CN115525465A - Fault point prediction method and system based on multiple failure analysis - Google Patents

Abstract

The invention relates to a fault point prediction method and a fault point prediction system based on multiple failure analysis, belongs to the technical field of failure analysis, and solves the problems that the existing failure analysis is not comprehensive and has low accuracy. Analyzing historical maintenance records to obtain abnormal points in fault points, deleting detection steps related to the abnormal points based on parts related to the detection steps and the lines to which the parts belong, supplementing detection steps related to other parts belonging to the same line, and obtaining actual detection steps; performing failure analysis according to the diagnostic data of the assembly part to be detected, if the failure analysis is primary failure analysis, acquiring an actual detection step, and if not, updating the actual detection step according to the fault point of the last failure analysis; obtaining a fault point according to the actual detection value, if the fault point is correct, ending failure analysis, otherwise, iterating until the maximum failure analysis times is reached, and obtaining a final fault point; and if the final fault point is incorrect, predicting the fault point according to the trained classification model. The accuracy of failure analysis is improved.

Description

Fault point prediction method and system based on multiple failure analysis

Technical Field

The invention relates to the technical field of failure analysis, in particular to a fault point prediction method and a fault point prediction system based on multiple failure analysis.

Background

The failure analysis is generally based on failure modes and phenomena, and through analysis and verification, the phenomenon of failure is simulated and reproduced, the reason of failure is found out, and the activity of the failure mechanism is excavated. Failure analysis has strong practical significance in the aspects of improving product quality, developing and improving technology, repairing products, arbitrating failure accidents and the like. With the large amount of equipment, the storage capacity of the industrial failure analysis data will grow exponentially.

Generally, a fault industrial assembly part needs to be subjected to initial diagnosis by a test machine to obtain a fault phenomenon, then a corresponding failure analysis standard flow is found for failure analysis, then a fault point is found, the fault point is maintained, and then whether the maintenance is good or not is judged, if the maintenance is good, the process is ended. If the maintenance is not good, the initial diagnosis process needs to be carried out again to determine the basic fault phenomenon, and then circulation is carried out until failure analysis and maintenance are not carried out for a certain number of times.

The existing failure analysis standard process has more steps, can not quickly and accurately find a failure point, has the same standard process adopted by multiple failure analysis, and has low analysis accuracy.

Disclosure of Invention

In view of the foregoing analysis, embodiments of the present invention provide a fault point prediction method and system based on multiple failure analyses, so as to solve the problems of incomplete failure analysis and low accuracy in the existing failure analysis.

On one hand, the embodiment of the invention provides a fault point prediction method based on multiple failure analysis, which comprises the following steps:

analyzing historical maintenance records to obtain abnormal points in fault points, deleting detection steps related to the abnormal points based on parts and lines thereof related to all detection steps of each detection flow, and supplementing detection steps related to other parts belonging to the same line to obtain actual detection steps of each detection flow;

performing failure analysis according to the diagnostic data of the assembly to be detected, if the failure analysis is primary failure analysis, acquiring the actual detection step of the corresponding detection flow, and if not, updating the actual detection step of the corresponding detection flow according to the fault point of the last failure analysis; obtaining a fault point according to the obtained actual detection value, if the fault point is correct, ending the failure analysis, otherwise, iteratively carrying out the next failure analysis until the maximum failure analysis times is reached, and obtaining a final fault point;

and if the final fault point is incorrect, constructing data to be predicted according to multiple failure analysis, inputting the data to the corresponding trained classification model, and predicting the fault point.

Based on further improvement of the method, the historical maintenance records are analyzed to obtain abnormal points in the fault points, and the method comprises the following steps:

counting the yield of fault points in the maintenance record according to periods based on the historical maintenance record;

removing fault points with yield rate not conforming to sigma principle to obtain fault points to be clustered;

clustering fault points to be clustered by adopting a density clustering algorithm according to the yield of the fault points to be clustered in the same period and a preset neighborhood radius to obtain various types of clustering; taking the fault points in the category of which the number of the fault points is less than the number threshold value as outliers;

and taking the outlier with the yield smaller than the minimum yield threshold as an abnormal point.

