Automobile fault intelligent diagnosis and early warning method based on knowledge graph

Abstract

The invention discloses an automobile fault intelligent diagnosis and early warning method based on a knowledge graph, and the method comprises the steps: obtaining a vehicle fault maintenance case, and extracting data related to a fault; extracting a structured text from the unstructured text; constructing a vehicle fault knowledge graph; constructing a fault prediction model based on the automobile driving data; and constructing a vehicle fault early warning and cloud diagnosis system. According to the invention, based on the vehicle fault prediction model and the vehicle fault knowledge graph, a fault cloud diagnosis function in a vehicle operation process is realized. And the efficiency in the vehicle maintenance process and the safety in the vehicle driving process are greatly improved. And establishing an unstructured text structuring method based on rules and a TextCNN model, realizing text extraction and text value classification of the automobile fault maintenance case, and obtaining structured text data for constructing the knowledge graph. The invention provides a structured text extraction method of an unstructured text, which is beneficial to extracting information from more unstructured maintenance cases.

Description

Technical Field

[0001] The present invention relates to a vehicle fault diagnosis method, and in particular to a vehicle fault intelligent diagnosis and early warning method based on a knowledge graph. Background Art

[0002] The annual growth of automobile ownership has put forward higher requirements for vehicle fault diagnosis and maintenance. By collecting and analyzing automobile driving data, the operating status and performance parameters of each component of the vehicle can be monitored in real time, so as to timely discover potential fault signs or abnormal conditions. Traditionally, there has been a lot of data accumulation in the field of automobile failure on the Internet, which may exist in the form of official structured maintenance records or unstructured fault maintenance case texts. The structured maintenance records provided by the official are mostly internal corporate materials that are difficult to obtain, and the maintenance records of a single brand are not applicable to all brands and models; the online automobile failure cases are mostly unstructured texts, and it is still necessary to manually extract the relevant information from the failure, and it is necessary to extract the structured text to construct a knowledge graph system. Therefore, developing a method for early warning and cloud diagnosis of potential vehicle failures using the Internet and big data related technologies can reduce the number of failures during vehicle driving and improve the efficiency of fault maintenance. Through the automobile failure knowledge graph, the characteristic factors that affect a certain vehicle failure can be analyzed from the perspective of the vehicle component structure.

[0003] The Chinese patent application with application number 202311383305.0, "A method and system for intelligent diagnosis of automobile faults based on knowledge graph", obtains the faulty component information and preset knowledge graph of the target vehicle, generates a recommended collection of fault phenomena corresponding to the faulty component or a global collection of fault phenomena based on the acquired faulty component information and the preset knowledge graph, and finally, after generating the recommended collection of fault phenomena corresponding to the faulty component or the global collection of fault phenomena, the fault phenomenon corresponding to the faulty component of the target vehicle is determined under the triggering of the user confirmation command, and the cause, maintenance means and maintenance case of the fault phenomenon are obtained based on the preset knowledge graph. Disadvantages: Fault prediction is based on vehicle components and cannot be combined with vehicle driving related data.

[0004] The Chinese patent application "Method for Searching Automobile Maintenance Data Based on Knowledge Graph" with application number 202210956760.4 first obtains the text of automobile fault information input by the user through the fault identification unit, performs fault identification on the text of the input fault information to obtain fault pairs, and sends the obtained fault pairs to the candidate answer generation unit; by utilizing natural language processing and KBQA technology to solve the consultation of related information such as automobile fault repair and automobile fault causes; it solves the generalization problem of the statement table of the traditional robot question-and-answer system. Disadvantages: The solution mainly identifies the fault based on the fault text information input by the user, so as to match the entity relationship in the knowledge base to obtain the most matching N answers. If there is a problem with the user's fault text information description, it may cause the failure of fault identification, and the knowledge graph cannot match the appropriate answer or matches the wrong answer.

