Abnormal factor determination method, electronic equipment and storage medium

Abstract

The invention provides an abnormal factor determination method, electronic equipment and a storage medium, and the abnormal factor determination method comprises the steps: obtaining product data of a plurality of sample products, the product data of each sample product comprising a plurality of data categories, each data category comprising a plurality of data parameters, and the plurality of sample products comprising abnormal products and normal products; performing feature screening on the plurality of data categories of the plurality of sample products to obtain a plurality of feature data categories; based on a target parameter factor prediction model, performing difference analysis on the plurality of data parameters in the plurality of feature data categories, and determining a plurality of target parameter factors; and based on a weight training model, determining the weight of the plurality of target parameter factors, and determining an abnormal factor which causes the abnormal product to be abnormal from the plurality of target parameter factors according to the weight. The method and the device can improve the accuracy of determining the abnormal factors causing the abnormity of the product.

Description

Technical Field

[0001] The present application belongs to the field of intelligent manufacturing and relates to data analysis technology, and specifically to an abnormal factor determination method, electronic device and storage medium. Background Art

[0002] In the field of intelligent manufacturing, analyzing abnormal factors in the production data of abnormal products can help manufacturers determine the common causes of abnormal products, thereby making targeted improvements to the production process and improving product yield and production efficiency.

[0003] Since the data variables generated when producing products are independent variables and the number of abnormal products in the actual production process is relatively small, the association-finding algorithms based on specific logical relationships in other fields are not applicable to the association-finding between the independent variables generated by the above-mentioned abnormal products. The abnormal factor detection model trained using limited abnormal product data in the related technology has a low model accuracy, resulting in a low accuracy in analyzing the abnormal factors in the production data of abnormal products. Summary of the invention

[0004] In view of the above, it is necessary to propose an abnormal factor determination method, electronic device and storage medium, which can solve the problem of low accuracy of product abnormal factor analysis caused by the fact that the variables when producing products are independent variables and the number of abnormal products in the actual production process is small.

[0005] A first aspect of an embodiment of the present application provides a method for determining an abnormal factor, comprising: acquiring product data of a plurality of sample products, the product data of each of the sample products comprising a plurality of data categories, each of the data categories comprising a plurality of data parameters, and the plurality of sample products comprising abnormal products and normal products; performing feature screening on the plurality of data categories of the plurality of sample products to obtain a plurality of feature data categories; performing difference analysis on the plurality of data parameters in the plurality of feature data categories based on a target parameter factor prediction model to determine a plurality of target parameter factors; determining weights of the plurality of target parameter factors based on a weight training model, and determining the abnormal factor that causes the abnormal product to be abnormal from the plurality of target parameter factors according to the weights.

[0006] According to an embodiment of the present application, the target parameter factor prediction model is based on which a difference analysis is performed on multiple data parameters in the multiple feature data categories to determine multiple target parameter factors, including: inputting the data parameters of the multiple feature data categories into the target parameter factor prediction model to obtain prediction values ​​corresponding to each input feature data category; determining the difference between the data parameters of each input feature data category and the corresponding prediction value, and selecting from the multiple feature data categories a feature data category corresponding to a difference greater than or equal to a preset difference as a target parameter factor.

[0007] According to an embodiment of the present application, the method further includes: using the product data of the normal product to train the parameter factor prediction model to obtain the target parameter factor prediction model.

[0008] According to an embodiment of the present application, the multiple data categories include numerical categories and classification categories, and the feature screening of the multiple data categories of the multiple sample products includes: if the number of abnormal products among the multiple sample products is less than or equal to a preset number, feature screening of the numerical category; or, if the number of abnormal products among the multiple sample products is greater than the preset number, feature screening of the numerical category, and feature screening of the classification category.

[0009] According to an embodiment of the present application, the feature screening of the numerical categories includes: determining a probability density function corresponding to each of the numerical categories, the probability density function being used to indicate the distribution probability of each data parameter of each of the numerical categories; determining the distribution probability of the data parameters of any numerical category of the abnormal product based on the probability density function, and if the distribution probability of the data parameters of any numerical category is less than a preset probability threshold, taking any numerical category as the feature data category.