Based on the further improvement of the method, based on the parts and the lines thereof associated with all the detection steps of each detection flow, the detection steps related to abnormal points are deleted, and the detection steps associated with other parts belonging to the same line are supplemented to obtain the actual detection steps of each detection flow, and the method comprises the following steps of:

based on the failure analysis knowledge map, obtaining parts associated with each detection step and the circuits thereof under each detection process according to the part entities associated with each detection step entity and the circuit entities associated with the part entities; the abnormal point name is called as a part entity name, and a part entity corresponding to the abnormal point and a related line entity thereof are inquired to obtain a line to which the abnormal point belongs;

in all detection steps of each detection process, deleting a detection step corresponding to the abnormal point and a detection step associated with parts of the same line with the abnormal point;

and for each detection flow, respectively supplementing the detection steps related to other parts belonging to the same line according to the parts related to the rest detection steps and the line to which the parts belong, and updating the relationship among the detection steps to obtain the actual detection steps of each detection flow.

Based on the further improvement of the method, the diagnostic data of the assembly to be detected comprises the detection project names and the detection numerical values thereof corresponding to the multiple fault phenomena respectively, and the corresponding detection process is obtained according to the first detection project name.

Based on the further improvement of the method, the actual detection step of the corresponding detection process is updated according to the fault point of the last failure analysis, and the actual detection step comprises the following steps:

and acquiring detection steps related to all parts on the adjacent line of the line to which the fault point belongs according to the fault point of the last failure analysis, sequentially putting the sets to be supplemented into the detection steps, adding the sets to be supplemented into the actual detection steps of the corresponding detection process, and updating the relation among the detection steps.

Based on a further improvement of the above method, updating the relationship between the detection steps comprises:

acquiring a detection step corresponding to the fault point as a first operation step, and acquiring a corresponding detection step according to the next operation relation of the first operation step as a second operation step; and associating the next operation relation of the first operation step to the first detection step in the set to be supplemented, and associating the next operation relation of the last detection step in the set to be supplemented to the second operation step.

Based on further improvement of the method, the obtaining of the fault point according to the obtained actual detection value comprises the following steps:

taking the detection step with a preset first-step operation step code as a first-step operation based on the actual detection step of the corresponding detection flow;

and taking the obtained actual detection value of the first-step operation as a judgment result, identifying whether a fault point corresponding to the judgment result is empty or not according to the detection result associated with the first-step operation, if not, obtaining the fault point, otherwise, obtaining a detection step associated with the next-step operation relation of the first-step operation according to the actual detection value, and continuing to obtain the actual detection value of the detection step until the fault point of the detection step corresponding to the actual detection value is not empty.

Based on the further improvement of the method, the classification model is obtained by constructing and training a sample set of each detection flow according to the sample detection flow of each failure analysis of each detection flow based on historical diagnosis data, historical failure analysis records and historical maintenance records generated by multiple failure analyses; the sample detection process of each failure analysis of each detection process comprises the following steps:

if the failure analysis is the first failure analysis, according to the part associated with each detection step under the corresponding detection flow and the circuit to which the part belongs, the detection steps associated with other parts belonging to the same circuit are supplemented into the original detection step, and the relation among the detection steps is updated to obtain a sample detection flow; otherwise, according to the detection result associated with each detection step in the corresponding detection flow, acquiring the fault point corresponding to the judgment result, sequentially supplementing the detection steps associated with all parts on the adjacent line of the line to which each fault point belongs into the original detection steps, and updating the relationship among the detection steps to obtain the sample detection flow.

Based on the further improvement of the method, a sample set of each detection flow is constructed according to the sample detection flow of each failure analysis of each detection flow, and the method comprises the following steps:

acquiring a plurality of detection item names and detection values thereof in sequence based on each piece of historical diagnosis data generated by each failure analysis; converting the name of the detection project into a detection code, forming a pair of diagnosis information by the detection code and a detection value thereof, and splicing a plurality of pairs of diagnosis information in sequence to obtain one piece of diagnosis data of each failure analysis;

acquiring a failure analysis record corresponding to each piece of diagnosis data of each failure analysis based on the historical failure analysis record, corresponding an actual detection value in the failure analysis record to a sample detection step, acquiring a fault point of each failure analysis, and acquiring a fault point successfully repaired based on the corresponding historical maintenance record;

splicing each piece of diagnosis data, actual detection values corresponding to the sample detection steps and fault points of the same assembly in the same failure analysis as one-time detection information of the assembly, splicing the detection information for multiple times as a sample according to the preset maximum failure analysis times, and placing the corresponding fault points which are successfully maintained as classification results into a sample set of the corresponding detection process.

