A method for mining vehicle ECU fault correlation
By using vehicle ECU fault correlation mining methods and data cleaning, modeling, and information fusion technologies, the problem of rapid and accurate location of multiple ECU faults in intelligent connected vehicles was solved, achieving efficient fault source location and shortening repair time.
Patent Information
- Application Number
- CN202211367573.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-03
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-11-03
AI Technical Summary
Existing technologies cannot efficiently and quickly locate the correlation between multiple ECU faults in intelligent connected vehicles, resulting in wasted repair time and inaccurate location.
We employ a vehicle ECU fault correlation mining method. Through data cleaning and modeling, we calculate the support and association rules of frequent itemsets, use lattice structure pruning to optimize the calculation, and combine forward, backward and hybrid reasoning to perform fault diagnosis using multi-sensor information fusion.
It improves the efficiency of ECU fault location, shortens the overall vehicle repair time, and enhances the accuracy of fault detection and repair efficiency.
Smart Images

Figure CN115599075B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent networked vehicles, and in particular to a method for mining vehicle ECU fault association relationships. Background Art
[0002] Association rules reflect the knowledge that one event is interdependent or associated with other events. If there is a certain association between two or more attributes, then the attribute value of one of them can be predicted by the other attribute values. The initial application of association rules was in the market field. Large supermarkets collected and stored a large number of sales records through cash registers. Each record stored the transaction time, name of the transaction item, quantity and amount of the transaction items, etc. These records imply certain association rules. For example, among customers who purchased product A, 70% of the customers also purchased product B. These association rules can guide shopping mall managers to implement appropriate sales strategies, thereby increasing sales. With the continuous expansion of application fields, researchers are also continuously conducting in-depth research on association rules. The object of mining has gradually changed from the initial mining of Boolean data to the transition of mining methods for numerical data, and from serial algorithms to parallel algorithms.
[0003] With the rapid development of the Internet, the Internet of Things and other fields, the vehicle field has also entered the electronic and Internet of Things fields. With the rapid development of the Internet of Things, ECUs have gradually been added to various vehicle components. Each component is controlled and managed by its own ECU. When a fault occurs, the corresponding fault can be found by checking the ECU's fault code to quickly locate and repair the problem. Currently, intelligent connected vehicles can remotely report faults to the cloud platform, and users can obtain vehicle fault information through mobile phone software, text messages, public accounts, etc. The faults reported by the current ECU cannot be completely and accurately located. When multiple ECUs report faults, the fault of one ECU may cause the fault of other ECUs, which brings a lot of trouble to the location of the problem.
[0004] When a vehicle has multiple ECU faults, the maintenance personnel can only rely on their experience to make judgments, and cannot perform scientific and effective fault location quickly. This problem location method based on experience is not always effective, and it will waste a lot of time checking error codes caused by other non-fault points. In this case, maintenance personnel with less experience need to spend a lot of time checking each component, wasting a lot of time and unable to obtain accurate fault location.
[0005] Therefore, there is a need for an efficient method that can quickly find the correlation between multiple ECU faults and find the source of the fault based on the correlation, so as to reduce the overall maintenance time of the vehicle and improve the efficiency of maintenance personnel in locating the fault. Summary of the invention
[0006] (I) Technical solution
[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a vehicle ECU fault association relationship mining method, the specific mining method includes two parts, the first part is to clean the existing data and perform data modeling, the second part is to calculate the support of all frequent item sets and calculate the association rules generated by all frequent item sets respectively, the data modeling uses the data set reported remotely by the vehicle, the vehicle fault data of each vehicle is cleaned, the set of all faults generated in the time period before the fault is eliminated takes its maximum set as the frequent item set, its subset is not included in the data item, and any fault is eliminated and recalculated, and the maximum set after any fault is eliminated is taken as the frequent item set, until no subsequent fault is eliminated or no fault is eliminated.