[0010] According to an embodiment of the present application, the classification category includes a first category, and the feature screening of the classification category includes: determining a first proportion of the number of abnormal products corresponding to a first subcategory in the first category in the total number of sample products; determining a second proportion of the number of abnormal products corresponding to a second subcategory in the first category in the total number of sample products; if the deviation between the first proportion and the second proportion is greater than or equal to a preset deviation threshold, using the first category as the feature data category.

[0011] According to an embodiment of the present application, the multiple data categories include numerical categories and classification categories, wherein the numerical categories are used to characterize processing parameters or test parameters in the production process of the sample product; the classification categories are used to characterize at least one of the processing process type, processing fixture type, processing equipment type and detection equipment type in the production process of the sample product.

[0012] According to an embodiment of the present application, the multiple data categories include classification categories and numerical categories, the data parameters of the classification categories include classification values ​​and null values, and the data parameters of the numerical categories include null values ​​and null values; before performing feature screening on the multiple data categories of the multiple sample products, the method also includes preprocessing the product data, and the preprocessing includes: cleaning the product data of null values ​​according to the proportion of null values ​​in the product data; and / or filling the product data with null values ​​according to the category of each data category in the multiple data categories.

[0013] According to an embodiment of the present application, the null value cleaning of the product data based on the proportion of null values ​​in the product data includes: determining the proportion of null values ​​in each data category in all product data, and deleting from the product data the data categories corresponding to the null value proportions exceeding a preset threshold.

[0014] According to an embodiment of the present application, filling null values ​​in the product data according to the category of each data category in the multiple data categories includes: if the data category is a classification category, determining the mode of the data parameters of each classification category in the product data, and using the mode to update the null values ​​in the corresponding classification category; and / or if the data category is a numerical category, determining the statistics of the data parameters of each numerical category in the product data, and using the statistics to update the null values ​​in the corresponding numerical category.

[0015] According to an embodiment of the present application, before determining multiple target parameter factors in the multiple feature data categories based on the target parameter factor prediction model, the method also includes: numerically encoding the data parameters in the classification category and the data parameters in the numerical category in the feature data category, respectively, to convert them into numerical variables, wherein the format of the data parameters in the classification category includes at least one of text, symbols, and character strings.

[0016] According to an embodiment of the present application, the target parameter factor prediction model includes an autoencoder model trained based on an unsupervised algorithm.

[0017] According to an embodiment of the present application, an algorithm used to determine the weight of each target feature variable in the multiple target feature variables includes: Shapley plus interpretation algorithm.

[0018] According to a second aspect of an embodiment of the present application, there is provided an abnormal factor determination device, the abnormal factor determination device comprising: a data acquisition module, for acquiring product data of a plurality of sample products, the product data of each of the sample products comprising a plurality of data categories, each of the data categories comprising a plurality of data parameters, the plurality of sample products comprising abnormal products and normal products; a feature screening module, for performing feature screening on the plurality of data categories of the plurality of sample products to obtain a plurality of feature data categories; a difference analysis module, for performing difference analysis on the plurality of data parameters in the plurality of feature data categories based on a target parameter factor prediction model to determine a plurality of target parameter factors; a weight determination module, for determining weights of the plurality of target parameter factors based on a weight training model, and determining the abnormal factor causing the abnormal product to be abnormal from the plurality of target parameter factors according to the weights.

[0019] A third aspect of an embodiment of the present application provides an electronic device, including a memory and at least one processor, wherein the memory stores at least one instruction, and when the at least one instruction is executed by the at least one processor, the abnormal factor determination method is implemented.

[0020] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the abnormal factor determination method is implemented.

[0021] The abnormal factor determination method provided in the embodiment of the present application preliminarily determines the feature data category that causes the abnormal product by performing feature screening on the product data; performs difference analysis on the feature data category through the prediction model, and can determine the feature with a large data error with the normal product as the target parameter factor; determines the weight of the target parameter factor through the feature weight model, and thus determines the abnormal factor in the target parameter factor according to the weight. Through feature screening, difference analysis and feature weight analysis, the abnormal factors corresponding to the abnormal product can be more accurately identified, thereby realizing the correlation and cause search of the product data of the small sample abnormal product, quickly locating the common problems of the abnormal product, and improving the product productivity and production yield. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1A flowchart of a method for determining an abnormal factor provided in an embodiment of the present application.

[0024] Figure 2 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0025] Figure 3 A schematic diagram of a probability density function provided in accordance with an embodiment of the present application.