On the other hand, an embodiment of the present invention provides a failure point prediction system based on multiple failure analysis, including:

the detection step acquisition module is used for analyzing the historical maintenance records to obtain abnormal points in fault points, deleting the detection steps related to the abnormal points based on the parts and the lines thereof related to all the detection steps of each detection flow, and supplementing the detection steps related to other parts belonging to the same line to obtain the actual detection steps of each detection flow;

the failure analysis module is used for performing failure analysis according to the diagnostic data of the assembly part to be detected, acquiring the actual detection step of the corresponding detection flow if the initial failure analysis is performed, and updating the actual detection step of the corresponding detection flow according to the fault point of the last failure analysis if the initial failure analysis is performed; obtaining a fault point according to the obtained actual detection value, if the fault point is correct, ending the failure analysis, otherwise, iterating to carry out the next failure analysis until the maximum failure analysis times is reached, and obtaining a final fault point;

and the fault point prediction module is used for constructing data to be predicted according to multiple failure analyses when the final fault point obtained by the multiple failure analyses is incorrect, inputting the data to be predicted to the corresponding trained classification model and predicting the fault point.

Compared with the prior art, the invention can realize at least one of the following beneficial effects: analyzing abnormal points from historical maintenance records, deleting detection steps related to the abnormal points, and accelerating the positioning of fault points; with the increase of failure analysis times, a new detection step is supplemented according to the failure point, so that the situation that the detected part is incomplete or the accurate failure point cannot be positioned due to the damage of peripheral parts caused by the maintenance of the previous failure part is avoided, the detection step is more complete, and the comprehensiveness of failure analysis is realized; meanwhile, a classification model is trained by using historical data generated by multiple failure analysis, and the classification model is provided for a maintenance engineer to make a final maintenance attempt after multiple failure analysis, so that the value of the historical data is maximized, and the failure analysis is more automatic and more convenient.

In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.

Drawings

The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.

Fig. 1 is a flowchart of a failure point prediction method based on multiple failure analysis in embodiment 1 of the present invention.

Detailed Description

The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.

Example 1

A specific embodiment of the present invention discloses a failure point prediction method based on multiple failure analysis, as shown in fig. 1, including the following steps:

s11: and analyzing the historical maintenance records to obtain abnormal points in the fault points, and deleting the detection steps related to the abnormal points based on all the detection steps of each detection flow to obtain the actual detection steps of each detection flow.

The process of failure analysis of the defective assembly according to the failure analysis standard flow by the service engineer based on the diagnostic data is recorded in the historical failure analysis record, in which the failure analysis standard flow code, the defective assembly code, the detection value of each step, and the final failure point (i.e., the component) are described in detail. The historical maintenance record is a record of maintenance engineers performing maintenance according to the fault point in the historical failure analysis record, and records whether the fault point is repaired, the maintenance time and the like.

Specifically, analyzing the historical service records to obtain an abnormal point in the fault point includes:

counting the yield of fault points in the maintenance data according to periods based on historical maintenance records; removing fault points with yield rate not conforming to sigma principle to obtain fault points to be clustered; clustering fault points to be clustered by adopting a density clustering algorithm according to the yield of the fault points to be clustered in the same period and a preset neighborhood radius to obtain various types of clusters; taking the fault points in the category of which the number of the fault points is less than the number threshold value as outliers; and taking the outlier with the yield smaller than the minimum yield threshold as an abnormal point.

It should be noted that, according to whether the fault point in the historical maintenance record is repaired, the yield of the fault point, that is, the probability of the repaired fault point, is counted periodically. The period can be month or week, and the statistical range is determined according to the actual maintenance condition, such as the number of historical maintenance records and the frequency of maintenance. Illustratively, the yield of failure points for 6 consecutive months is counted.

After yields for all fault points are obtained, illustratively, the yield for the fault point needs to be removed from the fault point when the yield is less than the average-3 sigma using the normal distribution based three-sigma principle. And clustering the remaining fault points by using a DBSCAN algorithm according to the yield of the remaining fault points in the last month. Based on the clustering result, if the number of fault points in the category is less than the number threshold, the fault points in the category are outliers.

And comparing the yield of the outlier with the minimum yield threshold again, and if the yield is smaller than the minimum yield threshold, taking the outlier as an abnormal point.

It should be noted that all the detection steps of each detection process are already present in the failure analysis knowledge-graph. When the failure analysis knowledge graph is constructed, all detection steps and detection results under each detection process are extracted from manually written historical failure analysis files. The detection objects related to each detection step are part entities established in the knowledge graph and are in association relation with the corresponding detection steps, and the part entities are associated with circuit entities to obtain the circuits of the parts. When the detection step entities are instantiated, the next operation relationship among the detection step entities is initialized, and the next operation relationship includes the next operation when normal detection is carried out and the next operation when abnormal detection is carried out. The detection result of each detection step shows a fault point corresponding to normal and/or abnormal detection. For the operation step code of the first detection step of each detection flow, the operation step code is set as a preset operation step code, for example, the operation step codes are uniformly set as a001, and the operation step codes of other detection steps can be set according to a preset rule.

