Vehicle body clearance surface difference quality prediction method and system, storage medium and vehicle

Abstract

The invention provides a vehicle body clearance surface difference quality prediction method and system, a storage medium and a vehicle, and the method comprises the steps: obtaining the original sample data of a vehicle body clearance surface, the original sample data comprising essential factor data and result data; removing noise data in the result system data after clustering analysis to obtain result system training data; according to the result system data, selecting essential factor system training data influencing the quality of the gap surface difference from the essential factor system data; constructing a poor quality prediction model based on a random forest method according to the result system training data and the cause system training data; and predicting the vehicle body clearance surface difference quality according to the difference quality prediction model. The method comprises the following steps: screening original sample data to remove noise data in result system data and extract cause system training data influencing the gap surface difference quality, and then constructing a difference quality prediction model with very high accuracy by using the screened original sample data; and utilizing the difference quality prediction model to realize accurate prediction of the vehicle body clearance surface difference quality.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a vehicle body gap and flushness quality prediction method, system, storage medium and vehicle. Background Art

[0002] In the quality inspection of the automobile manufacturing industry, the quality of the gap and surface difference of the vehicle body is a key indicator to measure the appearance quality of the vehicle. Due to the complexity and multi-linkage of the automobile production process, the body still needs to go through multiple processes such as spraying and assembly after welding, which makes it difficult to maintain the initial matching state after welding (i.e. the gap and surface difference between the mating surfaces of the body) in the subsequent process links. At the same time, the unpredictable variation between processes often leads to the vehicle passing the early inspection of the production line, but a large number of precision problems in the subsequent processes.

[0003] In order to ensure the quality of the gap and flushness of the car body, the common method adopted by various automobile manufacturers is to increase the inspection between processes. The traditional quality inspection methods mainly include: manual measurement and robot measurement. For the above two inspection methods, the manual measurement method has the problems of low efficiency, complex operation, high manpower and time costs, and the robot measurement method has the problem of delayed feedback of the measurement and analysis results on the gap and flushness problem, which cannot detect the problem as early as possible, guide the improvement of the production end, and avoid the occurrence of large-scale gap and flushness size problems. This restricts the further improvement and optimization of the quality system of the automobile manufacturing industry. Therefore, in the field of body gap and flushness prediction, although many well-known automobile companies and suppliers are actively exploring the use of production data mining and digital technology to optimize the current quality inspection process, there is still room for improvement in prediction accuracy and universal applicability.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present invention is to provide a vehicle body gap and flushness quality prediction method, a storage medium and a vehicle, aiming to solve the technical problem of low accuracy in vehicle body gap and flushness quality prediction in the prior art.

[0006] To achieve the above object, the present invention proposes a vehicle body gap flushness quality prediction method, the vehicle body gap flushness quality prediction method comprising:

[0007] Acquire original sample data of the vehicle body gap surface, wherein the original sample data includes: cause system data and result system data;

[0008] Performing cluster analysis on the result system data, and removing noise data in the result system data after the cluster analysis to obtain result system training data;

[0009] Selecting factor system training data that affects the gap and flushness quality from the factor system data according to the result system data;

[0010] Constructing a poor quality prediction model based on the random forest method according to the result system training data and the factor system training data;

[0011] The vehicle body gap surface difference quality is predicted according to the difference quality prediction model.

[0012] Optionally, performing cluster analysis on the result system data and removing noise data in the result system data after the cluster analysis to obtain result system training data includes:

[0013] Integrate abnormal problem data in the result system data;

[0014] Constructing a cluster analysis model for matching abnormal problems according to the last occurrence time, repair consumption time and cumulative occurrence frequency in the abnormal problem data;

[0015] The abnormal problem data is clustered through the cluster analysis model to obtain the clustering results of the result system data;

[0016] Determining noise data in the result series data from the clustering results according to the Pareto principle;

[0017] The result is training data obtained by removing noise data from the result data.

[0018] Optionally, selecting factor system training data affecting the gap and flushness quality from the factor system data according to the result system data includes:

[0019] Determine the abnormal probability of gap and flushness quality abnormality based on the result system data;

[0020] Analyze the result system data to determine the maximum likelihood estimate of the gap and flushness quality problem and the posterior probability of the point data corresponding to the gap and flushness quality problem;

[0021] Determine the prior probability of the point data by using a Bayesian probability formula according to the abnormal probability, the maximum likelihood estimate and the posterior probability;

[0022] The factor system training data affecting the gap and flushness quality are selected from the factor system data according to the prior probability of the point data.

[0023] Optionally, constructing a poor quality prediction model based on a random forest method according to the result training data and the factor training data includes:

[0024] The result system training data and the factor system training data are split according to a preset splitting standard to obtain sample training set data and sample verification set data;

[0025] Constructing an initial poor quality prediction model based on the random forest method according to the sample training set data;

[0026] Using the sample validation set data to verify the stability and accuracy of the initial poor quality prediction model;

[0027] When the stability and accuracy verification is passed, a poor quality prediction model is obtained.

[0028] Optionally, after constructing the poor quality prediction model based on the result system training data and the factor system training data based on the random forest method, the method further includes:

[0029] Obtaining information gain of each decision node in the process of building the poor quality prediction model;

[0030] Determine the data weight of each point data in the factor system data according to the information gain of each decision node;

[0031] The data of each point is sorted and output according to the data weight.

