AVM picture anomaly analysis and fault tree generation method, medium and equipment

Abstract

The invention discloses an AVM picture abnormity analysis and fault tree generation method, medium and equipment, and the method comprises the steps: 1) obtaining reversing picture abnormity record information which comprises a picture abnormity type, a picture abnormity time point and a picture abnormity probability; 2) constructing a fuzzy fault tree model according to a reversing picture display system structure; 3) carrying out qualitative analysis on the picture abnormity fault tree to obtain a structural importance coefficient of each basic event of the reversing picture abnormity fault tree; 4) carrying out quantitative analysis on the picture abnormal fault tree, and calculating the fuzzy probability of occurrence of a fault tree top event and the fuzzy probability importance of a basic event; according to the method, a fuzzy fault tree analysis method is adopted to analyze abnormal pictures appearing in a reversing system, and a theoretical basis is provided for an enterprise to analyze the problem of reversing picture abnormity.

Description

Technical Field

[0001] This invention relates to the field of abnormal analysis of reversing screen in vehicle systems, and particularly to a method, medium, and device for AVM screen abnormal analysis and fault tree generation. Background Technology

[0002] During the development and maintenance of the Auto Reversing View (AVM) function in smart cockpit products, abnormal display issues often arise, such as black screens, green screens, flickering screens, cracked screens, and distorted screens. Developers typically use traditional analysis methods, relying on system logs, to address these problems. However, this approach has limitations. Existing logs usually only record direct anomalies, failing to capture potential anomalies or module-related issues. This deficiency means that the causes of some complex problems may not be reflected in the logs. After confirming the logs are correct, developers often pass the problem on to others, relying on experience to determine the issue's location, which carries the risk of errors. This cumbersome problem transfer and scheduling process ultimately makes these issues difficult to resolve, impacting project progress and user experience. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method, medium, and device for AVM screen anomaly analysis and fault tree generation. It employs fuzzy fault tree analysis to analyze abnormal screens appearing in the reversing system, providing a theoretical basis for enterprises to analyze reversing screen anomalies. This solves the problem of slow project development and difficulty in handling anomalies when there are no error records in the log systems of various modules, due to the lack of an effective way to locate the cause of the anomaly.

[0004] The objective of this invention is achieved as follows: A method for AVM screen anomaly analysis and fault tree generation, comprising the following steps: Step 1) Obtain abnormal recording information of the reversing screen, including the type of screen abnormality, the time point of screen abnormality, and the probability of screen abnormality; Step 2) Construct a fuzzy fault tree model based on the structure of the reversing screen display system. Take the screen abnormal event T as the top event. Based on the screen abnormal event as the root node of the fault tree, analyze the cause of the system fault in a logical reasoning manner. Take the most direct fault cause event after layer-by-layer subdivision as the basic event, and take all events located between the top event and the basic event as intermediate events. Use logic gates "AND", "OR" and "NOT" to connect the three types of events in sequence to form a complete fault tree model. Step 3) Perform qualitative analysis on the screen anomaly fault tree, calculate all minimum cut sets of the reversing screen anomaly events using the up-row method, and obtain the structural importance coefficient of each basic event in the reversing screen anomaly fault tree by approximate judgment of the minimum cut sets. Step 4) Quantitatively analyze the fault tree of the screen anomaly. Use the Delphi method to statistically evaluate the probability of each event by a panel of five domain experts. Combine the triangular fuzzy theory to calculate the fuzzy probability of the top event of the fault tree and the importance of the fuzzy probability of the basic events. Step 5) Determine the root cause of the abnormal event based on the results of qualitative and quantitative analysis, and obtain the fault propagation path of the abnormal event; Step 6) Construct a multi-model fusion intelligent fault diagnosis system. The system adopts a hierarchical modular design scheme, and each module exchanges data and calls functions through standardized interfaces.