Illustratively, table 1 is a failure analysis criteria flow generated from a failure analysis knowledge graph. The step numbers and the operation types in table 1 are directly derived from the attribute values of the detection step entities, the operation step descriptions are filled into the operation step description templates corresponding to visual inspection and measurement according to the associated part entities, the step numbers of the detection step entities corresponding to the next operation for detecting normal (0) and/or abnormal (1) are the next step, and the fault points are the fault parts corresponding to the detection result (0) when the detection is normal and/or the detection result (1) when the detection is abnormal according to the associated detection results.

Table 1 example of standard flow for failure analysis

In this embodiment, based on the line to which all the detection steps of each detection flow belong, deleting the detection step related to the outlier, and adding the detection steps belonging to the same line as the remaining detection steps, to obtain the actual detection steps of each detection flow, the method includes:

based on the failure analysis knowledge graph, obtaining a part corresponding to each detection step and a belonging line under each detection process according to the part entity associated with each detection step entity and the line entity associated with the part entity; the abnormal point name is called as a part entity name, and a part entity corresponding to the abnormal point and a related line entity thereof are inquired to obtain a line to which the abnormal point belongs;

in all detection steps of each detection process, deleting a detection step corresponding to the abnormal point and a detection step associated with parts of the same line with the abnormal point;

and for each detection flow, supplementing detection steps related to other parts belonging to the same line according to the parts related to the remaining detection steps and the belonging line, and updating the relation among the detection steps to obtain an actual detection step.

It should be noted that, updating the relationship between the detection steps is to establish a next operation relationship, which is normal or abnormal, corresponding to the next detection step according to the detection result associated with the entity in the previous detection step and according to the abnormal or normal determination result in the detection result, and includes:

when the judgment result corresponding to the previous detection step is normal and the fault point is not empty, establishing a next step operation relation for detecting the abnormality between the previous detection step and the next detection step; when the judgment result corresponding to the previous detection step is abnormal and the fault point is not empty, establishing a next operation relation for normal detection between the previous detection step and the next detection step; if the fault point corresponding to the judgment result corresponding to the previous detection step is empty, the next step operation relation of normal detection and abnormal detection is established simultaneously by the entity of the previous detection step and the entity of the next detection step; the entity in the previous detection step has both normal and abnormal judgment results, the fault points are not empty, the emptying judgment result is the fault point corresponding to the normal condition, and the next operation relation for detecting the normal condition is established between the previous detection step and the next detection step.

Illustratively, if part R3300 is an abnormal point, part R3301 and R3300 belong to the same line, a002, a003 and a007 are deleted in table 1, and the remaining detection steps associated with parts C9400, C9401, C9402 and C9403 belong to another line on which part C9404 is present from the failure analysis knowledge map, the detection step associated with part C9404 needs to be added to table 1 and placed after step a 006. When the relation among the detection steps is updated, because the judgment result corresponding to A001 is abnormal and the fault point is C9400, establishing the next operation relation for detecting the normality with the subsequent A004; the emptying step a006 determines that the result is a normal corresponding fault point C9404, and establishes a normal detection next operation relationship with the supplementing step, so as to obtain an actual detection step of the first analysis.

It should be noted that the yield of the abnormal point is very low, that is, when the abnormal point is located as a fault point, the assembly is difficult to maintain, so that the steps related to the abnormal point and the detection steps belonging to the same line with the abnormal point are deleted in all the detection steps, the detection steps with interference are reduced, the detection process is simplified, and a maintenance engineer can locate the accurate fault point according to the remaining detection steps.

S12: performing failure analysis according to the diagnostic data of the assembly to be detected, if the failure analysis is primary failure analysis, acquiring the actual detection step of the corresponding detection flow, and if not, updating the actual detection step of the corresponding detection flow according to the fault point of the last failure analysis; and obtaining a fault point according to the obtained actual detection value, if the fault point is correct, ending the failure analysis, otherwise, iteratively performing the next failure analysis until the maximum failure analysis times is reached, and obtaining a final fault point.

It should be noted that the diagnostic data of the assembly component to be detected includes the detection project names and the detection values thereof corresponding to the multiple fault phenomena, respectively, and the corresponding detection process is obtained according to the first detection project name.

In the embodiment, when the first failure analysis is performed, an actual detection step excluding a detection step related to an abnormal point is used, if a failure point obtained by the first failure analysis is verified by a maintenance engineer that an assembly part cannot be modified, the detection steps corresponding to parts around the failure point can be supplemented by using the failure point obtained by the last failure analysis when the second and subsequent failure analyses are performed for multiple times, and the problem that the accurate failure point cannot be located due to the fact that the detected parts are incomplete in an actual detection process or the fact that the parts around the failure point are damaged due to the fact that the parts around the failure point are maintained for the last time is avoided.