[0008] Preferably, the steps for calculating the support and association rules for all the frequent item sets are as follows:
[0009] First, the terms used in this invention are explained. Frequent item sets refer to a set of multiple fault codes that often appear together. Support is also called coverage, which is the percentage of the data set that contains this item set. support(A→B)=P(A∪B), support(A→B) is the possibility that attribute A simultaneously generates attribute B. The support attribute is used to quantitatively evaluate the frequent coverage of frequent item sets. The confidence is
[0010] Confidence is used to quantitatively evaluate the accuracy of frequent itemsets, and confidence(A→B) is the credibility of attribute A generating attribute B at the same time.
[0011] Furthermore, the above method of calculating all frequent item sets by traversal, that is, using a lattice structure to traverse all item sets, cannot be used in the actual production process because there are too many combinations of frequent item sets and the amount of calculation is too large. It is necessary to perform pruning in the process of calculating frequent item sets to improve the calculation efficiency. The pruning method is as follows:
[0012] Calculate the support and confidence of this frequent item set, filter the frequent item sets whose support is greater than or equal to the minimum support, and obtain the frequent item sets that meet the minimum support; the association rules should be filtered according to the confidence greater than or equal to the minimum confidence, and obtain the rules that meet the minimum confidence;
[0013] The above pruning rules can greatly reduce the amount of calculation, providing effective protection for actual production use.
[0014] Furthermore, in the item set {A, B, C, D}, according to the lattice structure frequent item sets, there are frequent item sets of the first layer {A, B}, {A, C}, {A, D}, {B, C}, {B, D}, {C, D}, the second layer {A, B, C}, {A, B, D}, {A, C, D}, {B, C, D}, and the third layer {A, B, C, D}. All the support and confidence are calculated based on all the above frequent item sets.
[0015] Furthermore, according to the Apriori law, a subset of a frequent item set is also a frequent item set, and a superset of an infrequent item set is not a frequent item set either, and pruning can be performed to reduce a large number of invalid computing tasks.
[0016] Furthermore, the itemsets have a total of data sets [3], [1,3], [1,3,4], [2,3,5], [1,2,3,5], [2,5], [1,3,4], [2,4,5]. According to the calculation method of frequent itemsets, we can get {support([5]): 0.5, support([3]): 0.75, support([1,2]): 0.125, support([1,5]): 0.125, support([3,5]): 0.25, support([4]): 0.375, support([2,3]): 0.25, support([2,5 ]):0.5,support([1]):0.5,support([1,3]):0.5,support([2]):0.5}, According to the confidence calculation method, when the minimum confidence value is set to 0.5, we can obtain the rule [(support([3]),support([1]),0.666666666666666666),(support([1]),support([3]),1.0),(support([5]),support([2]),1.0),(support([2]),support([5]),1.0)].
[0017] Furthermore, according to the above rules, the probability that attribute 3 will cause attribute 1 failure is 66.66%, the probability that attribute 1 will cause attribute 3 is 100%, the probability that attribute 5 will cause attribute 2 failure is 100%, and the probability that attribute 1 will cause attribute 5 failure is 100%. Therefore, it can be concluded that attributes 2, 5 and attributes 1, 3 have a strong correlation, and the occurrence of attribute 1 will inevitably cause attribute 3 failure, and attributes 2 and attributes 5 appear in pairs.
[0018] Furthermore, the data modeling also includes a knowledge source, an inference engine, an interpretation system, and a human-computer interface. The knowledge source includes a knowledge base, a model base, and a database. The knowledge base stores expert knowledge, experiential knowledge, and common sense. The model base stores mathematical models that describe the state and mechanism of the diagnosis object. The database stores the state data of the diagnosis object measured in real time, as well as various information used for the reasoning process.
[0019] Furthermore, the inference engine uses various rules to perform reasonable inferences based on the acquired information, obtains inference results, and outputs diagnostic conclusions. The inference strategies include forward inference, reverse inference, and forward and reverse mixed inference.
[0020] The forward reasoning is to first process the original data and infer the conclusion based on the knowledge of experts in the knowledge base;
[0021] The reverse reasoning: first propose a hypothesis based on the possible conclusion, then reversely look for evidence to support the conclusion layer by layer, and make a judgment on the hypothesis based on the evidence;
[0022] The reverse hybrid reasoning is to organically combine forward reasoning and reverse reasoning, usually using the method of first forward and then reverse;
[0023] The explanation system is a system that provides necessary answers according to the needs of users, such as displaying the reasoning process, explaining instructions, etc.