[0005] Furthermore, step 2) specifically includes: Step 2.1) Treat black screen A1, distorted screen A2, and flickering screen A3 as intermediate events in the fault tree; and classify power supply X1, clock X2, driver configuration X3, buffer management X4, power supply noise X5, power-on timing X6, signal interference X7, image format X8, performance X9, and refresh rate X1 as intermediate events in the fault tree. 10 Frame Drop X 11 Synchronous X 12 As a basic event in a fault tree; Step 2.2) Connect the top event to each intermediate event using an OR gate, i.e., T = A1|A2|A3; Intermediate events are connected to basic events via AND or OR gates, i.e., A1 = X1 | X2 | X3 | X4 | ( X5 & X6 ), A2 = X3 | X4 | X7 | X8, A3 = X4 | X5 | X7 | ( X9 & X 10 ) | ( X 11 & X 12 ).

[0006] Furthermore, step 3) specifically includes: Step 3.1) Calculate the minimum cut set using the upward method, where each element is an event name: T = A1 + A2 + A3 A1 = X1 + X2 + X3 + X4 + X5X6 A2 = X3 + X4 + X7 + X8 A3 = X4 + X5 + X7 + X9 X 10 +X 11 X 12 ; Step 3.2) Determine the structural importance coefficient based on the minimum cut set approximation. The judgment principle is: if a minimum cut set contains only one basic event, then that event has the highest structural importance; all basic events in the same minimum cut set have equal structural importance. If multiple minimal cut sets contain the same number of basic events and each event occurs at the same frequency, then these events have the same structural importance; if they occur at different frequencies, the events with higher frequencies have greater structural importance, and the events with lower frequencies have less structural importance. If the frequency of occurrence of basic events is the same in different cut sets, the basic events in the minimal cut set with fewer events have higher structural importance than those in the minimal cut set with more events. For complex cases where the number of events and the frequency of occurrence are different, an approximate discriminant is used for evaluation.

[0007] Furthermore, step 4) specifically includes: Step 4.1) Obtain the probability of each event as evaluated by experts in the field, use triangular fuzzy theory to fuzzify the probability of the basic events, represent the probability of the basic events of the fault tree with fuzzy numbers, and obtain the fuzzy probability value of each basic event and the left and right fluctuation ranges α and β values. Step 4.2) Describe the AND gate fuzzy operator and the OR gate fuzzy operator using fuzzy number operation formulas; , ) ; The fuzzy probability values ​​of the top event and intermediate events are obtained using the above formula; Step 4.3) Express the fuzzy probability of the top event using symbols. Then, in the case where the m-th basic event in the fault tree does not occur while all other events occur, the fuzzy probability of the top event at this point is calculated, and this value is represented by the symbol... Indicate; finally, use symbols. This represents the difference between the two values, which is the modulo probability importance of the m-th basic event.

[0008] Furthermore, step 5) specifically includes: Step 5.1) Obtain the recorded information of abnormal events in the current screen; Step 5.2) Based on the abnormal record information, combined with the results of quantitative and qualitative analysis, the basic events that caused the top event are investigated according to their probability, the root cause of the abnormal event is found, and the fault propagation link of the abnormal event is obtained.

[0009] Furthermore, step 6) specifically includes: constructing a multi-model fusion intelligent fault diagnosis system, the system including an input module, a natural language processing module, a knowledge reasoning module, a dynamic fault tree generator module, and an output module; Step 6.1) Receive multi-source fault information through the input module, build a RESTful API interface using the FastAPI framework, receive user-initiated reversing screen fault information processing requests through the FastAPI application, extract the data in the request and parse it into a predefined Pydantic model, and call the standardization module to standardize the successfully parsed text data. Step 6.2) Use the ALBERT pre-trained model in the natural language processing module to perform semantic understanding of the fault description in the reversing screen and extract the fault entities; Step 6.3) The knowledge reasoning module uses the Neo4j graph database to record fault knowledge, uses the FCI algorithm to learn the cause-effect graph between various events of the reversing screen fault, and generates fault hypotheses through case reasoning CBR. Step 6.4) The fault tree is drawn using Graphviz based on the fault hypothesis results output by the knowledge reasoning module through the dynamic fault tree generator module. Based on online learning technology, the fault analysis performance of the system is monitored. According to the qualitative and quantitative analysis results, when the prediction accuracy is lower than the threshold, the fault tree structure adjustment is automatically triggered, including logic gate replacement and probability adjustment. Step 6.5) The output module converts the fault analysis results of the reversing screen into a format that the front-end visualization library can understand based on the output information of the dynamic fault tree generator module. The front-end visualization component is called for rendering. Finally, the template engine records the analysis results into the template and generates an HTML report based on the analysis results.