It should be noted that, according to the failure point of the last failure analysis, the actual detection step of updating the corresponding detection flow includes:

and acquiring detection steps related to all parts on the adjacent line of the line to which the fault point belongs according to the fault point of the last failure analysis, sequentially putting the sets to be supplemented into the detection steps, adding the sets to be supplemented into the actual detection steps of the corresponding detection process, and updating the relation among the detection steps.

It should be noted that the detection step to be supplemented is related to the failure point of the previous failure analysis, so in this embodiment, the relationship between the detection step corresponding to the failure point of the previous failure analysis in the actual detection process and the detection step associated with the next operation is disconnected, and a relationship is established with the detection step in the set to be supplemented, so that when the next failure analysis is performed, after the detection step corresponding to the failure point of the previous failure analysis is performed, the detection step added in the set to be supplemented is performed, and the operation returns to the original next operation.

For example, in the analysis of the failure of the automobile parts, the failure point obtained by the first failure analysis is a brake, and a maintenance engineer performs maintenance on the brake, but the maintenance is not successful. And the brake drum, the friction plate and the contact between the brake drum and the friction plate are obtained by correlation query according to the knowledge map, so that the detection steps of the contact between the brake drum, the friction plate and the brake drum and the friction plate are added to the step of the first failure analysis, the detection steps are more perfect, and the comprehensiveness of the failure analysis is realized.

When failure analysis is carried out, a fault point is obtained according to the obtained actual detection value, and the method comprises the following steps:

taking a detection step with a preset first-step operation step code as a first-step operation based on the failure analysis knowledge graph;

and taking the obtained actual detection value of the first-step operation as a judgment result, identifying whether a fault point corresponding to the judgment result is empty or not according to the detection result associated with the first-step operation, if not, obtaining the fault point, otherwise, obtaining a detection step associated with the next-step operation relation of the first-step operation according to the actual detection value, and continuing to obtain the actual detection value of the detection step until the fault point of the detection step corresponding to the actual detection value is not empty.

Illustratively, in table 1, the description of the operation steps of the first step is displayed according to a preset first-step operation step code, such as a001, a maintenance engineer visually inspects the part C9400, if there is a problem, the actual detection value 1 is input according to the description of the operation steps, at this time, 1 corresponds to a fault point C9400 in table 1 as a determination result that the first step is abnormal, a failure analysis is completed, if there is no problem, the actual detection value 0 is input according to the description of the operation steps, at this time, there is no fault point, the detection step corresponding to the normal next-step operation is a002, the description of the operation steps of a002 is displayed, and the maintenance engineer inputs the actual detection value according to the visual inspection condition, and according to the fault point corresponding to a002 or the next-step operation relation until the fault point is obtained, and completes the failure analysis.

It should be noted that the maximum failure analysis number is determined according to the field of actual failure analysis and the quality of the assembly. For example, if the assembly is a motherboard, the maximum number of failures is set to 3.

S13: and if the final fault point is incorrect, constructing data to be predicted according to multiple failure analysis, inputting the data to the corresponding trained classification model, and predicting the fault point.

It should be noted that the classification model is obtained by constructing and training a sample set of each detection flow according to a sample detection flow of each failure analysis of each detection flow based on historical diagnostic data, historical failure analysis records and historical maintenance records generated by multiple failure analyses, and the classification model includes: random forest, xgboost, and graph neural network GNN.

It should be noted that, in the actual failure analysis, the first failure analysis is a detection step related to other parts belonging to the same line after the abnormal point-removing related detection step in all detection steps is supplemented, and the subsequent failure analysis is a step supplemented according to dynamic fault points, so that all steps of each failure analysis need to be considered when the classification model is trained. The sample detection process of each failure analysis of each detection process comprises the following steps:

if the failure analysis is the first failure analysis, according to the part associated with each detection step under the corresponding detection flow and the circuit to which the part belongs, the detection steps associated with other parts belonging to the same circuit are supplemented into the original detection step, and the relation among the detection steps is updated to obtain a sample detection flow; otherwise, according to the detection result associated with each detection step in the corresponding detection flow, acquiring the fault point corresponding to the judgment result, sequentially supplementing the detection steps associated with all parts on the adjacent line of the line to which each fault point belongs into the original detection steps, and updating the relationship among the detection steps to obtain the sample detection flow.