[0032] Optionally, after obtaining the original sample data of the vehicle body clearance surface, the method further includes:

[0033] Identify abnormal data in the original sample data by using an isolation forest algorithm or a clustering algorithm;

[0034] Removing abnormal data from the original sample data;

[0035] The original sample data after the abnormal data is removed is identified as the new original sample data.

[0036] Optionally, before constructing the poor quality prediction model based on the random forest method according to the result system training data and the factor system training data, the method further includes:

[0037] Convert all feature values ​​in the result system training data and the factor system training data to the same dimension;

[0038] Accordingly, the poor quality prediction model is constructed based on the random forest method according to the result system training data and the factor system training data, including:

[0039] A poor quality prediction model is constructed based on the random forest method according to the result training data and the factor training data in which all feature values ​​are in the same dimension.

[0040] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a vehicle body gap flushness quality prediction system, and the vehicle body gap flushness quality prediction system is used to execute the vehicle body gap flushness quality prediction method.

[0041] In addition, to achieve the above purpose, the present invention also provides a storage medium, on which a vehicle body gap flushness quality prediction program is stored, and when the vehicle body gap flushness quality prediction program is executed by a processor, the steps of the vehicle body gap flushness quality prediction method are implemented.

[0042] In addition, to achieve the above-mentioned purpose, the present invention also provides a vehicle, which includes: a memory, a processor, and a vehicle body gap flushness quality prediction program stored in the memory and executable on the processor, wherein the vehicle body gap flushness quality prediction program implements the steps of the vehicle body gap flushness quality prediction method as described above when executed by the processor.

[0043] The present invention provides a method, system, storage medium and vehicle for predicting the quality of gap face difference of a vehicle body. The method comprises: obtaining original sample data of the gap face of a vehicle body, wherein the original sample data comprises: factor system data and result system data; performing cluster analysis on the result system data, and removing noise data in the result system data after cluster analysis to obtain result system training data; selecting factor system training data that affects the gap face difference quality from the factor system data according to the result system data; constructing a poor quality prediction model based on the random forest method according to the result system training data and the factor system training data; and predicting the gap face difference quality of a vehicle body according to the poor quality prediction model. The original sample data is filtered to remove noise data in the result system data and the factor system training data that affects the gap face difference quality is extracted, and then a poor quality prediction model with high accuracy is constructed using the filtered original sample data, and then the poor quality prediction model is used to accurately predict the gap face difference quality of the vehicle body. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0045] Figure 1 A schematic diagram of the structure of a vehicle involved in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of a manual measurement method in the prior art;

[0047] Figure 3 It is a schematic diagram of the measurement method of a 3D measuring machine in the prior art;

[0048] Figure 4 It is a logical diagram of data collection processing and model application in the prior art;

[0049] Figure 5 A schematic diagram of the process flow of the first embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention;

[0050] Figure 6 A schematic diagram of the first process of the second embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention;

[0051] Figure 7 Schematic diagram of clustering abnormal problem data for cluster analysis model;

[0052] Figure 8 A second flow chart of the second embodiment of the vehicle body gap flushness quality prediction method proposed by the invention;

[0053] Fig. 9 A schematic diagram of the first process of the third embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention;

[0054] Fig.10 A schematic diagram of the second process of the third embodiment of the vehicle body gap flushness quality prediction method proposed by the invention;

[0055] Fig.11 A schematic diagram of the first process of the fourth embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention;

[0056] Fig.12 This is a schematic diagram of the second process flow of the fourth embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention.

[0057] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0058] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0059] 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.

[0060] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0061] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0062] Reference Figure 1 , Figure 1 The figure is a schematic diagram of the structure of a vehicle involved in an embodiment of the present invention.

[0063] like Figure 1 As shown, the vehicle may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), and the user interface 1003 may also include a standard wired interface and a wireless interface. The wired interface of the user interface 1003 may be a USB interface in the present invention. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) or a stable memory (Non-volatile Memory, NVM), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0064] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the vehicle, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0065] like Figure 1As shown, the memory 1005 identified as a computer storage medium may include an operating system, a network communication module, a user interface module, and a vehicle body gap and flushness quality prediction program.

[0066] exist Figure 1 In the vehicle shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the user device; the vehicle calls the body gap surface difference quality prediction program stored in the memory 1005 through the processor 1001, and executes the body gap surface difference quality prediction method provided in the embodiment of the present invention.

[0067] It should be understood that, in order to ensure the gap and surface difference quality of the car body, the method generally adopted by automobile manufacturers is to increase inter-process inspections. Traditional quality inspection methods mainly include: manual measurement and robot measurement.

[0068] Reference Figure 2 , although manual measurement is more timely, its accuracy and efficiency are limited by the skills of the operator, and it can only evaluate the current state of the body gap and face difference, and cannot predict the variation of subsequent processes, lacking foresight. Tools and precision: When using feeler gauges, gauges and handheld measuring instruments for manual measurement, their precision is often limited by the quality of the tools and the proficiency of the operators. For example, the accuracy of the feeler gauge is usually only about 0.05 mm, and the accuracy of the gauge and handheld measuring instrument is also affected by many factors such as environment, temperature and usage. Unable to monitor continuously: Manual measurement methods can only be performed at specific production stages, and continuous monitoring of body gap and face difference cannot be achieved. This means that if problems occur in subsequent processes, they may not be discovered in time, resulting in quality defects in batch products. The data generated by manual measurement needs to be manually recorded and processed, which is not only inefficient but also prone to errors. In addition, due to the lack of effective data analysis tools, these data are often difficult to be fully utilized to optimize the production process.

