retrieve the corresponding maintenance suggestions from the preset fault mapping table, and generate corresponding solutions for the user, such as suggesting to replace the faulty parts, adjust the system settings or provide maintenance suggestions. In addition, the maintenance suggestions can be personalized adjusted and optimized in combination with the actual situation and maintenance history of the vehicle.

[0069] It can be understood that in this application, by finding the maintenance suggestions mapped to the target fault sub-category according to the preset fault mapping table, not only the automatic diagnosis and location of faults and problems are realized, but also the efficiency and accuracy of generating maintenance suggestions are effectively improved, thereby improving the maintenance efficiency and maintenance accuracy.

[0070] In an implementation method, the vehicle fault identification method further includes: S600. Select or adjust the structure of the deep learning model according to the actual feature data of the preset ratio, detect the performance change of the deep learning model according to the user feedback or the internal evaluation content of the system, and trigger the update of the deep learning model; Alternatively, S700. After defining the set of actions to be taken by the deep learning model, use the reward function to evaluate the execution effect of the set of actions, and use the reinforcement learning algorithm to optimize the action execution strategy to optimize the execution effect of the set of actions.

[0071] Among them, the reinforcement learning algorithm includes: Q-learning and policy gradient methods; the set of actions includes: replacing components and adjusting parameters. Among them, Q-learning is a model-free, off-policy reinforcement learning algorithm, and its core idea is to dynamically update the Q value through the Bellman equation, and approximate the optimal decision by combining the exploration and exploitation strategies. The policy gradient method is one of the core algorithms of reinforcement learning, and its core idea is to directly optimize the policy function.

[0072] Exemplarily, in this application, whether in the model training or model prediction stage, there will be model optimization, that is, using technologies such as adaptive learning algorithms and enhanced learning algorithms to continuously learn and optimize the model, so that the model can adapt to different brands of cars and driving habits and environments in different regions.

[0073] Then, the adaptive learning algorithm is an algorithm that can automatically adjust the model parameters according to new data. In the intelligent vehicle maintenance assistant, the adaptive learning algorithm can be used to update and optimize the fault diagnosis model.

[0074] The adaptive learning algorithm mainly includes: the online learning stage, the feedback stage, and the model selection and optimization stage.

[0075] As new collected data arrives, the model needs to continuously update its parameters to adapt to the change of data distribution, so as to achieve online learning.

[0076] However, through the user feedback or the internal evaluation mechanism of the system, detect the change of the model performance and trigger the model update. And, according to the characteristics of the new data, select or adjust the model structure, such as increasing the number of network layers, changing the activation function, etc., to achieve model selection and optimization.

[0077] The reinforcement learning algorithm enables the agent to learn the optimal strategy during the interaction with the environment. In the intelligent vehicle maintenance assistant, reinforcement learning can be used to optimize the maintenance suggestions or maintenance strategies.

[0078] The reinforcement learning algorithm mainly includes: the state representation stage, the action space definition stage, the reward function design stage, and the policy optimization stage.

[0079] State representation: mainly represent the current vehicle state, fault information, etc. as the state in reinforcement learning.

[0080] Action space definition: mainly define the set of actions that the agent can take, such as replacing components or adjusting parameters, etc.

[0081] Reward function design: design a reward function to evaluate the effect of the actions taken by the agent, such as rewards after successful repairs, rewards for reduced repair costs, etc.

[0082] Policy optimization: use reinforcement learning algorithms (such as Q-learning, policy gradient methods, etc.) to optimize the agent's policy to maximize the cumulative reward.

[0083] It can be understood that in this application, by updating the deep learning model and optimizing the action execution policy therein, the intelligent vehicle repair assistant can continuously learn and optimize its performance, enabling it to adapt to different brands of vehicles and driving habits and environments in different regions, and thus providing more accurate and efficient repair services.

[0084] In one implementation method, the vehicle fault identification method further includes: Identify different target gestures or target voice information of the user, and correspondingly control the opening of preset target instructions to switch diagnostic items or control the display page.