It should be noted that if at least two failure analyses are performed in the actual situation, model training for the first failure analysis is not required.

When a sample set is constructed, historical data subjected to multiple failure analyses is selected based on historical diagnostic data, historical failure analysis records and historical maintenance records, and a fault point with successful maintenance is obtained. The failure point of successful repair is not limited by the maximum failure analysis number. When the failure analysis times of some assembly parts in the historical data are smaller than the maximum failure analysis times, sample data of the previous time can be supplemented according to the sample data format of the current time so as to increase the number of samples and improve the classification accuracy of the model.

Specifically, constructing a sample set of each detection flow according to the sample detection flow of each failure analysis of each detection flow, including:

(1) acquiring a plurality of detection item names and detection values thereof in sequence based on each piece of historical diagnosis data generated by each failure analysis; converting the name of the detection project into a detection code, forming a pair of diagnosis information by the detection code and the detection value thereof, and splicing a plurality of pairs of diagnosis information in sequence to obtain one piece of diagnosis data of each failure analysis.

Illustratively, one piece of diagnostic data is a-1c-5e10g100w5, wherein "acegw" is a detection code having an order and "-1-5101005" is a detection value corresponding to "acegw", respectively.

(2) And acquiring a failure analysis record corresponding to each piece of diagnosis data of each failure analysis based on the historical failure analysis record, corresponding an actual detection value in the failure analysis record to a sample detection step, acquiring a fault point of each failure analysis, and acquiring a fault point successfully repaired based on the corresponding historical repair record.

It should be noted that the actual detection values in the historical failure analysis records are usually 0 and 1, different detection steps are executed according to the actual detection values, and the actual detection values corresponding to the detection steps that are not executed are empty. And according to the operation step codes, the actual detection value is corresponding to the sample detection step. And deleting the samples with the null ratio larger than the null threshold value, and filling the null value with mode for the samples with the null ratio smaller than the null threshold value. Wherein the mode is obtained by statistics according to the detection records of respective detection processes.

Illustratively, the operational step description of step 1 is: visual inspection is performed on the C9400, and a problem input of '1' and a problem-free input of '0' are performed; if 1 is input, step 2 is executed, and if 0 is input, step 3 is executed. And when the actual detection value input by the maintenance engineer is 0, directly executing the step 3, wherein the actual detection value corresponding to the step 2 is empty, and if the actual detection value which is 1 in the detection process is more, supplementing the step 2 with the actual detection value 1.

(3) Splicing each piece of diagnosis data, actual detection values corresponding to the sample detection steps and fault points of the same assembly in the same failure analysis as one-time detection information of the assembly, splicing the detection information for multiple times as a sample according to the preset maximum failure analysis times, and placing the corresponding fault points which are successfully maintained as classification results into a sample set of the corresponding detection process.

And dividing the constructed sample set into a training set and a testing set, wherein the training set is used for training the classification model, and the testing set is used for performing performance test and accuracy calculation on the analysis model so as to prevent overfitting and under-training of the network on the training data set. These are of conventional use and will not be described in detail.

If the final fault point obtained in the step S12 is incorrect, data to be predicted is constructed according to multiple failure analysis and is input to a corresponding trained classification model, and the fault point is predicted.

It should be noted that the data to be predicted is constructed according to the format of the sample set according to the sample detection step corresponding to each failure analysis of the detection process, and includes:

acquiring diagnostic data, actual detection values and fault points of the assembly to be detected, which are obtained by failure analysis each time;

according to the detection step operation codes, corresponding actual detection values input by each failure analysis to the sample detection step of the current failure analysis of the current detection process, and filling the sample detection steps which cannot be corresponding according to the mode of the sample set of the current detection process to obtain detection data;

and sequentially splicing the diagnostic data, the detection data and the fault points of the assembly part to be detected, which are obtained by each failure analysis, to obtain data to be predicted.

Compared with the prior art, the fault point prediction method based on multiple failure analysis of the embodiment separates abnormal points from the historical maintenance records, deletes the detection steps related to the abnormal points, and quickens the positioning of the fault points; with the increase of failure analysis times, a new detection step is supplemented according to the failure point, so that the situation that the detected part is incomplete or the accurate failure point cannot be positioned due to the damage of peripheral parts caused by the maintenance of the previous failure part is avoided, the detection step is more complete, and the comprehensiveness of failure analysis is realized; meanwhile, a classification model is trained by using historical data generated by multiple failure analysis, and the classification model is provided for a maintenance engineer to make a final maintenance attempt after multiple failure analysis, so that the value of the historical data is maximized, and the failure analysis is more automatic and more convenient.