[0024] The human-machine interface is the interface between the fault diagnosis personnel and the system.
[0025] Furthermore, the support of the frequent itemsets can be used to collect signals from different channels through multiple sensors, and different feature information can be obtained from the same signal, avoiding biased diagnostic conclusions from different diagnostic approaches. The accuracy of diagnosis is greatly improved. Therefore, information fusion technology can be applied to fault diagnosis. The above is the process of automatically analyzing and synthesizing information from multiple sensors according to certain rules to complete the required judgment and decision-making.
[0026] (II) Beneficial effects
[0027] The present invention provides a method for mining the association relationship of vehicle ECU faults. The method has the following beneficial effects: the present invention proposes a method for mining the association relationship of vehicle ECU faults, which can efficiently find the association relationship between multiple ECU faults, and find their source faults according to their association relationship, thereby improving the efficiency of locating vehicle ECU faults and shortening the overall maintenance time of the vehicle. Compared with the existing troubleshooting method based on experience, the present invention finds the association relationship between ECU faults and their source faults with higher computing efficiency, thereby improving the efficiency of locating vehicle ECU faults and shortening the overall maintenance time of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 This is a diagram showing the relationship between all frequent itemsets in the traditional itemset {A, B, C, D};
[0029] Figure 2 Schematic diagram of lattice structure pruning optimization of all frequent itemsets of the item set {A, B, C, D} of the present invention. DETAILED DESCRIPTION
[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0031] Embodiment 1:
[0032] like Figure 1 and 2 As shown, an embodiment of the present invention provides a method for mining the association relationship of vehicle ECU faults, wherein the vehicle fault data of each vehicle is cleaned, the maximum set of all faults generated in the time period before the fault is eliminated is taken as the data set, and its subset is not included in the data set, and recalculates after any fault is eliminated, and takes the maximum set after any fault is eliminated as the data set, until no subsequent fault is eliminated or no fault is found;
[0033] The attributes of vehicle fault data are numbered as {1,2,3,4,5}. Some data in production are desensitized and used as examples to demonstrate the actual calculation process. The data in the examples include [3], [1,3], [1,3,4], [2,3,5], [1,2,3,5], [2,5], [1,3,4], [2,4,5]. According to the calculation method of frequent itemsets, we can get {support([5]): 0.5, support([3]): 0.75, support([1,2]): 0.125, support([1,5]): 0.125 ,support([3,5]):0.25,support([4]):0.375,support([2,3]):0.25,support([2,5]):0.5,support([1]):0.5,support([1,3]):0.5,support([2]):0.5}, According to the confidence calculation method, when the minimum confidence value is set to 0.5, the rule [(support([3]),support([1]),0.66666666666666666),(support([1]),sup port([3]),1.0),(support([5]),support([2]),1.0),(support([2]),support([5]),1.0)], according to the above rules, the probability that attribute 3 will cause attribute 1 failure is 66.66%, the probability that attribute 1 will cause attribute 3 failure is 100%, the probability that attribute 5 will cause attribute 2 failure is 100%, and the probability that attribute 1 will cause attribute 5 failure is 100%. Therefore, attributes 2, 5 and attributes 1, 3 have a strong correlation, and the appearance of attribute 1 will inevitably cause attribute 3 failure. Attributes 2 and 5 appear in pairs, and the minimum confidence value is When it is set to 0.7, we can get the rules [(support([1]), support([3]), 1.0), (support([5]), support([2]), 1.0), (support([2]), support([5]), 1.0)]. According to the above rules, we can get that the probability of attribute 1 generating attribute 3 is 100%, the probability of attribute 5 generating attribute 2 failure is 100%, and the probability of attribute 1 generating attribute 5 failure is 100%. Therefore, attributes 2, 5 and attributes 1, 3 have a strong correlation, and the appearance of attribute 1 will inevitably cause attribute 3 failure. Attributes 2 and 5 appear in pairs.