[0085] Exemplarily, there is a human-computer interaction interface in the intelligent vehicle repair assistant, including functions such as speech recognition, speech broadcast, and gesture recognition. For example, during the repair process, the repair personnel can use gestures to control the interface operations of the repair assistant, such as zooming in, zooming out, switching pages, etc., to improve the convenience of operation. In addition, during the diagnosis process, the repair personnel can use specific gestures to start or stop certain diagnostic functions, such as drawing a circle with a gesture to start the engine fault diagnosis, and swiping with a gesture to switch to the next diagnostic item, etc., so as to realize switching the current diagnostic item or controlling the display page according to specific gestures or voice information, such as the display size of the page.

[0086] Then, in order to recognize the target gesture, it is necessary to pre-define the association rules between the gesture and the repair instruction (i.e., the preset target instruction), and collect three types of corresponding training data to train relevant models (such as acoustic models and speech models, etc.).

[0087] For example, developers will design a series of specific gestures according to the operating habits and needs of maintenance personnel, and bind these gestures to corresponding maintenance instructions. When a maintenance personnel makes a certain gesture, the intelligent vehicle maintenance assistant will identify the type of the gesture and execute the corresponding maintenance instruction according to the predefined rules. At the same time, this association can also be optimized and adjusted through machine learning algorithms to improve the accuracy of gesture recognition and the execution efficiency of maintenance instructions. Similarly, to implement speech recognition, it is also necessary to establish an association rule between the target speech information and the preset target instruction, and the setting method is similar to the above method.

[0088] Among them, the implementation process of the speech recognition function is as follows: It mainly includes: the audio preprocessing stage and the training stage.

[0089] Audio preprocessing: mainly denoise, filter and extract features from the collected audio signals. For example, using Mel Frequency Cepstral Coefficients (MFCC), by simulating the non-linear perception of frequency by the human ear (Mel scale), the speech signal is converted into a low-dimensional feature vector, retaining key information such as formants and pitch of the speech.

[0090] The training stage includes: training the acoustic model: using a deep learning model (such as a recurrent neural network or a long short-term memory network) to train the acoustic model, mapping the audio features to text characters or words.

[0091] And, training the language model: using natural language processing technology to train the language model, which is used to predict and correct errors in the speech recognition results.

[0092] The implementation process of the speech broadcast function is as follows: It mainly includes: the text preprocessing stage and the speech formation stage.

[0093] Text preprocessing: mainly perform preprocessing such as word segmentation and phonetic annotation on the text to be broadcast.

[0094] Speech formation includes: speech synthesis: using speech synthesis technology (such as Google's Tacotron or Baidu's DeepVoice) to convert the text into a speech signal. And audio output: playing the generated speech signal through a speaker.

[0095] The implementation process of the gesture recognition function is as follows: It mainly includes: the image preprocessing stage, the feature extraction stage and the gesture classification stage.

[0096] Image preprocessing: perform preprocessing such as cropping, scaling and background removal on the gesture image captured by the camera.

[0097] Feature extraction: using a deep learning model (such as a convolutional neural network) to extract features from the gesture image.

[0098] Gesture classification: The extracted features are input into a classifier (such as a support vector machine (SVM), decision tree, random forest, or deep neural network (DNN)) to identify the gesture category.

[0099] It can be understood that in this application, by identifying the target gesture or target voice information to control the activation of the preset target instruction, the human-computer interaction experience of the operation is effectively improved, making it more convenient, fast, and accurate for the vehicle owner or technical maintenance personnel during use.

[0100] In one implementation method, an embodiment of this application provides a vehicle fault identification system, including: A processing module 10, configured to collect and process the target data of the vehicle to be detected to obtain processed data; A mapping module 20, configured to map the training data to the corresponding fault categories by using a basic neural network architecture to train a deep learning model; An extraction module 30, configured to use the deep learning model to extract features from the processed data to obtain actual feature data; A comparison module 40, configured to use the deep learning model to compare the actual feature data with the training data corresponding to the fault categories to find the target fault subcategories in the fault categories; An acquisition module 50, configured to obtain maintenance suggestions for the target fault subcategories according to the preset fault mapping table in the knowledge base.