[0026] Figure 4 A schematic diagram of a probability density function provided by another embodiment of the present application.

[0027] Figure 5 An example diagram of sub-category proportions provided for an embodiment of the present application.

[0028] Figure 6 An example diagram of sub-category proportions provided for another embodiment of the present application.

[0029] Figure 7 A principle block diagram of an abnormal factor determination device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing the embodiments in one embodiment and are not intended to limit this application.

[0032] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0033] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way. The following embodiments and features in the embodiments may be combined with each other without conflict.

[0034] Figure 1 1 is a flow chart of an abnormal factor determination method provided by an embodiment of the present application. The abnormal factor determination method is applied to electronic devices, such as Figure 2 The electronic device 10 in the flowchart specifically includes the following steps. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted.

[0035] Step S201, obtaining product data of a plurality of sample products.

[0036] In one embodiment, product data of multiple sample products can be obtained from a product quality inspection database provided by a product manufacturer (or "manufacturer"), and the product quality inspection database can be constructed in one or more servers. Multiple sample products may include abnormal products and normal products, and it is known whether each sample product is an abnormal product or a normal product. For example, the sample product may be an electronic device accessory or a complete product, for example, it may be a screen, an electronic device housing component, or other manufactured or processed industrial products. As an example, an abnormal product (or positive sample, NG (not good) sample) may be a screen with bubbles, and a normal product (or negative sample, OK sample) may be a screen without bubbles.

[0037] In one embodiment, the number of multiple sample products can be set according to actual needs. For example, the method provided in the embodiment of the present application can be applied to determine the abnormal factors in the data of a small number of abnormal products. The number of sample products can be an integer less than 1000, for example, 100, 400, etc., and the embodiment of the present application does not impose specific restrictions on this. Generally speaking, the number of negative samples will be much larger than the number of positive samples. Therefore, the number of product data corresponding to negative samples in the product data of multiple sample products is relatively large (for example, accounting for 98%). Subsequent related model training can be performed based on the product data corresponding to the large number of negative samples, thereby meeting the training requirements of the model.

[0038] In one embodiment, the product data of each sample product includes multiple data categories, each data category includes multiple data parameters, and the multiple data categories may include two major categories according to the types of the data parameters, specifically including a numerical category and a classification category.

[0039] In one embodiment, the numerical category is used to characterize the processing parameters or test parameters in the production process of the sample product, such as the processing time and processing temperature of the product, etc., wherein each numerical category is used to characterize a processing parameter or test parameter. The data parameters of the numerical category may include numerical values, which may be quantitative data or variable data, which may be continuous values ​​or discrete values, and which have measurability and operability. For example, the data parameter corresponding to the processing temperature of the product may be a value between 15 degrees Celsius and 70 degrees Celsius. When performing data statistics on the data parameters of the numerical category, the mean, median, mode, standard deviation, etc. may be calculated to describe the central tendency, degree of dispersion, etc. of the data parameters of the numerical category.

[0040] In one embodiment, the classification category is used to characterize at least one of the processing process type (e.g., welding process, assembly process, CNC process), processing fixture type (e.g., welding fixture, assembly fixture), processing equipment type (e.g., welding equipment, assembly equipment, film sticking equipment) and testing equipment (e.g., pressure testing, air tightness testing, water tightness testing, size testing equipment) in the production process of the sample product. As an example, the classification category may include the processing machine, processing point, processing line, processing fixture code, processing public mold code, etc. of the product, wherein each classification category is used to characterize a processing process type, processing fixture type and processing equipment type. The data parameter corresponding to the classification category may include a classification value, and the classification value may include classification data or quality data. The format of the classification value includes at least one of text or symbol, character string, which is used to represent the category of the parameter rather than a specific value. For example, the data parameter corresponding to the XXX processing point of the product may include classification values ​​such as A-1 processing point, A-2 processing point, etc. For example, the data parameter corresponding to the welding processing point of the product may be A-1 welding point, A-2 welding point, etc. When performing data statistics on categorical values, you can use mode, frequency statistics, percentage calculations, etc. to describe the distribution of categorical values ​​in the population.

[0041] In an example, the product data of the plurality of sample products may include a plurality of data rows and a plurality of data columns, wherein each data row corresponds to a sample product; the product data includes a plurality of data columns, wherein each data column corresponds to a data category.