[0010] Furthermore, step 6.2) specifically includes: The natural language processing module uses the ALBERT pre-trained model to segment the standardized text data. During the vocabulary construction stage, the sub-word units are optimized using the segmentation score function formula, as shown below: ; freq(input_word) is the frequency of the word in the training corpus, and len(input_word) is the length of the word in characters; Based on a trained vocabulary, the text is decomposed into a token sequence, and the padded special tokens are mapped to IDs. Then, the sequence is padded or truncated to obtain the format of the model input. Finally, the natural language input by the user is converted into data that the machine learning model can understand and process.

[0011] Furthermore, step 6.3) specifically includes: the knowledge reasoning module receives structured fault event data from the natural language processing module, and stores fault knowledge using the Neo4j graph database, where nodes represent entities, including the type of reversing screen fault event, the probability of reversing screen fault occurrence, the scenario of reversing screen fault occurrence, information about the vehicle with reversing screen fault, and reversing screen fault cases; edges represent relationships between entities, including causal relationships between reversing screen fault events, similarity of reversing screen fault types, and reversing screen status; and the FCI algorithm is used to learn the dataset composed of data extracted from Neo4j to obtain a causal graph between various reversing screen fault events. The FCI algorithm identifies causal relationships between variables through conditional independence tests and uses the Fisher-Z test statistic to determine independence. ; Where Z0 is the Fisher-Z value associated with 0, n is the number of samples, and Z0 is the number of samples. r The value of the sample correlation coefficient r after Fisher-Z transformation; when the reversing screen malfunctions, the event of the current malfunction is extracted, and then cases containing these events are searched in Neo4j. The solutions in the historical cases are applied to the current malfunction to obtain preliminary hypotheses. The hypothesis steps are adjusted according to the actual information of the current malfunction until the malfunction conditions are met. The adjusted solution is stored in the database as a new malfunction hypothesis, realizing the generation of malfunction hypotheses through case reasoning.

[0012] The present invention also provides an AVM screen anomaly analysis and fault tree generation medium, characterized in that it stores a computer program, which, when executed by a processor, implements the above-mentioned AVM screen anomaly analysis and fault tree generation method.

[0013] The present invention also provides an AVM screen anomaly analysis and fault tree generation device, including a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) By analyzing the fuzzy fault tree, the present invention enables developers to quickly locate the basic events that cause the abnormal events when encountering abnormal problems in the reversing screen, thus avoiding the limitation of traditional analysis methods where fault problems are difficult to locate.

[0015] 2) This invention uses fuzzy fault tree to list the causes of abnormal events in the form of basic events. Analyzing the basic events can identify potential problems in the system and prevent and pay attention to the prevention of basic events, which can improve the stability of the reversing system. In contrast, traditional analysis methods only maintain system stability when problems occur.

[0016] 3) This invention provides a clear fault analysis framework through fuzzy fault tree, which explains the relationship between various modules. Some problems are caused by multiple modules, but traditional analysis methods are usually limited to the analysis of a single module, providing a theoretical basis for team collaboration.

[0017] 4) The multi-model fusion intelligent fault diagnosis system constructed in this invention can dynamically construct a fault tree based on the input fault information and correct the fault tree model based on the diagnosis results, thereby improving data accuracy. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is the overall flowchart of the present invention.