[0005] The Chinese patent application No. 201410683965.5, "An Intelligent Diagnosis Method for Electric Vehicle Faults Based on CAN Information", obtains the occurrence time of the fault to be diagnosed recorded by the system; retrieves the CAN information source data within a certain period of time before and after the occurrence time of the fault to be diagnosed; determines the time period t0-t1 to which the actual occurrence time T of the fault to be diagnosed belongs; verifies the CAN information source data within the time period t0-t1 and a certain period of time before t0; finds the actual cause of the fault to be diagnosed, and feeds it back to the display screen through the CAN bus. Disadvantages: Fault information data can only be retrieved through CAN bus information, and faults cannot be identified and solutions cannot be found based on user input.

[0006] The Chinese patent application numbered 201110311011.8, "A Method for Intelligent Fault Diagnosis, Warning and Alarm of Automobiles", uses a GPS module to obtain vehicle location information, and a CAN bus to obtain vehicle status information, and sends the vehicle status information to a call center. When the call center diagnoses a vehicle fault, it sends a warning message to the car audio host. When the car audio host detects a dangerous vehicle status signal, it automatically dials the call center's dedicated emergency rescue number, and sends the vehicle's location information and status information to the call center. Disadvantages: Faults cannot be identified based on user input and vehicle driving data, and solutions cannot be provided.

[0007] The Chinese patent application with application number 202311318461.9, "An Intelligent Diagnosis Method for New Energy Vehicles Based on Fault Tree", reads gateway data, bus data, ECU data, fault codes and intelligently diagnoses, uses fault tree analysis to judge and repair faults, and only requires maintenance personnel to follow the prompts to find the fault point of the new energy vehicle, changing from the traditional fixed mode to the mode of selecting fault phenomena, that is, starting directly from the fault phenomenon, making fault diagnosis faster, and at the same time, each fault scan takes a short time, and the fault point is more accurately located. Based on the vehicle network, ECU, fault code, diagnostic data stream and internal bus signal of the vehicle, a comprehensive analysis of vehicle faults is performed, effectively reducing the probability of constantly disassembling parts to determine whether the accessories are normal. After the vehicle is repaired, it can be quickly scanned to intelligently determine whether the vehicle fault has been eliminated. Disadvantages: Faults cannot be identified and solutions cannot be found based on user input, and there is no recommendation of fault causes and solutions.

[0008] The Chinese patent application with application number 201810911168.6, "A Cloud-based Intelligent Diagnosis and Matching System for Automobile Faults", requires the user to first input the detection data of the detected vehicle through the detection module, and use the automatic identification function of the work number to identify the vehicle series and model information, and then use the vehicle-specific decoder to connect to the diagnostic connector of the vehicle diagnostic socket to collect the data of the automobile electronic control fault phenomenon. The data is transmitted to the storage module through the transmission module, and the algorithm is notified. Finally, the algorithm module obtains the detection data from the storage module and performs artificial intelligence diagnosis through the vehicle series and model, matching the original parameter data. The detection data is deduced through the original parameter data matching and algorithm calculation to obtain the diagnosis result big data through the diagnosis result, matching the maintenance method and solution through the maintenance method, matching the accessories and consumables. Disadvantages: This method mainly uses structured data, and the mining of the dynamic characteristics of unstructured events of automobile failures is relatively insufficient. Summary of the invention

[0009] Purpose of the invention: The purpose of the present invention is to provide a simple, efficient and accurate method for intelligent diagnosis and early warning of automobile faults based on knowledge graphs.

[0010] To realize the early warning and cloud diagnosis functions of vehicle faults, the present invention first uses crawler technology to obtain automobile fault repair cases from the Internet, then extracts structured text from unstructured text, and constructs an automobile fault knowledge graph. Then, based on the real-time driving data of the vehicle, typical faults are screened, and the vehicle fault knowledge graph is used to screen the factors affecting the faults. After cleaning the vehicle driving data, different time series deep learning models are used to predict the faults. Finally, the early warning and cloud diagnosis functions of vehicle faults are realized based on the driving data of the vehicle.

[0011] Technical solution: The automobile fault intelligent diagnosis and early warning method based on knowledge graph described in the present invention comprises the following steps:

[0012] Step 1: Acquire online automobile repair cases through crawlers, filter fault extraction rules based on the automobile parts-specific dictionary, and extract text data related to fault causes, fault parts, and fault solutions according to the fault extraction rules.