[0069] Reference Figure 3, the specification deviation of key points of the car body can be measured with high precision by 3D measuring machine. 3D measuring robots usually use laser scanning or structured light technology, and their accuracy can reach the micron level, which is much higher than the manual measurement method. Although 3D measuring machines have excellent measurement accuracy advantages, they also have limitations when used in the automobile production process. First, the measurement speed of 3D measuring robots is relatively slow. It may take 2-3 hours to perform a complete 3D scan of a car body. In actual production, it is impossible to use 3D measuring machines to conduct comprehensive measurements on every car. The general method is to sample the same batch of products. Second, the gap and flushness problem is usually related to a series of dimension chains. The amount of data generated by 3D measurement is huge and complex. Due to the gap and flushness problem and the complexity of the dimension chain, and the pre-set upper and lower limits of the specifications may not fully reflect the subsequent accuracy problems, it is necessary to do complex and time-consuming data analysis based on experience to find problems. Third, the measurement process of the 3D measuring machine is usually carried out after the welding process is completed. In the fast-paced production rhythm, the existing 3D measuring machine usage solution, the measurement and analysis results cannot be fed back to the production line in time, which may lead to the production of batches of unqualified products.

[0070] The above two inspection methods have the problems of low efficiency, complex operation, high manpower and time costs in manual measurement, and the problem of gap and flushness in robot measurement is that the feedback of measurement and analysis results on the gap and flushness problem is delayed, and the problem cannot be discovered as early as possible to guide the improvement of the production end and avoid the gap and flushness size problems in large quantities. This restricts the further improvement and optimization of the quality system of the automobile manufacturing industry. Therefore, in the field of body gap and flushness prediction, many well-known car companies and suppliers are actively exploring the use of production data mining and digital technology to optimize the current quality inspection process. For example, some car companies have introduced handheld laser measurement technology and are equipped with corresponding data management systems. Such systems can indeed efficiently replace the tedious manual measurement data processing, thereby greatly improving the efficiency and accuracy of data processing. However, their functions are limited to reflecting the current static condition of the car body, and they mainly rely on experience-based control charts for data management, so they are relatively limited in predictive ability and adaptability. Some car companies also try to build a "linear deviation model" based on 3DCS software to simulate the deviation relationship, and reversely analyze it through mathematical methods to obtain the threshold of quality monitoring, in order to achieve more refined quality management. Similarly, the welding IMARS process monitoring system also adopts a similar mechanism to monitor and predict the matching of the vehicle body. However, this type of method has an obvious shortcoming: they are highly dependent on fixed theoretical formulas and manually adjusted parameters to establish simulation models. Taking the IMARS monitoring system as an example, its theoretical prediction formula contains multiple variables and factors. The theoretical formula is: accuracy problem = 290 measurement error + UCF frequency limit + lack of QTF factor + IMARS data call problem + IMARS logic problem + formula accuracy problem. See data acquisition processing and model application logic for details. Figure 4 However, the actual operation of the system has the following problems: first, some of the calculation logic of its early warning formula is biased, resulting in low accuracy; second, once the production process changes, the accuracy of the simulation model will drop significantly, requiring a lot of time and resources to adjust parameters or build a new model; finally, the system fails to fully consider the impact of process variation, making the quality monitoring process appear relatively isolated and fragmented.

[0071] Therefore, the automotive industry has made some progress in gap and flushness prediction, but there is still room for improvement in prediction accuracy and universal applicability.

[0072] In order to solve the problem of low accuracy in the prediction of the gap and flushness quality of the vehicle body, a first embodiment of the method for predicting the gap and flushness quality of the vehicle body of the present invention is proposed based on the structure of the vehicle.

[0073] Reference Figure 5 , Figure 5 It is a schematic flow chart of the first embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention.

[0074] In this embodiment, the vehicle body gap and flushness quality prediction method includes:

[0075] Step S10: obtaining original sample data of the vehicle body gap surface, wherein the original sample data includes cause system data and result system data.

[0076] It should be understood that in this embodiment and the following embodiments, the body gap flushness quality prediction system can be used as the execution subject, and the body gap flushness quality prediction system can execute processes such as data collection, processing, analysis, and model building. The body gap flushness quality prediction system can be set in a server. In addition, the body gap flushness quality prediction system can also establish a connection with an external device to obtain relevant data, such as a 3D measuring machine.

[0077] It should be noted that the body gap surface refers to two surfaces with gaps on the body, and the relative position between the two body gap surfaces can change. For example, the frame and the door are two body gap surfaces. When the door is closed, there is only one gap between the door and the frame. The original sample data refers to the body data during the vehicle manufacturing process. The original sample data includes factor system data and result system data, wherein the factor system data refers to the point data of the key points of the body skeleton after the body welding is completed and before the body is sprayed or assembled. The factor system data will affect the quality of the body gap surface difference. For example, there is a raised point on the A-pillar on the frame. Under the condition of normal execution of the subsequent spraying and assembly process, there will also be a raised point at the A-pillar of the finished vehicle, which will lead to poor quality of the body gap surface difference. The result system data is the data corresponding to the final gap surface difference final detection result after the vehicle is assembled, such as whether there is a problem of excessive gap deviation in the door. Among them, the factor system data corresponds to the result system data, and one factor system data corresponds to one result system data. The original sample data is usually historical data, which includes a large amount of factor series data and the same amount of result series data as the factor series data, thereby ensuring sufficient data volume.