[0020] Figure 2 This is a schematic diagram of the AVM screen anomaly fault tree of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 The method for AVM screen anomaly analysis and fault tree generation shown includes the following steps: Step 1) Obtain abnormal recording information of the reversing screen, including the type of abnormality, the time point of the abnormality, and the probability of the abnormality. Based on the above information, the specific events causing the abnormality are black screen, distorted screen, and flickering screen. Black screen is related to power supply, clock, driver configuration, buffer management, power supply noise, and power-on sequence; distorted screen is related to driver configuration, buffer management, signal interference, and image format; flickering screen is related to buffer management, power supply noise, signal interference, performance, refresh rate, frame drops, and synchronization. Construct a fault tree for the reversing screen abnormality based on the relationships between the above events, such as... Figure 2 As shown in Table 1, the events causing abnormalities in the reversing camera display are listed. Table 1. Causes of Abnormal Reversing Camera Display .

[0023] Step 2) Construct a fuzzy fault tree model based on the structure of the reversing screen display system. Take the screen abnormal event T as the top event. Based on the screen abnormal event as the root node of the fault tree, analyze the cause of the system fault in a logical reasoning manner. Take the most direct fault cause event after layer-by-layer subdivision as the basic event, and take all events located between the top event and the basic event as intermediate events. Use logic gates "AND", "OR" and "NOT" to connect the three types of events in sequence to form a complete fault tree model. Step 2.1) Treat black screen A1, distorted screen A2, and flickering screen A3 as intermediate events in the fault tree; and classify power supply X1, clock X2, driver configuration X3, buffer management X4, power supply noise X5, power-on timing X6, signal interference X7, image format X8, performance X9, and refresh rate X1 as intermediate events in the fault tree. 10 Frame Drop X 11 Synchronous X 12 As a basic event in a fault tree; Step 2.2) Connect the top event to each intermediate event using an OR gate, i.e., T = A1|A2|A3; Intermediate events are connected to basic events via AND or OR gates, i.e., A1 = X1 | X2 | X3 | X4 | ( X5 & X6 ), A2 = X3 | X4 | X7 | X8, A3 = X4 | X5 | X7 | ( X9 & X 10 ) | ( X 11 & X 12 ).

[0024] Step 3) Perform qualitative analysis on the screen anomaly fault tree, calculate all minimum cut sets of the reversing screen anomaly events using the up-row method, and obtain the structural importance coefficient of each basic event in the reversing screen anomaly fault tree by approximate judgment of the minimum cut sets. Step 3.1) Calculate the minimum cut set using the upward method, where each element is an event name: T = A1 + A2 + A3 A1 = X1 + X2 + X3 + X4 + X5X6 A2 = X3 + X4 + X7 + X8 A3 = X4 + X5 + X7 + X9 X 10 +X 11 X 12 ; Step 3.2) Determine the structural importance coefficient based on the minimum cut set approximation. The judgment principle is: if a minimum cut set contains only one basic event, then that event has the highest structural importance; all basic events in the same minimum cut set have equal structural importance. If multiple minimal cut sets contain the same number of basic events and each event occurs at the same frequency, then these events have the same structural importance; if they occur at different frequencies, the events with higher frequencies have greater structural importance, and the events with lower frequencies have less structural importance. If the frequency of basic events is the same in different cut sets, the basic events in the minimal cut set with fewer events have higher structural importance than those in the minimal cut set with more events. For complex cases where both the number of events and the frequency of occurrence are different, an approximate discriminant is used for evaluation. In this embodiment, a qualitative analysis of the screen anomaly fault tree is performed. The up-row method is used to calculate all the minimum cut sets of the reversing screen anomaly events. The structural importance coefficient of each basic event in the reversing screen anomaly fault tree is obtained by approximating the minimum cut sets. Minimal cut sets: {X1}, {X2}, {X3}, {X4}, {X5}, {X7}, {X8}, {X5 X6}, {X9 X 10}、{X 11 X 12}; The structural importance coefficients are arranged in the following order: X4 > X3 = X7 > X5 > X1 = X2 = X8 > X6 = X9 = X 10 =X 11 =X 12 .