[0042] In one example, the product data of multiple sample products may be presented in a table format, which may be referred to as a product data table. The number of data rows in the product data table is equal to the number of sample products, and the number of data columns in the product data table is equal to the number of all data categories. For example, the product data table may include a total of 50 columns of classification categories, and may include a total of 21 columns of numerical categories. For example, as shown in Table 1, a product data table provided in one embodiment of the present application is provided.

[0043] Table 1

[0044]

[0045]

[0046] For example, as shown in Table 2, it is a product data table provided in another embodiment of the present application.

[0047] Table 2

[0048]

[0049]

[0050]

[0051] Referring to Table 1 and Table 2, each row in the product data table corresponds to a sample product, the header of each column indicates the data category of the corresponding column, and each table element indicates the data parameter of the data category of the sample product in the corresponding row direction in the corresponding column direction. Among them, the classification categories in Table 1 include: product quality inspection results, product quality inspection line type N, SS fixture, XXX processing point and YYY processing point, the data categories in Table 1 include processing time, and the classification categories in Table 2 include: product resolution, product configuration model, laser processing step 1 machine number, laser processing step 2 machine number, sealing operation step 1 machine number, sealing operation step 2 machine number and cleaning test packaging step machine number.

[0052] In one embodiment, due to various problems in data collection, such as data collection omissions, equipment detection failures, data conversion errors, etc., the data parameters of the classification category and the numerical category may also include null values ​​(or missing values). In order to ensure the accuracy and completeness of the data, appropriate pre-processing measures such as null value cleaning and null value filling can be taken to minimize the impact of null values ​​on subsequent data analysis results.

[0053] In one embodiment, the method further includes preprocessing the product data, the preprocessing including: cleaning the product data of null values ​​according to the proportion of null values ​​in the product data; and / or filling the product data with null values ​​according to the category of each data category in multiple data categories.

[0054] In one embodiment, according to the proportion of null values ​​in the product data, performing null value cleaning on the product data includes: determining the null value proportion of each data category in all product data, and deleting the data category corresponding to the null value proportion exceeding a preset threshold from the product data. The preset threshold can be set according to actual needs or data categories, such as 0.1, 0.2, and this application does not make specific restrictions on this.

[0055] In one example, the null value ratio of each data category can be determined as follows: determine the first number of null value tables in the data column corresponding to any data category in the product data table; and take the ratio of the first number to the number of data rows in the product data table (i.e., the total number of sample products) as the null value ratio corresponding to any data category. For example, if the first number of null value tables in the data column corresponding to the data category XXX processing acupoints is 26, and the number of data rows in the product data table is 100, then the null value ratio corresponding to the data category XXX processing acupoints is 26 / 100=0.26, which exceeds the preset threshold value of 0.2, and the data column corresponding to the data category XXX processing acupoints can be deleted as a whole. For example, the data column of the original product data table includes 106 data categories (e.g., 106 columns of data). After deleting the severely missing data columns during the null value cleaning process, the product data table can include the remaining 71 data categories (e.g., 71 columns of data, which can include 50 columns of numerical categories and 21 columns of classification categories).

[0056] By performing null value cleaning on product data, it can help eliminate noise and outliers in product data, improve the quality and reliability of data, and thus improve the prediction accuracy and generalization ability of subsequent models obtained based on data training.

[0057] In one embodiment, null value filling can be performed after performing null value cleaning, or null value filling can be performed on data columns corresponding to data categories whose null value ratio does not exceed a preset threshold while performing null value cleaning. Different methods can be used to fill null values ​​for data of numerical or categorical categories.

[0058] If the data category is a categorical category, determine the mode of the data parameters of each categorical category in the product data, and use the mode to update the null values ​​in the corresponding categorical category; and / or if the data category is a numerical category, determine the statistics (such as mean, median, mode, etc.) of the data parameters of each numerical category in the product data, and use the statistics to update the null values ​​in the corresponding numerical category.

[0059] In one example, if the data category is XXX processing acupoints, the number of occurrences of each data parameter in the data column corresponding to the XXX processing acupoints can be counted. For example, the number of occurrences of the A-1 processing acupoints is 50 times, and the number of occurrences of the A-2 processing acupoints is 30 times. Since the A-1 processing acupoints has the largest number of occurrences, it can be determined that the mode of the data parameters corresponding to the XXX processing acupoints is the A-1 processing acupoints, and the A-1 processing acupoints are used to fill the empty values ​​in the data column corresponding to the XXX processing acupoints.