[0078] In a specific implementation, the acquisition of the original sample data can be done by first acquiring one of the factor system data separately, and then acquiring the other one after one of the data is acquired. The factor system data can use a 3D measuring machine to obtain high-precision body skeleton point data, and then use high-precision 3D multi-axis measuring machine equipment to conduct a comprehensive and detailed measurement of the body skeleton. This method can capture the precise position coordinate data of each key part of the body, thereby obtaining a set of accurate factor system data. A large amount of factor system data can be obtained in the same way. The result system data can be obtained by measuring the outer contour of the body and the key positions of the matching after the vehicle is assembled, and evaluating whether the gap and face difference are qualified to obtain the result system data. Similarly, by measuring vehicles after different assemblies in the same way, a large amount of result system data can be obtained. In addition, when a data storage structure is set, the original sample data can also be directly obtained by reading the data storage structure.

[0079] Step S20: performing cluster analysis on the result system data, and removing noise data in the result system data after the cluster analysis to obtain result system training data.

[0080] It should be understood that when the original sample data is obtained, there are some interferences in the original sample data due to human factors, measurement failure, equipment failure or accidental conditions, which leads to certain interferences in the result data of the original sample data. Noise data is the result data caused by this part of interference. For example, in the process of measuring the result data, the hand of the measurement engineer shakes involuntarily, resulting in inaccurate noise data in the result data.

[0081] It should be noted that cluster analysis is to divide data into certain categories and classify data into different classes or clusters. Therefore, the data in the same cluster has great similarities, while the data between different clusters have great differences. For example, due to accidental data changes, the result data is that the quality of the body spacing difference is unqualified, then there are fewer result data in this category, and the result data in this category can be identified as noise data. Structural training data is the result data that can be directly used to build an accurate model after removing the noise data.

[0082] In a specific implementation, cluster analysis is performed on the result system data according to a preset standard, and then each type of result system data after the cluster analysis is identified to confirm whether the type of result system data is noise data. In the case of noise data, the noise data in the result system data is removed to obtain the result system training data. For example, cluster analysis is performed on the result system data according to the causes of the body gap and face difference quality, and then the number of result system data specifically included in each type of data after the cluster analysis is determined, and the result system data is confirmed to be accidental data, i.e., noise data. Finally, the accidental data in the result system data is filtered out to obtain the result system training data.

[0083] Step S30: selecting factor system training data that affects the gap and flushness quality from the factor system data according to the result system data.

[0084] It should be noted that the factor system training data is the factor system data obtained after screening the factor system data, which can directly build an accurate model. Because there is a certain correlation between the factor system data and the result system data. For example, there are multiple point data in the factor system data on the door, and the abnormality of each point data may cause the result system data of the door to be reflected as the body gap face difference quality is unqualified. However, the factor system data may include some point data that will not affect the gap face difference quality, such as the point data of the internal point of the frame. This point data will not affect the gap face difference quality, and it is meaningless to use this point data to build a model. In addition, when the body is divided, the factor system data of the left door will not affect the result system data of the right door; the factor system data of the front door will not affect the result system data of the rear door. Therefore, in the case of a part of the body, the factor system data that does not belong to the part can also be filtered out, so as to accurately train the model for the part of the body.

[0085] Step S40: constructing a poor quality prediction model based on the random forest method according to the result training data and the factor training data.

[0086] It can be understood that the random forest method is an integrated learning method that improves the accuracy and stability of the model by building multiple decision trees and combining their prediction results. The core of the random forest method is to build a large number of decision trees and combine the prediction results of these trees by voting or averaging to improve the overall prediction performance.

[0087] It should be noted that the poor quality prediction model is a model that predicts the body gap surface difference quality after the body is assembled before the body is painted or assembled. The poor quality prediction model is a big data model. When the result system training data and the factor system training data are sufficient and accurate, the poor quality prediction model can output a body gap surface difference quality prediction result with high accuracy.

[0088] In the specific construction process, each set of factor system training data and result system training data can be selected to establish each decision tree using the random forest method, and then each established decision tree can be combined to obtain a poor quality prediction model.

[0089] Step S50: predicting the vehicle body gap surface difference quality according to the difference quality prediction model.

[0090] It is understandable that after the poor quality prediction model is constructed, after the body welding is completed, the factor system data of the current body can be directly collected and then the factor system data can be input into the poor quality prediction model to predict the body gap surface difference quality. Of course, in the specific pre-charging process, the body can be predicted as a whole, and the body can also be divided into different areas, and then predictions can be made for different areas. For example, all the point data of the body in the factor system data can be directly input into the poor quality prediction model, and the body gap surface difference quality of the entire body can be directly obtained; the point data at the door in the factor system data can be directly input into the poor quality prediction model, and the body gap surface difference quality at the door can be directly obtained.

[0091] In the present embodiment, a method for predicting the quality of gap face difference of a vehicle body is provided, and the method for predicting the quality of gap face difference of a vehicle body comprises: obtaining original sample data of the gap face of a vehicle body, wherein the original sample data comprises: factor system data and result system data; performing cluster analysis on the result system data, and removing the noise data in the result system data after cluster analysis to obtain result system training data; selecting factor system training data that affects the quality of gap face difference from the factor system data according to the result system data; constructing a poor quality prediction model based on the random forest method according to the result system training data and the factor system training data; and predicting the quality of gap face difference of a vehicle body according to the poor quality prediction model. The original sample data is filtered to remove the noise data in the result system data and the factor system training data that affects the quality of gap face difference is extracted, and then a poor quality prediction model with high accuracy is constructed using the filtered original sample data, and then the poor quality prediction model is used to realize accurate prediction of the quality of gap face difference of a vehicle body.