[0013] Step 2: Use rule-based and TextCNN methods to extract automobile fault structured text from unstructured text. The steps include:

[0014] 2.1 Extract fault-related sentences through special nouns in the automotive field, and build fault text extraction rules based on the frequency of occurrence of context words.

[0015] 2.2 Use fault extraction rules to extract fault-related sentences, and use the TextCNN model to classify the fault sentences by value.

[0016] 2.3 Finally, use Jieba word segmentation to extract fault-related information in valuable sentences and complete the extraction of structured text data.

[0017] Step 3: Build the automobile failure knowledge graph through neo4j.

[0018] Establish a connection between Python and Neo4j, use the structured automobile fault text dataset extracted from automobile fault repair cases to create corresponding entity nodes, and extract the dataset required for building the automobile fault knowledge graph.

[0019] Step 4: Construct a fault prediction model based on vehicle driving data. The steps include:

[0020] 4.1 Based on the vehicle driving data, the data is preprocessed by splitting and reorganizing the faulty and non-faulty data, converting the missing value filling format, and classifying the faults.

[0021] 4.2 Filter the influencing feature quantities of vehicle faults through the retrieval method of knowledge graph, and construct the fault data set through time series extraction to provide a data basis for the construction of fault diagnosis model.

[0022] 4.3 After that, MC-DCNN and ResNet neural networks are used for fault prediction training, and the best model is used to predict vehicle faults, and a fault prediction model based on vehicle driving data is constructed.

[0023] Step 5: Build a vehicle fault warning and cloud diagnosis system.

[0024] First, the vehicle driving data is preprocessed and feature extracted, and then fault prediction is performed. If there is no fault, the output is no fault; if there is a fault, the faulty component is input into the knowledge graph for retrieval, and the cause of the fault and maintenance method are output and an alarm is issued.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. The present invention combines fault detection with vehicle driving related data, solving the problems of automobile maintenance relying solely on manual experience, automobile maintenance data knowledge waste, and knowledge difficulty in sharing.

[0027] 2. The present invention will realize the cloud fault diagnosis function during vehicle operation based on the vehicle fault prediction model and vehicle fault knowledge graph. It will provide maintenance methods and suggestions for potential faults during vehicle operation, greatly improving the efficiency of vehicle maintenance and the safety of vehicle driving.

[0028] 3. The present invention will establish an unstructured text structuring method based on rules and TextCNN model, realize the text extraction and text value classification of automobile fault repair cases, and obtain structured text data for building knowledge graphs. It provides a structured text extraction method for unstructured text, which is conducive to extracting information from more unstructured repair cases. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Extracting flow charts for structured text data;

[0030] Figure 2 Build a flow chart for the automobile failure knowledge graph;

[0031] Figure 3 is a flow chart of the method of the present invention;

[0032] Figure 4 This is a diagram of the TextCNN model architecture;

[0033] Figure 5 This is the ResNet time series classification model structure. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be further described below.

[0035] Step 1: Acquire online automobile repair cases through crawlers, filter fault extraction rules based on the automobile parts-specific dictionary, and extract text data related to fault causes, fault parts, and fault solutions according to the fault extraction rules.

[0036] Vehicle fault repair cases are obtained through web crawler technology. Python crawler technology is used to establish a connection with the website first, and then the web page is read in a loop through programming settings. The crawled data is stored after DNS resolution, and the extracted automobile fault repair cases are used as the original data set for building a vehicle fault knowledge graph, and structured text extraction is performed based on this.

[0037] Step 2: Use rule-based and TextCNN methods to extract automobile fault structured text from unstructured text. The steps include:

[0038] 2.1 Extract fault-related sentences through automobile field proper nouns, and build fault text extraction rules according to the frequency of occurrence of context words. The present invention uses the automobile maintenance proper word library provided by Sogou and converts it into txt text format.

[0039] 2.2 Use fault extraction rules to extract fault-related sentences, and use the TextCNN model to classify the fault sentences by value.