[0034] According to the above correlation between fault attributes, it can be used in actual production. When attribute 1 and 3 faults appear in pairs, the fault of attribute 1 can be directly checked, thereby reducing the time for troubleshooting and improving the accuracy of fault location, and also reducing the time for fault detection and improving the efficiency of vehicle maintenance. The present invention uses Windows 10 version 190918363.1556 and processor 11th Gen Intel (R) Core (TM) i7-1185G7 @ 3.00GHz 1.80GHz for calculation;
[0035] The data modeling also includes a knowledge source, an inference engine, an interpretation system, and a human-computer interface. The knowledge source includes a knowledge base, a model base, and a database. The knowledge base stores expert knowledge, empirical knowledge, and common sense. The model base stores mathematical models describing the state and mechanism of the diagnosis object. The database stores the state data of the diagnosis object measured in real time, as well as various information used in the inference process.
[0036] The inference engine uses various rules to perform reasonable inferences based on the acquired information, obtains inference results, and outputs diagnostic conclusions. The inference strategies include forward inference, reverse inference, and forward and reverse mixed inference.
[0037] The forward reasoning is to first process the original data and infer the conclusion based on the knowledge of experts in the knowledge base;
[0038] The reverse reasoning: first propose a hypothesis based on the possible conclusion, then reversely look for evidence to support the conclusion layer by layer, and make a judgment on the hypothesis based on the evidence;
[0039] The reverse hybrid reasoning is to organically combine forward reasoning and reverse reasoning, usually using the method of first forward and then reverse;
[0040] The explanation system is a system that provides necessary answers according to the needs of users, such as displaying the reasoning process, explaining instructions, etc.
[0041] The human-machine interface is the interface between the fault diagnosis personnel and the system;
[0042] The support of the frequent item set can be collected by multiple sensors from different channels, and different feature information can be obtained from the same signal, so as to avoid biased diagnosis conclusions drawn by different diagnostic approaches, and the accuracy of diagnosis is greatly improved. Therefore, information fusion technology can be applied to fault diagnosis;
[0043] Among them, support and confidence are important criteria for measuring whether the obtained association rules are meaningful. The size of support reflects the importance of association rules in the data set. What we need to mine are rules with higher than the minimum support. If the support of the rule is less than the minimum support, it means that the mined rules are meaningless and not general. The size of confidence is used to indicate the credibility of the association rules. If the confidence of the rule is less than the minimum confidence, it means that the mined rules are unreliable and have no reference value. The purpose of data mining is to mine valuable rules that are generally representative and highly reliable, so as to achieve the goal of discovering useful knowledge and ultimately meet the needs of users.
[0044] Furthermore, the present invention proposes a vehicle ECU fault correlation mining method, which can efficiently find the correlation between multiple ECU faults and find their source faults according to their correlation, thereby improving the efficiency of locating vehicle ECU faults and shortening the overall vehicle maintenance time;
[0045] Unless otherwise defined, all terms used in this specification have the same meaning as understood by persons skilled in the art to which the present invention belongs. The terms used in this specification are only used to describe the embodiments of the present invention and are not intended to limit the present invention.
[0046] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A vehicle ECU fault association mining method, the specific mining method includes two parts, the first part is to clean the existing data and perform data modeling, the second part is to calculate the support of all frequent item sets and calculate the association rules generated by all frequent item sets, the characteristics are: The data modeling uses the data set reported remotely by the vehicle, cleans the vehicle fault data of each vehicle, takes the maximum set of all faults generated in the time period before the fault is eliminated as the frequent item set, and its subset is not included in the data item, and recalculates after any fault is eliminated, and takes the maximum set after any fault is eliminated as the frequent item set, until no subsequent fault is eliminated or there is no fault; The method of calculating the support of all frequent item sets and the association rules generated by all frequent item sets includes: using the support attribute to quantitatively evaluate the frequent coverage of frequent item sets, wherein the support is also called coverage, which is the percentage of the item set included in the data set, support(A→B)=P(A∪B), support(A→B) is the property that attribute A generates attribute B at the same time, The confidence level is: , Confidence is used to quantitatively evaluate the accuracy of frequent itemsets. Confidence(A→B) is the credibility of attribute A generating attribute B at the same time. Specifically, the vehicle fault data attributes are numbered, and all vehicle fault attributes are numbered as data {1, 2, 3, 4, 5}. The item sets include the data sets [3], [1, 3], [1, 3, 4], [2, 3, 5], [1, 2, 3, 5], [2, 5], [1, 3, 4], [2, 4, 5]. According to the calculation method of frequent item sets, we can get {support([5]): 0.5, support([3]): 0.75, support([1, 2]): 0.125, support([1, 5]): 0.125, support([3, 5]): 0.25, support([4]): 0.375, support([2, 3]): 0.25, support([2, 5]): 0.5, support([1]):0.5, support([1, 3]):0.5, support([2]):0.5}, according to the confidence calculation method, when the minimum confidence value is set to 0.5, we can obtain the rules [(support([3]), support([1]), 0.666666666666666666), (support([1]), support([3]), 1.0), (support([5]), support([2]), 1.0), (support([2]), support([5]), 1.0)].