[0025] Step 4) Quantitatively analyze the fault tree of the screen anomaly. Use the Delphi method to statistically evaluate the probability of each event by a panel of five domain experts. Combine the triangular fuzzy theory to calculate the fuzzy probability of the top event of the fault tree and the importance of the fuzzy probability of the basic events. Step 4.1) Obtain the probability of each event as evaluated by experts in the field, use triangular fuzzy theory to fuzzify the probability of the basic events, represent the probability of the basic events of the fault tree with fuzzy numbers, and obtain the fuzzy probability value of each basic event and the left and right fluctuation ranges α and β values. A quantitative analysis of the fault tree of the abnormal image was performed. Table 2 shows the probability of each event occurring as subjectively evaluated by a panel of five domain experts using the Delphi method. Table 2 Fuzzy Probability Values ​​of Basic Events .

[0026] Step 4.2) Describe the AND gate fuzzy operator and the OR gate fuzzy operator using fuzzy number operation formulas; , ) ; The fuzzy probabilities of the top event and intermediate events are obtained according to the above formula, as shown in Table 3. Table 3. Fuzzy Probabilities of Top Event and Intermediate Event ; Step 4.3) Express the fuzzy probability of the top event using symbols. Then, in the case where the m-th basic event in the fault tree does not occur while all other events occur, the fuzzy probability of the top event at this point is calculated, and this value is represented by the symbol... Indicate; finally, use symbols. The difference between the two values ​​represents the modulo probability importance of the m-th basic event, as shown in Table 4. Table 4 Importance of Fuzzy Probabilities of Basic Events .

[0027] Step 5) Determine the root cause of the abnormal event based on the results of qualitative and quantitative analysis, and obtain the fault propagation path of the abnormal event; Step 5.1) Obtain the recorded information of abnormal events in the current screen; Step 5.2) Based on the abnormal record information, combined with the results of quantitative and qualitative analysis, the basic events that caused the top event are investigated according to their probability, the root cause of the abnormal event is found, and the fault propagation link of the abnormal event is obtained. Given the known anomaly in the reversing camera display, the probability of each basic event occurring, as determined by the above analysis, is as follows: X4 > X3 > X7 > X1 > X5 > X2 > X6 > X11 = X12 > X9 = X10, therefore it can be inferred that the most likely cause of abnormal reversing screen is buffer management, followed by driver configuration.

[0028] Step 6) Construct a multi-model fusion intelligent fault diagnosis system. The system adopts a hierarchical modular design scheme, and each module exchanges data and calls functions through standardized interfaces. A multi-model fusion intelligent fault diagnosis system is constructed, comprising an input module, a natural language processing module, a knowledge reasoning module, a dynamic fault tree generator module, and an output module. Step 6.1) Receive multi-source fault information through the input module, build a RESTful API interface using the FastAPI framework, receive user-initiated reversing screen fault information processing requests through the FastAPI application, extract the data in the request and parse it into a predefined Pydantic model, and call the standardization module to standardize the successfully parsed text data. Step 6.2) Use the ALBERT pre-trained model in the natural language processing module to perform semantic understanding of the fault description in the reversing screen and extract the fault entities; The natural language processing module uses the ALBERT pre-trained model to segment the standardized text data. During the vocabulary construction stage, the sub-word units are optimized using the segmentation score function formula, as shown below: freq(input_word) is the frequency of the word in the training corpus, and len(input_word) is the length of the word in characters. Based on a trained vocabulary, the text is decomposed into a token sequence, and the padded special tokens are mapped to IDs. Then, the sequence is padded or truncated to obtain the format of the model input. Finally, the natural language input by the user is converted into data that the machine learning model can understand and process.