[0092] Based on the above first embodiment, a second embodiment of the vehicle body gap flushness quality prediction method of the present invention is proposed. Figure 6 , Figure 6This is a schematic diagram of the first process flow of the second embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention.

[0093] In this embodiment, the step S20 further includes:

[0094] Step S201: Integrate the abnormal problem data in the result system data.

[0095] It should be understood that abnormal problem data refers to a collection of result system data in which there are problems with the quality of the body gap and flushness in the result system data. Abnormal problem data includes abnormal result system data caused by abnormal tendency characteristics of the cause system data and abnormal result system data caused by noise data. Abnormal data is part of abnormal problem data. In the process of determining the noise data in the result system data, the abnormal problem data needs to be determined first.

[0096] In a specific implementation, the result system data may be checked one by one to see whether they are abnormal problem data indicating that the vehicle body gap and flushness quality is unqualified, and then all abnormal problem data obtained after checking one by one may be integrated to obtain the abnormal problem data.

[0097] Step S202: constructing a cluster analysis model for matching abnormal problems according to the last occurrence time, repair consumption time and cumulative occurrence frequency in the abnormal problem data.

[0098] It can be understood that the last occurrence time refers to the moment when the abnormal problem data last appeared abnormal, and the last occurrence time can reflect the urgency of the abnormality in the body gap and surface difference quality. The rework time refers to the time required to return to the repair shop or manufacturer for repair of the abnormal problem data. The rework time is directly related to the cost of repair. Therefore, the rework time can reflect the economy of the abnormality in the body gap and surface difference quality. The cumulative occurrence frequency refers to the frequency of occurrence of the abnormal problem data within a certain period of time. The higher the cumulative occurrence frequency, the more serious the situation of the abnormal problem data. The cluster analysis model of matching abnormal problem is a model used to perform cluster analysis on abnormal problem data. The cluster analysis model of matching abnormal problem can cluster different types of abnormal problem data, thereby dividing different types of abnormal problem data.

[0099] In a specific implementation, the last occurrence time, repair consumption time and cumulative occurrence frequency of abnormal problems can be extracted, and then a cluster analysis model of matching abnormal problems can be constructed based on the k-means algorithm (taking K=3).

[0100] Step S203: clustering the abnormal problem data using a cluster analysis model to obtain clustering results of the result system data.

[0101] It is understandable that, referring to Figure 7 , when the cluster analysis model is determined, the abnormal problem data can be clustered through the cluster analysis model, and the mean closest to the median point is continuously taken. The calculation result is the aggregation of the data, so as to obtain the clustering result of the abnormal problem data.

[0102] Step S204: determining noise data in the result series data from the clustering results according to the Pareto principle.

[0103] It should be understood that the Pareto principle is a data selection principle. In this embodiment, the Pareto principle can be used to determine the noise data in the result data by the types and specific quantities of different abnormal problem data. For example, some abnormal problem data are accidental data, and the number of occurrences of the accidental data is small. When the number of accidental data is less than a certain threshold, the abnormal problem data corresponding to the accidental data can be determined as noise data.

[0104] In this embodiment, abnormal problem data with a long rework time, the highest cumulative occurrence frequency and still occurring in recent years can be identified as abnormal problem data caused by the main cause data; and for abnormal problem data with a long rework time, the highest cumulative occurrence frequency but occasional batch problems, the abnormal problem data is likely to be abnormal problem data caused by accidental events, so the abnormal problem data is noise data; similarly, for abnormal problem data with a very short rework time, the abnormal problem data is likely to be abnormal problem data caused by measurement errors, and the abnormal problem data is also identified as noise data.

[0105] Step S205: removing noise data from the result system data to obtain result system training data.

[0106] It is understandable that since noise data will have a certain impact on the accuracy of the poor quality prediction model, when the noise data is determined, the noise data in the result system data can be removed to obtain the result system data that does not include the noise data, that is, the result system training data.

[0107] In addition, refer to Figure 8 , Figure 8 This is a second flow chart of the second embodiment of the vehicle body gap flushness quality prediction method proposed by the invention. In this embodiment, step S30 includes:

[0108] Step S301: Determine the abnormal probability of gap and flushness quality abnormality based on the result data.

[0109] It is understandable that inaccurate factor data will also affect the accuracy of the poor quality prediction model. Therefore, in order to obtain an accurate poor quality prediction model, the factor data needs to be processed.

[0110] It should be noted that there is a direct correlation between the result data and the cause data. When processing the cause data, it is necessary to combine the result data. Among them, the abnormal probability is the probability of abnormal problem data appearing in the result data. The abnormal probability can be directly determined based on the specific value of the abnormal problem data and the specific value of the result data.

[0111] Step S302: Analyze the result system data to determine the maximum likelihood estimation of the gap and flushness quality problem and the posterior probability of the point data corresponding to the gap and flushness quality problem.

[0112] It should be understood that in this embodiment, the prior probability of the point data in the factor system data can be determined by the result system data and the Bayesian probability formula, and then the prior probability can be used to determine the factor system training data in the factor system data that affects the gap face difference quality.