[0101] Exemplarily, if the vehicle fault identification method is applied to a vehicle fault identification system, then the implementation process of the vehicle fault identification system is as follows: First, call the camera to take pictures of the body or components of the vehicle to be detected to obtain image data, or connect a data cable or OBD connection port to obtain the operating parameters of target components such as the engine, tires, or braking system during operation to obtain target data. And perform preprocessing such as noise removal, missing value processing, or numerical normalization on the target data to obtain processed data.

[0102] Then, enter the model training stage, that is, map the training data to the corresponding fault categories by using a basic neural network architecture, that is, perform feature extraction on the training data and then perform a linear transformation, and then convert it into the probability distribution of each fault category to implement the training of the basic neural network architecture to obtain a deep learning model (for example: convolutional neural network, recurrent neural network, fully connected neural network, autoencoder, or reinforcement learning model, etc.), and deploy the deep learning model to the intelligent vehicle maintenance assistant.

[0103] Secondly, the model prediction stage begins, that is, using the deep learning model to calculate the similarity between the actual feature data and the training data corresponding to the fault category, so as to identify the most likely target fault subcategory in the fault category.

[0104] Finally, search for the pre-stored preset fault mapping table in the knowledge base to obtain the maintenance suggestions corresponding to the target fault subcategory. For example, the maintenance suggestions stored in the preset fault mapping table for the engine overheating fault include: checking the cooling system or replacing the radiator, etc.

[0105] It can be understood that in this application, after mapping the training data to the corresponding fault category, using the obtained deep learning model, comparing the actual feature data extracted by the feature extraction with the training data corresponding to the fault category, searching for the target fault subcategory in the fault category, and searching for the maintenance suggestions corresponding to the target fault subcategory in the preset fault mapping table, thus realizing the automatic diagnosis and positioning of vehicle faults and problems, and quickly and accurately giving the corresponding maintenance suggestions, thereby providing an efficient and convenient maintenance service.

[0106] Moreover, the intelligent vehicle maintenance assistant of this application is easy to use, can easily solve problems, and avoids the human and material costs required by traditional maintenance methods. And it also has the characteristic of high intelligence. This assistant uses AI technologies such as deep learning and image recognition, and can realize the automatic diagnosis and solution of faults and problems, avoiding human errors and delays in the maintenance process.

[0107] This application also provides a terminal device. Exemplarily, the terminal device includes a processor and a memory. Among them, the memory stores a computer program, and the processor runs the computer program, so that the terminal device executes the functions of the above vehicle maintenance suggestion acquisition method or each module in the above vehicle maintenance suggestion acquisition system.

[0108] Among them, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc., and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0109] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electric Erasable Programmable Read-Only Memory (EEPROM), etc. Among them, the memory is used to store a computer program, and after receiving an execution instruction, the processor can execute the computer program accordingly.

[0110] This application also provides a computer-readable storage medium for storing the computer program used in the above terminal device. For example, the computer-readable storage medium may include, but is not limited to: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0111] In several embodiments provided by this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and structure diagrams in the drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to multiple embodiments of this application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in an alternative implementation, the functions marked in the blocks can occur in a different order from that marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the structure diagram and / or flowchart, as well as the combination of blocks in the structure diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0112] In addition, in each embodiment of this application, the various functional modules or units can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0113] When the above-mentioned functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or a part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a smart phone, a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application.

[0114] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application.

Claims

1. A method for obtaining vehicle maintenance suggestions, characterized in that, Including: After collecting and processing the target data of the vehicle to be detected, the processed data is obtained; Using the basic neural network architecture to map the training data to the corresponding fault categories to train a deep learning model; Using the deep learning model, extracting features from the processed data to obtain actual feature data; Using the deep learning model, comparing the actual feature data with the training data corresponding to the fault categories, and searching for the target fault subcategories in the fault categories; Obtaining the maintenance suggestions for the target fault subcategories according to the preset fault mapping table in the knowledge base.