[0029] Step 6.3) The knowledge reasoning module uses the Neo4j graph database to record fault knowledge, uses the FCI algorithm to learn the cause-effect graph between various events of the reversing screen fault, and generates fault hypotheses through case reasoning CBR. The knowledge reasoning module receives structured fault event data from the natural language processing module and stores fault knowledge using the Neo4j graph database. Nodes represent entities, including the type of reversing screen fault event, the probability of the reversing screen fault occurring, the scenario in which the reversing screen fault occurs, information about the vehicle with the reversing screen fault, and case studies of the reversing screen fault. Edges represent relationships between entities, including causal relationships between reversing screen fault events, similarity between reversing screen fault types, and reversing screen status. The FCI algorithm is used to learn the dataset composed of data extracted from Neo4j, resulting in a causal graph between various reversing screen fault events. The FCI algorithm identifies causal relationships between variables through conditional independence tests and uses the Fisher-Z test statistic to determine independence. ; Where Z0 is the Fisher-Z value associated with 0, n is the number of samples, and Z0 is the number of samples. rThe value of the sample correlation coefficient r after Fisher-Z transformation; when the reversing screen malfunctions, the event of the current malfunction is extracted, and then cases containing these events are searched in Neo4j. The solutions in the historical cases are applied to the current malfunction to obtain preliminary hypotheses. The hypothesis steps are adjusted according to the actual information of the current malfunction until the malfunction conditions are met. The adjusted solution is stored in the database as a new malfunction hypothesis, realizing the generation of malfunction hypotheses through case reasoning.

[0030] Step 6.4) The fault tree is drawn using Graphviz based on the fault hypothesis results output by the knowledge reasoning module through the dynamic fault tree generator module. Based on online learning technology, the fault analysis performance of the system is monitored. According to the qualitative and quantitative analysis results, when the prediction accuracy is lower than the threshold, the fault tree structure adjustment is automatically triggered, including logic gate replacement and probability adjustment. The dynamic fault tree generator module receives system fault data and prior knowledge, uses Graphviz to graphically describe the relationships between various fault events, adjusts parameters based on the latest data using online learning technology, and monitors the accuracy of fault prediction in real time through qualitative and quantitative analysis. Qualitative analysis evaluates the correctness and feasibility of the fault tree structure, while quantitative analysis assesses the performance of the prediction model by calculating metrics such as prediction accuracy and error rate. If the prediction accuracy fails to reach a preset threshold, automatic adjustment of the fault tree is triggered.

[0031] Step 6.5) Based on the output information of the dynamic fault tree generator module, the output module converts the fault analysis results of the reversing screen into a format that the front-end visualization library can understand, calls the front-end visualization component for rendering, and finally uses the template engine to record the analysis results into the template and generate an HTML report based on the analysis results. The generated report will include the following: a visual structure diagram of the fault tree, a detailed description of each fault event and logic gate, and the occurrence probability and fault risk analysis results of each event.

[0032] This invention provides an AVM screen anomaly analysis and fault tree generation method, medium, and device. By analyzing the fuzzy fault tree, developers can quickly locate the basic events that cause the anomaly when encountering a reversing screen anomaly, avoiding the limitation of traditional analysis methods where fault problems are difficult to locate. Fuzzy fault trees list the causes of abnormal events as basic events. Analyzing these basic events can identify potential problems in the system and prevent and address them, thereby improving the stability of the reversing system. In contrast, traditional analysis methods only maintain system stability when problems occur. Fuzzy fault trees provide a clear framework for fault analysis, illustrating the relationships between various modules. Some problems are caused by multiple modules, but traditional analysis methods are usually limited to the analysis of a single module, providing a theoretical basis for team collaboration. The intelligent fault diagnosis system that integrates multiple models can dynamically construct fault trees based on the input fault information and correct the fault tree model based on the diagnosis results, effectively improving data accuracy.

[0033] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. An AVM picture anomaly analysis and fault tree generation method, characterized in that, The method comprises the following steps: Step 1) acquiring the reverse image abnormality record information, including the image abnormality type, the image abnormality time point and the image abnormality probability; Step 2) constructing a fuzzy fault tree model according to the structure of the reverse image display system, taking the image abnormality event T as the top event, analyzing the causes of system failure in a logical reasoning manner on the basis that the image abnormality event is the root node of the fault tree, taking the most direct fault cause event after layer-by-layer subdivision as a basic event, taking all events between the top event and the basic event as intermediate events, and connecting the three kinds of events in sequence by using the logical gates "and", "or" and "not" symbols to form a complete fault tree model; Step 3) performing qualitative analysis on the image abnormality fault tree, calculating all the minimal cut sets of the reverse image abnormality event by using the uplink method, and obtaining the structural importance coefficients of each basic event of the reverse image abnormality fault tree through the minimal cut set approximation judgment; Step 4) performing quantitative analysis on the image abnormality fault tree, subjectively evaluating the probabilities of each event by using the Delphi method to statistically evaluate the probabilities of each event by a group of five experts in the field, and combining the triangular fuzzy theory to calculate the fuzzy probability of the top event of the fault tree and the fuzzy probability importance of the basic event; Step 5) determining the root cause of the abnormal event according to the results of the qualitative analysis and the quantitative analysis, and obtaining the fault propagation link of the abnormal event; Step 6) constructing an intelligent fault diagnosis system based on multi-model fusion, and the system adopts a hierarchical modular design scheme, and each module exchanges data and calls functions through standardized interfaces.