[0060] In another example, if the numerical category is the processing temperature of the product, the statistics of all data parameters in the data column corresponding to the processing temperature of the product can be counted. For example, the average of all temperature values ​​in the data column corresponding to the processing temperature of the product is 30 degrees Celsius, and 30 degrees Celsius can be used to fill the empty values ​​in the data column corresponding to the processing temperature of the product.

[0061] By filling in null values, we can make up for the information loss caused by missing data, make the data set more complete and accurate, help maintain data consistency in the data set, and ensure that data between different data sources can be compared and integrated with each other, so that the data has higher availability and practicality, so as to improve the prediction accuracy and generalization ability of the subsequent model obtained based on data training.

[0062] Step S202, feature screening is performed on multiple data categories of multiple sample products to obtain multiple feature data categories.

[0063] In one embodiment, in order to determine the cause of abnormal products, product data may be feature screened, thereby determining data most relevant to abnormal products from multiple data categories as feature data categories. Since data categories include two major categories: numerical categories and classification categories, data of different major categories may be selectively screened according to the number of abnormal products.

[0064] In one embodiment, feature screening of multiple data categories of multiple sample products includes: if the number of abnormal products in the multiple sample products is less than or equal to a preset number, feature screening of the numerical category. The preset number can be set according to actual needs, for example, the preset number can be an integer less than 9, such as 1, 2, etc., and the present application does not impose specific restrictions on this.

[0065] In one embodiment, data parameters of numerical categories can usually be directly quantified, and abnormal values ​​may be more easily reflected in data parameters of numerical categories. For example, the temperature of a product is much lower or higher than the normal temperature range of other similar products, which may be an obvious abnormality. Therefore, when the number of abnormal products is small (for example, less than or equal to a preset number), we can focus on the features of numerical categories that can more directly reflect abnormalities and perform feature screening on the data of numerical categories.

[0066] In one embodiment, feature screening of numerical categories includes: determining a probability density function corresponding to each numerical category, the probability density function is used to indicate the distribution probability of each data parameter of each numerical category; determining the distribution probability of data parameters of any numerical category of abnormal products according to the probability density function, and if the distribution probability of data parameters of any numerical category is less than a preset probability threshold, then any numerical category is used as a feature data category. The preset probability threshold can be set according to actual needs, such as 0.2, and this application does not impose specific restrictions on this.

[0067] In one embodiment, the probability density function corresponding to each value category in all pre-processed product data can be determined based on a kernel density estimation (KDE) algorithm. The probability density function calculated by the KDE algorithm can indicate the distribution probability of each data parameter of each value category.

[0068] For example Figure 3 As shown, it is a schematic diagram of the probability density function provided by an embodiment of the present application, wherein the horizontal axis represents the value of the data parameter, and the vertical axis represents the distribution probability of the data parameter, for example, the horizontal axis represents the value of the processing temperature, and the vertical axis represents the distribution probability of the processing temperature. It can be seen that the values ​​of the processing temperature of most normal samples are concentrated around 50, while the values ​​of the processing temperature of a small number of abnormal samples (the position indicated by the dotted line) are concentrated around 10, and the distribution probability of the values ​​of the processing temperature of the abnormal samples being concentrated around 10 is relatively small, less than the probability threshold of 0.2, so it can be determined that the concentration of the values ​​of the processing temperature around 10 is a small probability factor causing the abnormal samples to appear, and it can be further determined that the characteristic data category causing the abnormal samples to appear abnormal may be the processing temperature.

[0069] For example Figure 4As shown, it is a schematic diagram of a probability density function provided by another embodiment of the present application, wherein the horizontal axis represents the value of the data parameter, and the vertical axis represents the distribution probability of the data parameter, for example, the horizontal axis represents the value of the processing temperature, and the vertical axis represents the distribution probability of the processing temperature. It can be seen that the values ​​of the processing temperature of most normal samples are concentrated around 50, and the values ​​of the processing temperature of a small number of abnormal samples (the position indicated by the dotted line) are also concentrated around 50, and the distribution probability that the values ​​of the processing temperature of the abnormal samples are concentrated around 50 is the same as the distribution probability that the values ​​of the processing temperature of the normal samples are concentrated around 50, which is greater than the probability threshold of 0.2. Therefore, it can be determined that the concentration of the processing temperature values ​​around 50 is not a small probability factor that causes the abnormal samples to appear, and it can be further determined that the characteristic data category that causes the abnormal samples to appear abnormal is not the processing temperature.