Example 2

The other embodiment of the invention discloses a fault point prediction system based on multiple failure analysis, so that the fault point prediction method based on multiple failure analysis in the embodiment 1 is realized. The concrete implementation of each module refers to the corresponding description in embodiment 1. The system comprises:

the detection step acquisition module is used for analyzing the historical maintenance records to obtain abnormal points in fault points, deleting the detection steps related to the abnormal points based on the parts and the lines thereof related to all the detection steps of each detection flow, and supplementing the detection steps related to other parts belonging to the same line to obtain the actual detection steps of each detection flow;

the failure analysis module is used for performing failure analysis according to the diagnostic data of the assembly part to be detected, acquiring the actual detection step of the corresponding detection flow if the initial failure analysis is performed, and updating the actual detection step of the corresponding detection flow according to the fault point of the last failure analysis if the initial failure analysis is performed; obtaining a fault point according to the obtained actual detection value, if the fault point is correct, ending the failure analysis, otherwise, iteratively carrying out the next failure analysis until the maximum failure analysis times is reached, and obtaining a final fault point;

and the fault point prediction module is used for constructing data to be predicted according to multiple failure analyses when the final fault point obtained by the multiple failure analyses is incorrect, inputting the data to be predicted to the corresponding trained classification model, and predicting the fault point.

Since the parts of the failure point prediction system based on multiple failure analysis and the generation method in the present embodiment can be referred to each other, which is a repeated description, the description is omitted here. Since the principle of the embodiment of the system is the same as that of the embodiment of the method, the embodiment of the system also has the corresponding technical effect of the embodiment of the method.

Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.

The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.

Claims (10)

1. A fault point prediction method based on multiple failure analysis is characterized by comprising the following steps:

analyzing historical maintenance records to obtain abnormal points in fault points, deleting detection steps related to the abnormal points based on parts and lines thereof related to all detection steps of each detection flow, and supplementing detection steps related to other parts belonging to the same line to obtain actual detection steps of each detection flow;

performing failure analysis according to the diagnostic data of the assembly to be detected, if the failure analysis is primary failure analysis, acquiring the actual detection step of the corresponding detection flow, and if not, updating the actual detection step of the corresponding detection flow according to the fault point of the last failure analysis; obtaining a fault point according to the obtained actual detection value, if the fault point is correct, ending the failure analysis, otherwise, iterating to carry out the next failure analysis until the maximum failure analysis times is reached, and obtaining a final fault point;

and if the final fault point is incorrect, constructing data to be predicted according to multiple failure analysis, inputting the data to the corresponding trained classification model, and predicting the fault point.

2. The multiple failure analysis-based failure point prediction method of claim 1, wherein analyzing historical service records to obtain abnormal points in failure points comprises:

counting the yield of fault points in the maintenance data according to periods based on historical maintenance records;

removing fault points with yield rate not conforming to the sigma principle to obtain fault points to be clustered;

clustering fault points to be clustered by adopting a density clustering algorithm according to the yield of the fault points to be clustered in the same period and a preset neighborhood radius to obtain various types of clustering; taking the fault points in the category with the quantity of the fault points smaller than the quantity threshold value as outliers;

and taking the outlier with the yield smaller than the minimum yield threshold as an abnormal point.

3. The failure point prediction method based on multiple failure analysis according to claim 2, wherein the actual detection step based on the detection steps associated with all the detection steps of each detection flow and the circuit to which the detection step is associated is obtained by deleting the detection step associated with the abnormal point and supplementing the detection steps associated with other parts belonging to the same circuit, and comprises:

based on the failure analysis knowledge map, obtaining parts associated with each detection step and the circuits thereof under each detection process according to the part entities associated with each detection step entity and the circuit entities associated with the part entities; the abnormal point name is called as a part entity name, and a part entity corresponding to the abnormal point and a related line entity thereof are inquired to obtain a line to which the abnormal point belongs;

deleting the detection step corresponding to the abnormal point and the detection step associated with the parts of which the abnormal points belong to the same line in all the detection steps of each detection flow;

and for each detection flow, supplementing detection steps related to other parts belonging to the same line according to the parts related to the rest detection steps and the lines to which the parts belong, and updating the relationship among the detection steps to obtain the actual detection steps of each detection flow.

4. The failure point prediction method based on multiple failure analysis according to claim 1, wherein the diagnostic data of the assembly to be detected comprises detection project names and detection numerical values thereof corresponding to the multiple failure phenomena, respectively, and a corresponding detection process is obtained according to a first detection project name.