2. The AVM picture anomaly analysis and fault tree generation method of claim 1, wherein, The step 2) specifically comprises: Step 2.1) Black screen A1, garbled screen A2, flashing screen A3 as intermediate events in the fault tree; power supply X1, clock X2, drive configuration X3, Buffer management X4, power supply noise X5, power-on timing X6, signal interference X7, image format X8, performance X9, refresh rate X 10 , frame loss X 11 , synchronization X 12 as basic events in the fault tree; Step 2.2) connecting the top event and each intermediate event through an or gate, i.e. T = A1| A2 | A3; The intermediate events are connected to the individual basic events by means of AND gates, i.e. A1 = X1 | X2 | X3 | X4 | (X5 & X6), A2 = X3 | X4 | X7 | X8, A3 = X4 | X5 | X7 | (X9 & X10) | (X11 & X12). 10 11 12 ​​​ 3. The method of claim 1, wherein the method further comprises: The step 3) specifically comprises: Step 3.1) calculating the minimal cut sets according to the uplink method, wherein each element is an event name: T = A1+ A2 +A3 A1 = X1 + X2 +X3 + X4+X5X6 A2= X3+X4 +X7 + X8 A3 = X4 + X5 + X7 + X9 X 10 +X 11 X 12 ; Step 3.2) obtaining the structural importance coefficients according to the minimal cut set approximation judgment, and the judgment principle is that if a minimal cut set contains only one basic event, the event has the maximum structural importance; all basic events in the same minimal cut set have equal structural importance; if multiple minimal cut sets contain the same number of basic events and the number of occurrences of each event is consistent, the structural importance of these events is the same; if the number of occurrences is different, the higher the number, the greater the structural importance, and the lower the number, the smaller the structural importance; if the basic events in different cut sets have the same occurrence frequency, the basic events in the minimal cut set with fewer events have higher structural importance than the minimal cut set with more events; for complex situations where the number of events and the number of occurrences are different, the approximation discriminant is used for evaluation.

4. The AVM picture anomaly analysis and fault tree generation method of claim 1, wherein, The step 4) specifically comprises: Step 4.1) obtaining the probabilities of each event evaluated by experts in the field, and using the triangular fuzzy theory to fuzz the probabilities of the basic events, representing the probabilities of the basic events of the fault tree by fuzzy numbers to obtain the fuzzy probability values of each basic event and the left and right fluctuation intervals α and β values; Step 4.2) The AND gate fuzzy operator and the OR gate fuzzy operator are described by the fuzzy number operation formula. , ) The fuzzy probability values of the top event and the intermediate event are obtained according to the above formula; Step 4.3) The fuzzy probability of the top event is denoted by , then the fuzzy probability of the top event is calculated under the condition that the mth basic event does not occur while the other events occur, which is denoted by ; finally, the difference between the two is denoted by , which is the fuzzy probability importance of the mth basic event.

5. The AVM picture anomaly analysis and fault tree generation method of claim 1, wherein, The step 5) specifically comprises: Step 5.1) obtaining the record information of the current picture abnormal event; Step 5.2) according to the abnormal record information, combining the quantitative analysis and qualitative analysis results, the basic events causing the top event are investigated according to the probability size, and the root cause of the abnormal event is found out, and the fault propagation link of the abnormal event is obtained.