[0070] In one embodiment, if the number of abnormal products is large, the features that rely only on the numerical category may not be sufficient to fully resolve the source of the abnormal products, and feature screening can be performed in combination with the numerical category and the classification category. Therefore, feature screening of multiple data categories of multiple sample products includes: if the number of abnormal products in the multiple sample products is greater than a preset number, feature screening of the numerical category and feature screening of the classification category.

[0071] In one embodiment, the classification category includes a first category, and feature screening of the classification category includes: determining a first proportion of the number of abnormal products corresponding to the first sub-category in the first category in the total number of sample products; determining a second proportion of the number of abnormal products corresponding to the second sub-category in the first category in the total number of sample products; if the deviation between the first proportion and the second proportion is greater than or equal to a preset deviation threshold, the first category is used as a feature data category. Among them, the first category is used to distinguish the classification categories of other categories, and the first category can also include multiple other sub-categories such as the third sub-category. The preset deviation threshold can be set according to actual needs, for example, 0.01, 0.02, etc., and this application does not impose specific restrictions on this.

[0072] For example Figure 5As shown, it is an example diagram of the sub-classification proportion provided by an embodiment of the present application. The first category can be XXX processing acupoints, and XXX processing acupoints can include two sub-classifications (or two classification values), for example, the first sub-classification is the A-1 position, and the second sub-classification is the A-2 position. Among them, the total number of sample products is 400, the number of sample products corresponding to the A-1 position of the XXX processing acupoints is 100, of which the number of NG products is 3, that is, the first proportion is 3 / 100; the number of sample products corresponding to the A-2 position of the XXX processing acupoints is 300, of which the number of NG products is 1, that is, the second proportion is 1 / 300; the deviation between the first proportion and the second proportion is 8 / 300, which is greater than the preset deviation threshold of 0.01, so it can be determined that the A-1 position and the A-2 position of the XXX processing acupoints have obvious differences, and the XXX processing acupoints are used as feature data categories.

[0073] For example Figure 6 As shown, it is an example diagram of the sub-classification proportion provided by another embodiment of the present application. The first category can be a YYY processing acupoint, and the YYY processing acupoint can include two sub-classifications (or two classification values), for example, the first sub-classification is the A-1 position, and the second sub-classification is the A-2 position. Among them, the total number of sample products is 400, the number of sample products corresponding to the A-1 position of the YYY processing acupoint is 200, of which the number of NG products is 2, that is, the first proportion is 2 / 200; the number of sample products corresponding to the A-2 position of the YYY processing acupoint is 200, of which the number of NG products is 2, that is, the second proportion is 2 / 200; the deviation between the first proportion and the second proportion is 0, which is less than the preset deviation threshold of 0.01, so it can be determined that the A-1 position and the A-2 position of the YYY processing acupoint do not have a difference, and the YYY processing acupoint is not used as a feature data category.

[0074] In one embodiment, feature screening of multiple data categories includes feature screening of multiple data columns in a product data list. Data columns that do not belong to feature data categories may be deleted to obtain an updated product data list, and the updated product data list may be used to provide a basis for subsequent data analysis.

[0075] Through the above embodiments, data of different major categories can be selectively screened according to the number of abnormal products, and targeted common feature analysis can be performed in combination with feature distribution and numerical relationship to achieve preliminary determination of abnormal feature data, which is convenient for training related models in subsequent processes.

[0076] Step S203: Based on the target parameter factor prediction model, a difference analysis is performed on a plurality of data parameters in a plurality of feature data categories to determine a plurality of target parameter factors.

[0077] In one embodiment, in order to make the product data available for model training, the data parameters of different categories may be numerically encoded first to make the different data parameters comparable, and then the numerically encoded data may be used for model training. Specifically, before step S203, the method provided in the embodiment of the present application further includes the following steps:

[0078] Step S301 , numerically encode the data parameters in the classification category and the data parameters in the numerical category in the feature data category respectively, so as to convert them into numerical variables respectively.