2. The method according to claim 1, characterized in that: When calculating all frequent item sets through the traversal method, it is necessary to perform pruning in the process of calculating frequent item sets to improve the calculation efficiency. The pruning method is as follows: Calculate the support and confidence of this frequent item set, filter out the frequent item sets with support greater than the minimum support, and obtain the frequent item sets that meet the minimum support. The association rules should be filtered according to the confidence greater than the minimum confidence to obtain the rules that meet the minimum confidence.
3. The method according to claim 1 or 2, characterized in that: In the item set {A,B,C,D}, according to the lattice structure frequent itemsets have the first layer of frequent itemsets {A,B}, {A,C}, {A,D}, {B,C}, {B,D}, {C,D}, the second layer {A,B,C}, {A,B,D}, {A,C,D}, {B,C,D}, and the third layer {A,B,C,D}. All support and confidence are calculated based on all the above frequent itemsets.
4. The method according to claim 3, characterized in that: According to the Apriori law, a subset of a frequent item set is also a frequent item set, and a superset of an infrequent item set is not a frequent item set. Pruning can be performed to reduce a large number of invalid computing tasks.
5. The method according to claim 1, characterized in that: According to the above rules, the probability that attribute 3 causes attribute 1 failure is 66.66%, the probability that attribute 1 causes attribute 3 is 100%, the probability that attribute 5 causes attribute 2 failure is 100%, and the probability that attribute 1 causes attribute 5 failure is 100%. It can be concluded that attributes 2, 5 and attributes 1, 3 have a strong correlation, and the appearance of attribute 1 will inevitably cause the failure of attribute 3, and attributes 2 and 5 appear in pairs.
6. The method according to claim 1, characterized in that: The data modeling also includes knowledge source, inference engine, interpretation system, and human-computer interface. The knowledge source includes knowledge base, model base and database. The knowledge base stores expert knowledge, experience knowledge and common sense. The model base stores mathematical models that describe the state and mechanism of the diagnosis object. The database stores the state data of the diagnosis object measured in real time, as well as various information used for the reasoning process.
7. The method according to claim 6, characterized in that: The inference engine uses various rules to perform reasonable inferences based on the acquired information, obtains inference results, and outputs diagnostic conclusions. The inference strategies include forward inference, reverse inference, and forward and reverse mixed inference. The forward reasoning is to first process the original data and infer the conclusion based on the knowledge of experts in the knowledge base; The reverse reasoning: first propose a hypothesis based on the possible conclusion, then reversely look for evidence to support the conclusion layer by layer, and make a judgment on the hypothesis based on the evidence; The reverse hybrid reasoning is to organically combine forward reasoning and reverse reasoning, using the method of first forward and then reverse; The explanation system is a system that provides necessary answers according to the needs of users; The human-machine interface is the interface between the fault diagnosis personnel and the system.
8. The method according to claim 1, characterized in that: The support of the frequent itemsets can be obtained by collecting signals from different channels through multiple sensors, and different feature information can be obtained from the same signal, thereby avoiding biased diagnosis conclusions drawn by different diagnostic approaches.