2. The method for obtaining vehicle maintenance suggestions according to claim 1, wherein, The process of obtaining the training data includes: Obtaining the historical case data in the historical fault data, and obtaining the simulation data of the faulty components in the vehicle when the simulated vehicle fails; Combining the simulation data, the historical case data, and the obtained preset proportion of the target data to obtain the training data; Wherein, the historical fault data includes: historical case data, and the general feature data of the target sensor under the historical fault phenomenon; the historical case data includes: the historical fault phenomenon, the historical fault code, and the maintenance record; the simulation data includes: the simulated fault phenomenon data and the simulated image data.

3. The method for obtaining vehicle maintenance suggestions according to claim 1, wherein, It also includes: According to the preset proportion of the actual feature data, selecting or adjusting the structure of the deep learning model, and detecting the performance change of the deep learning model according to the user feedback or the system internal evaluation content, and triggering the update of the deep learning model; Or, after defining the action set that the deep learning model is to take, evaluating the execution effect of the action set using the reward function, and optimizing the action execution strategy using the reinforcement learning algorithm to optimize the execution effect of the action set; Wherein, the reinforcement learning algorithm includes: Q-learning and policy gradient methods; the action set includes: replacing components and adjusting parameters.

4. The method for obtaining vehicle maintenance suggestions according to claim 1, wherein The using the basic neural network architecture to map the training data to the corresponding fault categories includes: Using the fully connected layer in the deep learning model to map the feature vector of the training data to the target feature map; Using the activation function layer in the deep learning model to convert the target feature map into the probability distribution result of the fault category; Obtaining the fault category corresponding to the maximum probability in the probability distribution result; Wherein, the deep learning model includes: convolutional neural network, recurrent neural network, fully connected neural network, autoencoder or reinforcement learning model.

5. The method for obtaining vehicle maintenance suggestions according to claim 1, wherein The using the deep learning model to extract features from the processed data to obtain actual feature data includes: Using the convolutional layer and pooling layer in the deep learning model to extract the image features and spatial features in the processed data to obtain the actual feature data; Among them, the image features include: edges, textures, and shapes; the spatial features include: layouts and positions; the actual feature data includes at least one of edges, textures, shapes, measurement data of the target sensor, color changes and brightness differences of the faulty components, and contrast of local areas; among them, the measurement data of the target sensor includes at least one of temperature, pressure, and vibration data; the vibration data includes at least one of acceleration, velocity, and displacement.

6. The method for obtaining vehicle maintenance suggestions according to claim 2, wherein, The comparing the actual feature data with the training data corresponding to the fault category to find the target fault subcategory in the fault category includes: Obtaining the generalized feature data and actual feature data of the target sensor corresponding to each fault subcategory in the fault category; Calculating the similarity between the generalized feature data and the actual feature data corresponding to each target sensor respectively; Taking the fault subcategory corresponding to the maximum similarity as the target fault subcategory; Among them, the generalized feature data is the common feature of faults of corresponding categories, including the generalized feature data of the target sensor under historical fault phenomena, simulated fault phenomena, or actual operation.

7. The method for obtaining vehicle maintenance suggestions according to claim 1, wherein The post-processing of the target data of the vehicle to be detected to obtain processed data includes: Taking pictures of the target position and target components of the vehicle to be detected, and obtaining the operating parameters of the target components to obtain the target data; Performing preprocessing on the target data to obtain the processed data; Among them, the target position includes: the exterior and interior of the vehicle body; the target components include: the engine, braking system, and tires.

8. A vehicle maintenance advice acquisition system, characterized in that, Includes: A processing module for post-processing the target data of the vehicle to be detected to obtain processed data; A mapping module for mapping the training data to the corresponding fault category by using a basic neural network architecture to train a deep learning model; An extraction module for extracting actual feature data from the processed data by using the deep learning model; A comparison module for comparing the actual feature data with the training data corresponding to the fault category by using the deep learning model to find the target fault subcategory in the fault category; An acquisition module for obtaining maintenance suggestions for the target fault subcategory according to a preset fault mapping table in the knowledge base.

9. A terminal device, characterized in that, Includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, and when the computer program is executed by the processor, the steps of the vehicle maintenance suggestion acquisition method according to any one of claims 1-7 are implemented.

10. A computer-readable storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a processor, the steps of the vehicle maintenance suggestion acquisition method according to any one of claims 1-7 are implemented.

Images