[0040] Considering text data as a one-dimensional image, the input text is first converted into a word vector matrix through the word vector encoding layer, and then the feature is extracted through the convolution layer. The convolution layer consists of a convolution kernel and an activation function. The convolution layer constructs a one-dimensional convolution kernel and convolves with the text matrix within the convolution window to extract text features. The pooling layer compares the text features and extracts the most obvious features. After the convolution layer and the pooling layer extract the features, the connection layer is used to integrate the feature information. It acts as a classifier of the model, converting the extracted text features into category features, and finally enters the classification output result, that is, the Softmax function is used to output the category with the highest probability as the classification result.

[0041] 2.3 Finally, jieba word segmentation is used to extract fault-related information in valuable sentences to complete the extraction of structured text data. For example, for sentences containing fault solutions, jieba word segmentation can extract verbs and nouns in the sentence, which are generally the corresponding maintenance actions and maintenance objects.

[0042] Step 3: Build the automobile failure knowledge graph through neo4j.

[0043] Establish a connection between Python and Neo4j, use the structured automobile fault text dataset extracted from automobile fault repair cases to create corresponding entity nodes, and extract the dataset required for building the automobile fault knowledge graph.

[0044] Using the structured automobile failure text dataset extracted from automobile failure repair cases, the corresponding entity nodes are created and expressed in triple form as: <failure phenomenon, leads to, cause of failure>, <failure cause, affects, repair part>, <repair method, applied to, faulty part>.

[0045] Step 4: Construct a fault prediction model based on vehicle driving data. The steps include:

[0046] 4.1 Based on the vehicle driving data, the data is preprocessed by splitting and reorganizing the faulty and non-faulty data, converting the missing value filling format, and classifying the faults.

[0047] For example, the air intake data is 1.0 kg / s. Although this data representation method is convenient for human reading in some cases, it is not suitable for machine learning and statistical analysis. Use the regular expression method in Python to extract the numerical values ​​and convert the data type of each column to float type for subsequent model construction.

[0048] 4.2 Filter the influencing feature quantities of vehicle faults through the retrieval method of knowledge graph, and construct the fault data set through time series extraction to provide a data basis for the construction of fault diagnosis model.

[0049] The vehicle driving data is extracted according to the fault feature items to extract the factors related to the vehicle fault. When it is necessary to obtain feature data for a specific fault moment, the end point of the observation window will be located at the moment when the fault occurs. The continuous feature data covered in the window is the feature set at the moment of the fault. At the same time, it is necessary to randomly extract 50 minutes of fault-free driving data to form a fault-free data set in order to form a complete prediction data set.

[0050] 4.3 After that, the best ResNet neural network model is used for fault prediction training to build a fault prediction model based on vehicle driving data.

[0051] The input of the multivariate time series classification involved in the present invention is multiple one-dimensional subsequences. It is necessary to separate the multivariate time series into univariate time series, and perform feature learning on each univariate sequence separately. In this paper, the residual neural network consists of a residual block, a residual network and a classifier. Each residual block consists of three convolutional layers and their corresponding batch normalization (Batch Normalization) and ReLU activation functions, and also contains a shortcut connection (shortcut connection). The residual network consists of three residual blocks, and the number of channels in each block gradually increases. The ResNet time series classification model consists of multiple residual blocks.

[0052] Step 5: Build a vehicle fault warning and cloud diagnosis system.

[0053] First, the vehicle driving data is preprocessed and feature extracted, and then fault prediction is performed. If there is no fault, the output is no fault; if there is a fault, the faulty component is input into the knowledge graph for retrieval, and the cause of the fault and maintenance method are output and an alarm is issued.

[0054] The vehicle fault takes the original driving data of the vehicle as input, performs a series of analysis and calculation processing on the vehicle driving data according to the vehicle fault diagnosis model based on driving data constructed in step 2 and the vehicle fault knowledge graph constructed in step 3, and outputs the final fault cloud diagnosis result to the user.