[0113] It should be noted that the Bayesian probability formula P(Y|X)=P(X|Y)*P(Y) / P(X); wherein P(Y|X) is the posterior probability, P(X|Y) is the maximum likelihood estimate, P(X) is the abnormal probability, and P(Y) is the prior probability. When determining the prior probability, it is necessary to obtain the maximum likelihood estimate of the gap face difference quality problem and the posterior probability of the point data corresponding to the gap face difference quality problem; in this embodiment, when the result system data is known, the maximum likelihood estimate of the gap face difference quality problem and the posterior probability of the point data corresponding to the gap face difference quality problem can be obtained by analyzing the result system data.

[0114] Step S303: determining the prior probability of the point data by using the Bayesian probability formula according to the abnormal probability, the maximum likelihood estimate and the posterior probability.

[0115] It can be understood that, when the abnormal probability, the maximum likelihood estimate and the posterior probability are determined, the Bayesian probability formula can be directly used to calculate the prior probability of the point data in the factor system data.

[0116] Step S304: selecting factor system training data that affects the gap and flushness quality from the factor system data according to the prior probability of the point data.

[0117] It can be understood that, considering that there is a certain relationship between the prior probability and the factor system data, when the prior probability is obtained through the result system data, the prior probability can be used to determine the factor system data that causes abnormality in the result system data. For example, when it is determined that there is abnormal problem data in the result system data, the point data corresponding to the abnormal problem data can be selected according to the prior probability. The point data will affect the quality of the gap difference, and then cause the corresponding abnormal problem data in the result system data. Therefore, when the prior probability of the point data is determined, the prior probability of the point data can be directly used to select the factor system training data that affects the quality of the gap difference from the factor system data; and then remove the factor system data that will not affect the quality of the gap difference from the factor system data to obtain the factor system training data.

[0118] Based on the first embodiment or the second embodiment, a third embodiment of the vehicle body gap flushness quality prediction method of the present invention is proposed. Fig. 9 , Fig. 9 This is a schematic diagram of the first process flow of the third embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention.

[0119] In this embodiment, step S40 includes:

[0120] Step S401: split the result system training data and the factor system training data according to a preset splitting standard to obtain sample training set data and sample verification set data.

[0121] It should be understood that when constructing a poor quality prediction model, in order to ensure the accuracy and stability of the poor quality prediction model, the result system training data and part of the factor system training data used to construct the poor quality prediction model can be used as verification data, that is, sample verification set data. When the model is built, the sample verification set data can be used to verify the model so as to obtain a poor quality prediction model that meets the requirements for stability and accuracy.

[0122] The sample training set data is the data used to construct and train the poor quality prediction model; the sample validation set data is the data used to verify the stability and accuracy of the poor quality prediction model. The preset splitting standard is a standard pre-set for splitting the result system training data and the factor system training data. The preset splitting standard can be set by the user or by the system according to the results of the model training.

[0123] In the specific division process, the result system training data and the factor system training data can be preprocessed first, and the data in the result system training data and the factor system training data can be matched one by one to establish a data matrix corresponding to the result system training data and the factor system training data; then most of the data matrix is ​​selected as the sample training set data, and a smaller part is selected as the sample verification set data. For example, 90% of the data in the data matrix is ​​selected as the sample training set data, and 10% of the data is selected as the sample verification set data.

[0124] Step S402: constructing an initial poor quality prediction model based on the random forest method according to the sample training set data.

[0125] It should be noted that the initial poor quality prediction model is a poor quality prediction model that has been constructed but has not been verified for stability and accuracy. The initial poor quality prediction model has low accuracy and stability in predicting the body gap and surface difference quality.

[0126] It can be understood that after the result system training data and the factor system training data are split, the sample training set data can be directly used to construct an initial poor quality prediction model using the random forest method.

[0127] Step S403: using the sample validation set data to verify the stability and accuracy of the initial poor quality prediction model.

[0128] It is understandable that after the initial poor quality prediction model is built, the sample validation set data previously split out can be directly used to verify the initial poor quality prediction model. Among them, stability refers to the ability of the initial poor quality prediction model to stably output the body gap surface difference quality prediction results; accuracy refers to the accuracy of the initial poor quality prediction model in predicting the body gap surface difference quality.

[0129] In the specific verification process, the factor system data in the sample verification set data can be input into the initial poor quality prediction model one by one, and then the prediction results output by the initial poor quality prediction model are compared with the result system data in the sample verification set data, and the accuracy of the initial poor quality prediction model is determined according to the comparison results; and the stability of the initial poor quality prediction model is determined according to the amount of data of the factor system data input into the initial poor quality prediction model and the amount of data of the prediction results output by the initial poor quality prediction model.

[0130] Step S404: when the stability and accuracy verification are passed, a poor quality prediction model is obtained.

[0131] It is understandable that after verification, it is necessary to test whether the stability and accuracy of the initial poor quality prediction model meet the use requirements. You can first set the verification standards for accuracy and stability, and then confirm whether the actual verification results of the stability and accuracy of the initial poor quality prediction model meet the set verification standards, and then determine whether the initial poor quality prediction model has passed the verification. For example, during accuracy verification, you can set a verification standard that the initial poor quality prediction model needs to reach more than 90%, and when the accuracy of the initial poor quality prediction model is greater than 90% during actual verification, it can be determined that the accuracy verification has passed.

[0132] Reference Fig.10 , Fig.10 This is a schematic diagram of the second process flow of the third embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention.

[0133] In this embodiment, after step S40, the following steps are further included:

[0134] Step S60: Obtaining the information gain of each decision node in the process of constructing the poor quality prediction model.