[0079] In one embodiment, since different data categories have different data types (e.g., the data types of numerical values ​​and classification values ​​are different), and different data parameters are used to characterize different parameter categories (e.g., the A-1 position and the A-2 position of the XXX processed acupuncture point are different), different data parameters may not be comparable. To solve the above problem, the data parameters in the classification category and the data parameters in the numerical category in the feature data category can be numerically encoded respectively to be converted into numerical variables respectively, wherein the format of the data parameters in the classification category includes at least one of text, symbols, and character strings.

[0080] In one example, the Bayesian Target Encoding (BTE) algorithm can be used to numerically encode the data parameters in the numerical category. Specifically, the category_encoders library in the Python programming language can be used to implement the above encoding process. When the data parameters in the numerical category are numerically encoded, the original values ​​can be retained, and the data parameters of the classification category after numerical encoding can also be processed with the data parameters in the numerical category using the standardization or normalization method, so that the data of each data category are comparable, so that they can be used for subsequent model training. For example, the numerical variable corresponding to the A-1 position of the XXX processing acupoint can be 0.11, and the numerical variable corresponding to the processing temperature of 50 degrees Celsius can be 0.15. In other embodiments, the character encoding algorithm in the H2O algorithm can also be used to implement the above numerical encoding. The H2O algorithm can use the characteristics of the enumerated data type (or factor type) to map a string to a numerical value, and the mapping does not need to be based on the dictionary order of the string or any other fixed order, so that data preprocessing is more flexible and helps subsequent data analysis.

[0081] In one example, the numerical encoding algorithm can take into account the distribution (e.g., frequency of occurrence) and other statistical measures (e.g., variance, skewness, etc.) of the target variable (e.g., any categorical value), so the same character may be converted into different numerical values ​​after numerical encoding. The numerical encoding algorithm helps to more accurately represent the relationship between the parameters of each category of data, making the data of each data category comparable.

[0082] For example, as shown in Table 3, the data in Table 2 are obtained after performing operations such as numerical encoding and normalization.

[0083] Table 3

[0084]

[0085]

[0086] Step S303: Use product data of normal products to train the parameter factor prediction model to obtain a target parameter factor prediction model.

[0087] In one embodiment, the parameter factor prediction model includes a preset autoencoder model (Autoencoder), and the target parameter factor prediction model includes an autoencoder model trained based on an unsupervised algorithm, wherein the product data of a normal product may include numerical variables after numerical encoding of each data category.

[0088] In one embodiment, when constructing a target parameter factor prediction model based on an autoencoder model obtained through unsupervised algorithm training, the autoencoder model can be used as a model for feature learning or data dimensionality reduction, and can perform feature learning or data dimensionality reduction on the original input data (e.g., product data of normal products) to obtain a low-dimensional representation or encoding feature of the original input data; then, the original input data of the autoencoder (e.g., product data of normal products) can be obtained by decoding and reconstructing the low-dimensional representation or encoding feature. It can be understood that the model can predict the normal value of the data parameter of each feature data category by learning the specific value of the data parameter of each feature data category of the product data of normal products. For example, the input product data of normal products can be 0.7001, and the corresponding prediction value output by the target parameter factor prediction model can be 0.7002, wherein the deviation between the input data and the corresponding prediction value meets the preset loss threshold (e.g., 0.05).

[0089] Step S305, input the data parameters of multiple feature data categories into the target parameter factor prediction model to obtain the prediction value corresponding to each input feature data category; determine the difference between the data parameters of each input feature data category and the corresponding prediction value, and select the feature data category corresponding to the difference greater than or equal to the preset difference from the multiple feature data categories as the target parameter factor.

[0090] In one embodiment, the difference between the data parameter of each input feature data category and the corresponding predicted value can be = |data parameter of each input feature data category - corresponding predicted value| / data parameter of each input feature data category. For example, the data parameter of the input feature data category can be 0.15, and the corresponding predicted value can be 0.2, then the difference can be = |0.15-0.2| / 0.15≈0.33. The preset difference can be set according to actual needs, for example, 0.3, and the embodiment of the present application does not make specific restrictions on this.