[0055] According to the data preprocessing method in step 2, complete the processing of invalid values ​​and missing values ​​in the data, and process the data types of each feature column in the csv file. Finally, extract the feature items of the data to provide data for the subsequent fault prediction model. Determine whether the vehicle has a fault through the prediction results of the ResNet model. If the prediction results of the model show that the vehicle is running normally, that is, no fault is detected, then the system will output the result of "no fault". If the model predicts that there is a specific type of fault, the system will proceed to the next step. In this case, the model will output the fault type and the specific faulty parts, and input them into the fault repair knowledge graph for retrieval, call the py2neo library, use the Cypher query language for the queried faulty parts, execute the query in the Neo4j database, and find the relevant fault causes and repair methods. Extract the unit where "faulty parts" in the fault knowledge graph are faulty parts, and obtain all the fault causes and solutions corresponding to the faulty parts through the relationship network of the knowledge graph, and output them to the user.

[0057] Figure 4 The figure is a schematic diagram of the TextCNN model architecture. The entire model consists of four parts: input layer (word embedding layer), convolution layer, pooling layer and fully connected layer. When the TextCNN model performs text classification, it first converts the input text into a word vector matrix (Wi, W j The convolutional layer is composed of a convolution kernel and an activation function. The convolutional layer extracts text features by constructing a one-dimensional convolution kernel and performing convolution with the text matrix in the convolution window. The pooling layer compares the text features and extracts the most obvious features. After the convolutional and pooling layers extract the features, the connection layer integrates the feature information. It acts as a classifier of the model, converting the extracted text features into category features, and finally outputs the classification results, that is, the Softmax function is used to output the category with the highest probability as the classification result.

[0058] Figure 5 This is the ResNet time series classification model structure. The entire model consists of three residual blocks, and the number of channels in each residual block gradually increases. A single residual block consists of three convolutional layers and their corresponding batch normalization (BatchNormalization) and ReLU activation functions, and also contains a shortcut connection.

[0059] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A method for intelligent diagnosis and early warning of automobile faults based on knowledge graph, characterized in that: The steps include: (1) Obtain vehicle failure repair cases and extract data related to the failure; (2) Extract structured text from unstructured text; (3) Construct vehicle fault knowledge graph; (4) Construct a fault prediction model based on vehicle driving data; (5) Build a vehicle fault warning and cloud diagnosis system.

2. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 1 is characterized in that: The method of step (1) is as follows: obtaining online automobile maintenance cases through a crawler, screening fault extraction rules based on a proprietary dictionary of automobile parts, and extracting text data related to fault causes, fault parts, and fault solutions according to the fault extraction rules.

3. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 1 is characterized in that: The method of step (2) is to extract automobile fault structured text from unstructured text using a rule-based and TextCNN method.

4. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 3 is characterized in that: The extraction method is: 2.1 Extract fault-related sentences through automobile-related nouns, and construct fault text extraction rules based on the frequency of occurrence of context words; 2.2 Use fault extraction rules to extract fault-related sentences, and use the TextCNN model to classify the fault sentences by value; 2.3 Finally, use Jieba word segmentation to extract fault-related information in valuable sentences and complete the extraction of structured text data.

5. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 1 is characterized in that: The step (3) constructs a knowledge graph of automobile failures through neo 4-j.

6. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 5 is characterized in that: The specific construction method is to establish a connection between Python and Neo4j, use the structured automobile fault text dataset extracted from automobile fault repair cases, create corresponding entity nodes, and extract the dataset required for the construction of the automobile fault knowledge graph.

7. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 1 is characterized in that: The construction method of step (4) is: 4.1 Based on vehicle driving data, pre-process the data by splitting and reorganizing fault and non-fault data, converting missing value filling format, and classifying faults; 4.2 Filter the influencing features of vehicle faults through the retrieval method of knowledge graph, and construct the fault data set through time series extraction to provide a data basis for the construction of fault diagnosis model; 4.3 After that, MC-DCNN and ReSNet neural networks are used for fault prediction training, and the best model is used to predict vehicle faults, and a fault prediction model based on vehicle driving data is constructed.

8. The method for intelligent diagnosis and early warning of automobile faults based on knowledge graph according to claim 1 is characterized in that: The construction method of the step (5) is as follows: first, the vehicle driving data is preprocessed and feature extracted, and then fault prediction is performed. If there is no fault, the output is no fault; if there is a fault, the faulty component is input into the knowledge graph for retrieval, the cause of the fault and the maintenance method are output, and an alarm is issued.

Images