[0135] It should be understood that the influence of each point data in the factor system data on the body gap flushness quality is not the same. For example, the factor system data at the door position includes seven point data, one of which has a small influence on the body gap flushness quality. Even if there is a large error in the point data, the body gap flushness quality may still meet the body gap flushness quality standard; however, even if there is a small error in another point data, it may cause the body gap flushness quality to still fail to meet the body gap flushness quality standard. It can be seen that different point data in the factor system data have different influences on the body gap flushness quality.

[0136] It should be noted that the information gain of the decision node is the change in information entropy brought by each decision node on the decision tree during the construction of the quality difference prediction model. The information gain generated by the point data after being added to the decision point is not the same. The greater the information gain, the greater the impact of the point data on the body gap surface difference quality.

[0137] In a specific implementation, during the construction of the poor quality prediction model, the information gain of each point data after being added can be determined according to the information entropy before the point data is added and the information entropy after the point data is added.

[0138] Step S70: determining the data weight of each point data in the factor system data according to the information gain of each decision node.

[0139] It should be noted that the data weight of the point data refers to the degree of influence of the point data on the quality of the body gap and flushness. The greater the data weight of the point data, the greater the influence on the quality of the body gap and flushness. Point data with a larger data weight may have a small deviation, which may also cause the body gap and flushness quality to not meet the standard; while point data with a smaller data weight may have a large deviation, but the body gap and flushness quality may still meet the standard. Based on the data weight, it can be understood which point data needs to be focused on after the body welding is completed.

[0140] It is understandable that the point data will directly affect the information gain of the decision node, and the information gain can directly reflect the impact on the quality of the body gap and flushness. Therefore, when the information gain of each decision node is determined, the degree of influence of each point data on the quality of the body gap and flushness can be determined. In the specific implementation, the data weight of each point data in the factor system data can be directly determined according to the information gain of each decision node; the greater the information gain, the greater the data weight corresponding to the point data.

[0141] Step S80: sorting and outputting the data of each point according to the data weight.

[0142] It can be understood that when the data weight of each point data is determined, each data weight can be fed back to the client. After understanding the data weight of the specific point data, the user can determine the influence of each point data in the factor system data on the vehicle body gap and face difference quality according to the data weight.

[0143] When the data weight of each point data in the factor system data is determined, the weight of each point data can be sorted and then fed back. At this time, the data weight of each point can be directly understood, and then in the process of vehicle body processing, the points with larger data weight can be focused on according to the data weight of the point data.

[0144] Based on any one of the first to third embodiments, a fourth embodiment of the vehicle body gap flushness quality prediction method of the present invention is proposed. Fig.11 , Fig.11 This is a schematic diagram of the first process flow of the fourth embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention.

[0145] In this embodiment, after step S10, the following steps are further included:

[0146] Step S101: identifying abnormal data in the original sample data by using an isolation forest algorithm or a clustering algorithm.

[0147] It should be understood that after the original sample data is acquired, the original sample data is historical data, and the original sample data may include abnormal data, including defective data, duplicate data, etc. In the case where the original sample data includes abnormal data, constructing a poor quality prediction model will lead to reduced accuracy or stability of the model. For example, in the case where there is only factor data but no result data corresponding to the factor data, the factor data will cause the model to be unable to output normal result data, which will have a greater impact on the stability of the model.

[0148] It should be noted that the isolation forest algorithm is an algorithm for mining abnormal data or outlier data. The isolation forest algorithm can isolate the abnormal data in the original sample data, and then identify the abnormal data in the original sample data. The clustering algorithm is an algorithm that clusters data according to attributes. The clustering algorithm can continuously aggregate data with the same attributes to the same place, while data with different attributes or large attribute differences will not be aggregated to the same place. Therefore, the above isolation forest algorithm or clustering algorithm can identify abnormal data in the original sample data.

[0149] Step S102: removing abnormal data from the original sample data.

[0150] Step S103: the original sample data after the abnormal data is removed is identified as new original sample data.

[0151] It should be understood that abnormal data will affect the stability and accuracy of the poor quality prediction model. Therefore, before preprocessing the original sample data, the abnormal data in the original sample data can also be removed to obtain new original sample data. In the subsequent steps, the new original sample data is directly processed and the poor quality prediction model is constructed to obtain an accurate and stable poor quality prediction model.

[0152] In addition, in this embodiment, referring to Fig.12 , Fig.12 This is a schematic diagram of the second process of the fourth embodiment of the vehicle body gap and flushness quality prediction method proposed by the invention. Before step S40, it also includes:

[0153] Step S41: convert all feature values ​​in the result system training data and the factor system training data into the same dimension.

[0154] It should be understood that the result system training data and the factor system training data are data used to directly construct the poor quality prediction model. However, a large amount of result system training data and the factor system training data are needed in the process of constructing the poor quality prediction model. In the case of a large amount of data, there may be different system training data and factor system training data that may be in different dimensions. Therefore, before using the result system training data and the factor system training data to construct the poor quality prediction model, the result system training data and the factor system training data must be converted to the same dimension to avoid confusion in model construction or insufficient model accuracy due to different data dimensions during model construction.

[0155] It should be noted that dimension refers to the specific representation form of the result system training data and the factor system training data, such as data units, the number of data in a unit array, two-dimensional data, three-dimensional data, etc.

[0156] In the specific implementation, you can first select a data dimension standard and then convert all the result system training data and factor system training data to this dimension. Of course, you can also first count the dimensions of the result system training data and factor system training data, and use the dimension of the result system training data and factor system training data with relatively large data volume as the standard dimension, and then convert the dimension of the result system training data and factor system training data that are not in this dimension.