5. The failure point prediction method based on multiple failure analysis according to claim 1, wherein the step of updating the actual detection of the corresponding detection process according to the failure point of the previous failure analysis comprises:

and acquiring detection steps related to all parts on the adjacent line of the line to which the fault point belongs according to the fault point of the last failure analysis, sequentially putting the sets to be supplemented into the detection steps, adding the sets to be supplemented into the actual detection steps of the corresponding detection process, and updating the relation among the detection steps.

6. The multiple failure analysis-based failure point prediction method of claim 5, wherein updating the relationship between the detection steps comprises:

acquiring a detection step corresponding to a fault point, wherein the detection step is used as a first operation step, and acquiring a corresponding detection step according to the next operation relation of the first operation step, and the detection step is used as a second operation step; and associating the next operation relation of the first operation step to the first detection step in the set to be supplemented, and associating the next operation relation of the last detection step in the set to be supplemented to the second operation step.

7. The multiple failure analysis-based failure point prediction method according to claim 1, wherein the obtaining a failure point according to the obtained actual detection value comprises:

taking the detection step with a preset first-step operation step code as a first-step operation based on the actual detection step of the corresponding detection flow;

and taking the obtained actual detection value of the first-step operation as a judgment result, identifying whether a fault point corresponding to the judgment result is empty or not according to the detection result associated with the first-step operation, if not, obtaining the fault point, otherwise, obtaining a detection step associated with the next-step operation relation of the first-step operation according to the actual detection value, and continuing to obtain the actual detection value of the detection step until the fault point of the detection step corresponding to the actual detection value is not empty.

8. The failure point prediction method based on multiple failure analysis according to claim 1, wherein the classification model is obtained by constructing and training a sample set of each detection process according to a sample detection process of each failure analysis of each detection process based on historical diagnosis data, historical failure analysis records and historical maintenance records generated by multiple failure analysis; the sample detection process for each failure analysis of each detection process comprises the following steps:

if the failure analysis is the first failure analysis, according to the part associated with each detection step under the corresponding detection flow and the circuit to which the part belongs, the detection steps associated with other parts belonging to the same circuit are supplemented into the original detection step, and the relation among the detection steps is updated to obtain a sample detection flow; otherwise, acquiring the fault point corresponding to the judgment result according to the detection result associated with each detection step in the corresponding detection process, sequentially supplementing the detection steps associated with all parts on the adjacent line of the line to which each fault point belongs into the original detection step, and updating the relation among the detection steps to obtain the sample detection process.

9. The method for predicting fault points based on multiple failure analysis according to claim 8, wherein the constructing a sample set of each detection process according to the sample detection process of each failure analysis of each detection process comprises:

acquiring a plurality of detection item names and detection values thereof in sequence based on each piece of historical diagnosis data generated by each failure analysis; converting the name of the detection project into a detection code, forming a pair of diagnosis information by the detection code and a detection value thereof, and splicing a plurality of pairs of diagnosis information in sequence to obtain one piece of diagnosis data for each failure analysis;

acquiring a failure analysis record corresponding to each piece of diagnosis data of each failure analysis based on the historical failure analysis record, corresponding an actual detection value in the failure analysis record to a sample detection step, acquiring a fault point of each failure analysis, and acquiring a fault point successfully repaired based on the corresponding historical repair record;

splicing each piece of diagnosis data, actual detection values corresponding to the sample detection steps and fault points of the same assembly in the same failure analysis as one-time detection information of the assembly, splicing the detection information for multiple times as a sample according to the preset maximum failure analysis times, and placing the corresponding fault points which are successfully maintained as classification results into a sample set of the corresponding detection process.

10. A failure point prediction system based on multiple failure analysis, comprising:

the detection step acquisition module is used for analyzing the historical maintenance records to obtain abnormal points in fault points, deleting the detection steps related to the abnormal points based on the parts and the lines thereof related to all the detection steps of each detection flow, and supplementing the detection steps related to other parts belonging to the same line to obtain the actual detection steps of each detection flow;

the failure analysis module is used for performing failure analysis according to the diagnostic data of the assembly part to be detected, acquiring the actual detection step of the corresponding detection flow if the initial failure analysis is performed, and updating the actual detection step of the corresponding detection flow according to the fault point of the last failure analysis if the initial failure analysis is performed; obtaining a fault point according to the obtained actual detection value, if the fault point is correct, ending the failure analysis, otherwise, iterating to carry out the next failure analysis until the maximum failure analysis times is reached, and obtaining a final fault point;

and the fault point prediction module is used for constructing data to be predicted according to multiple failure analyses when the final fault point obtained by the multiple failure analyses is incorrect, inputting the data to be predicted to the corresponding trained classification model, and predicting the fault point.

Images