6. The AVM picture anomaly analysis and fault tree generation method of claim 1, wherein, The step 6) specifically comprises: constructing a multi-model fusion intelligent fault diagnosis system, the system comprises an input module, a natural language processing module, a knowledge reasoning module, a dynamic fault tree generator module and an output module; Step 6.1) receiving multi-source fault information through the input module, using FastAPI framework to build RESTful API interface, receiving user initiated reverse picture fault information processing request through FastAPI application, extracting data in the request and parsing into predefined Pydantic model, calling standardization module to standardize the parsed text data; Step 6.2) through the natural language processing module, using ALBERT pre-training model to understand the semantics of the fault description of the reverse picture and extract fault entities; Step 6.3) through the knowledge reasoning module, using Neo4j graph database to record fault knowledge, using FCI algorithm to learn the causal graph between each event of the reverse picture fault, and generating fault hypothesis through case-based reasoning CBR; Step 6.4) through the dynamic fault tree generator module, using Graphviz to draw fault tree according to the fault hypothesis result output by the knowledge reasoning module, and based on online learning technology, monitoring the fault analysis performance of the system, according to the qualitative analysis and quantitative analysis results, when the prediction accuracy is lower than the threshold, automatically triggering fault tree structure adjustment, including logic gate replacement and probability adjustment; Step 6.5) through the output module, according to the output information of the dynamic fault tree generator module, the fault analysis result of the reverse picture is converted into a format that can be understood by the front-end visualization library, and the front-end visualization component is called for rendering, finally using template engine to record the analysis result in the template and generating HTML report according to the analysis result.

7. The method of claim 6, wherein the AVM picture abnormality analysis and fault tree generation method is characterized by, The step 6.2) specifically comprises: The natural language processing module uses ALBERT pre-training model to segment the standardized text data, and the vocabulary table construction stage optimizes the sub-word unit through the segmentation score function formula, as shown in the formula below: ； Freq(input_word) is the frequency of the word appearing in the training corpus, and len(input_word) is the character length of the word; Based on the trained vocabulary table, the text is decomposed into token sequence, and the filled special token is mapped to ID, then the sequence is filled or truncated, and the format of the model input is obtained, finally the user input natural language is converted into data that can be understood and processed by machine learning model.

8. The method of claim 6, wherein the AVM picture abnormality analysis and fault tree generation method is characterized by, The step 6.3) specifically comprises: the knowledge reasoning module receives the structured fault event data from the natural language processing module, and stores the fault knowledge by using a Neo4j graph database, wherein a node represents an entity, and there are a reversing picture fault event type, a probability of occurrence of the reversing picture fault, a scene of occurrence of the reversing picture fault, information of a reversing picture fault vehicle, and a reversing picture fault case; an edge represents a relationship between entities, and there are a reversing picture fault event causal relationship, a similarity of reversing picture fault types, and a reversing picture state; a data set composed of data proposed from the Neo4j is learned by using an FCI algorithm, and a causal graph between each event of the reversing picture fault is obtained. The FCI algorithm identifies causal relationships between variables by conditional independence tests, using the Fisher-Z test statistic for independence: ; wherein Z0 is the Fisher-Z value of 0 correlation, n is the sample number, Z r is the Fisher-Z transformed value of sample correlation coefficient r; when the reverse picture failure occurs, the current fault event is extracted, then the cases containing these events are found in Neo4j, the solution in the historical cases is applied to the current fault to obtain a preliminary hypothesis, the hypothesis is adjusted according to the actual information of the current fault until the fault condition is met, and the adjusted solution is stored in the database as a new fault hypothesis, so that the fault hypothesis is generated through case reasoning.

9. An AVM picture anomaly analysis and fault tree generation medium, characterized by, A computer program is stored, and the computer program is executed by a processor to implement the AVM picture abnormality analysis and fault tree generation method in any one of claims 1-8.

10. An AVM picture anomaly analysis and fault tree generation device, characterized by, A computer program is stored, and the computer program is executed by a processor to implement the AVM picture abnormality analysis and fault tree generation method in any one of claims 1-8.

Images