[0157] Correspondingly, the step S40 is step S40': constructing a poor quality prediction model based on the random forest method according to the result system training data and the factor system training data whose characteristic values ​​are in the same dimension.

[0158] It can be understood that after the dimensions of the result system training data and the factor system training data are unified, each group of factor system training data and result system training data can be selected to establish each decision tree using the random forest method, and then each established decision tree can be combined to obtain a poor quality prediction model.

[0159] In addition, to achieve the above-mentioned purpose, the present invention also provides a vehicle body gap flushness quality prediction system, which is used to execute the vehicle body gap flushness quality prediction method recorded in any of the above-mentioned embodiments.

[0160] In addition, to achieve the above purpose, the present invention also provides a storage medium, on which a vehicle body gap flushness quality prediction program is stored, and when the vehicle body gap flushness quality prediction program is executed by a processor, the steps of the vehicle body gap flushness quality prediction method are implemented.

[0161] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting vehicle body gap and flushness quality, characterized in that: The vehicle body gap and flushness quality prediction method comprises: Acquire original sample data of the vehicle body gap surface, wherein the original sample data includes: cause system data and result system data; Performing cluster analysis on the result system data, and removing noise data in the result system data after the cluster analysis to obtain result system training data; Selecting factor system training data that affects the gap and flushness quality from the factor system data according to the result system data; Constructing a poor quality prediction model based on the random forest method according to the result system training data and the factor system training data; The vehicle body gap surface difference quality is predicted according to the difference quality prediction model.

2. The method for predicting vehicle body gap flushness quality as claimed in claim 1, characterized in that: The clustering analysis is performed on the result system data, and the noise data in the result system data after the cluster analysis is removed to obtain the result system training data, including: Integrate abnormal problem data in the result system data; Constructing a cluster analysis model for matching abnormal problems based on the last occurrence time, repair consumption time and cumulative occurrence frequency in the abnormal problem data; The abnormal problem data is clustered through the cluster analysis model to obtain the clustering results of the result system data; Determining noise data in the result series data from the clustering results according to the Pareto principle; The result is training data obtained by removing noise data from the result data.

3. The method for predicting vehicle body gap flushness quality as claimed in claim 1, characterized in that: Selecting factor system training data that affects the gap and flushness quality from the factor system data according to the result system data includes: Determine the abnormal probability of gap and flushness quality abnormality based on the result system data; Analyze the result system data to determine the maximum likelihood estimate of the gap and flushness quality problem and the posterior probability of the point data corresponding to the gap and flushness quality problem; Determine the prior probability of the point data by using a Bayesian probability formula according to the abnormal probability, the maximum likelihood estimate and the posterior probability; The factor system training data affecting the gap and flushness quality are selected from the factor system data according to the prior probability of the point data.

4. The method for predicting vehicle body gap flushness quality as claimed in claim 1, characterized in that: The method of constructing a poor quality prediction model based on the result training data and the factor training data based on a random forest method includes: The result system training data and the factor system training data are split according to a preset splitting standard to obtain sample training set data and sample verification set data; Constructing an initial poor quality prediction model based on the random forest method according to the sample training set data; Using the sample validation set data to verify the stability and accuracy of the initial poor quality prediction model; When the stability and accuracy verification is passed, a poor quality prediction model is obtained.

5. The method for predicting vehicle body gap flushness quality as claimed in claim 1, characterized in that: After the poor quality prediction model is constructed based on the random forest method according to the result system training data and the factor system training data, the method further includes: Obtaining information gain of each decision node in the process of building the poor quality prediction model; Determine the data weight of each point data in the factor system data according to the information gain of each decision node; The data of each point is sorted and output according to the data weight.

6. The method for predicting vehicle body gap flushness quality as claimed in claim 1, characterized in that: After obtaining the original sample data of the vehicle body gap surface, the method further includes: Identify abnormal data in the original sample data by using an isolation forest algorithm or a clustering algorithm; Removing abnormal data from the original sample data; The original sample data after the abnormal data is removed is identified as the new original sample data.

7. The method for predicting vehicle body gap flushness quality as claimed in claim 6, characterized in that: Before constructing the poor quality prediction model based on the random forest method according to the result training data and the factor training data, the method further includes: Convert all feature values ​​in the result system training data and the factor system training data to the same dimension; Accordingly, the poor quality prediction model is constructed based on the random forest method according to the result system training data and the factor system training data, including: A poor quality prediction model is constructed based on the random forest method according to the result training data and the factor training data in which all feature values ​​are in the same dimension.

8. A vehicle body gap and flushness quality prediction system, characterized in that: The vehicle body gap flushness quality prediction system is used to execute the vehicle body gap flushness quality prediction method according to any one of claims 1-7.

9. A storage medium, characterized in that: The storage medium stores a vehicle body gap flushness quality prediction program, and when the vehicle body gap flushness quality prediction program is executed by the processor, the steps of the vehicle body gap flushness quality prediction method according to any one of claims 1 to 7 are implemented.

10. A vehicle, characterized in that: The vehicle comprises: a memory, a processor and a vehicle body gap flushness quality prediction program stored in the memory and executable on the processor, wherein the vehicle body gap flushness quality prediction program implements the steps of the vehicle body gap flushness quality prediction method as claimed in any one of claims 1 to 7 when executed by the processor.

Images