[0091] In one embodiment, the data parameters of multiple feature data categories of abnormal products can be input into the target parameter factor prediction model to obtain the prediction value corresponding to each input feature data category. Since the model is trained using the product data of normal products, the prediction value of the model tends to output the value corresponding to the data parameter corresponding to the normal product. If the difference between the prediction value of a certain feature data category and the data parameter of the abnormal product actually input into the model is large, the feature data category can be determined as the target parameter factor, that is, the target parameter factor that causes the abnormal product to appear.

[0092] Step S204, based on the weight training model, determine the weights of multiple target parameter factors, and determine the abnormal factor that causes the abnormal product to be abnormal from the multiple target parameter factors according to the weights.

[0093] In one embodiment, the algorithm used to determine the weight of each target feature variable in multiple target feature variables includes: Shapley Value or SHAP value algorithm. Specifically, the weight training model can be a prediction model based on a supervised learning algorithm (such as logistic regression, random forest, gradient boosting machine, etc.), and the weight training model can learn to predict the output (for example, the main abnormal factor causing product abnormality among multiple target parameter factors) based on the input features (for example, multiple target parameter factors); during the model training process, the algorithm will learn the weight of each feature, where the weight represents the relative importance of each feature to the model prediction result; when the model training is completed, the contribution of each feature to the model prediction result can be determined by calculating the SHAP value of each feature, that is, the weight of each target parameter factor, so as to determine the abnormal factor causing abnormal product abnormality from multiple target parameter factors according to the weight.

[0094] In other embodiments, the above-mentioned weight determination can also be implemented based on other models, including but not limited to: Strategic Decision Toolkit (SDT), Local Interpretable Model-agnostic Explanations (LIME), Individual Conditional Expectation / Partial Dependence Plot (ICE / PDP), etc., and this application does not make any specific restrictions on this.

[0095] In one example, as shown in Table 4, the weights of the target parameter factors can be sorted in descending order, and the first number of target parameter factors in the sorted sequence are determined as abnormal factors. The first number can be set according to actual needs, for example, the first number can be 3, and this application does not impose specific restrictions on this.

[0096] Table 4

[0097] Target parameter factor Weight Product resolution 25.5％ Product configuration model 20.4％ Laser processing step 1 machine number 18.1％ Laser processing step 2 machine number 16.5％ Sealing operation steps 1 Machine number 13.2％ Sealing operation step 2 machine number 4.7％ Cleaning Test Packaging Steps Machine Number 1.6％

[0098] The abnormal factor determination method provided in the embodiment of the present application is suitable for determining abnormal factors of small sample product data (for example, 2 abnormal products). By performing feature screening on product data, the feature data category that causes the abnormal product to appear is preliminarily determined; the feature data category is analyzed by the prediction model to determine the feature with a large data error with the normal product as the target parameter factor; the weight of the target parameter factor is determined by the feature weight model, and the abnormal factor in the target parameter factor is determined according to the weight. Through feature screening, difference analysis and feature weight analysis, the abnormal factors corresponding to the abnormal product can be more accurately identified, thereby realizing the correlation and cause search of the product data of the small sample abnormal product, quickly locating the common problems of the abnormal product, and improving product productivity and production yield.

[0099] Figure 7 It is a structural diagram of an abnormal factor determination device provided in one embodiment of the present application.

[0100] In some embodiments, the abnormal factor determination device 80 may include a plurality of functional modules composed of computer program segments. The computer programs of the various program segments in the abnormal factor determination device 80 may be stored in a memory of an electronic device and executed by at least one processor to perform (see Figure 1 Description) The function determined by the anomaly factor.

[0101] In this embodiment, the abnormal factor determination device 80 can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a data acquisition module 801, a feature screening module 802, a difference analysis module 803, and a weight determination module 804. The module referred to in this application refers to a series of computer program segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functional implementation method of each module in the abnormal factor determination device 80 can refer to the above definition of the abnormal factor determination method, and will not be repeated here.

[0102] The data acquisition module 801 is used to acquire product data of a plurality of sample products, wherein the product data of each sample product includes a plurality of data categories, each of the data categories includes a plurality of data parameters, and the plurality of sample products include abnormal products and normal products;

[0103] The feature screening module 802 is used to perform feature screening on the multiple data categories of the multiple sample products to obtain multiple feature data categories;

[0104] The difference analysis module 803 is used to perform difference analysis on the multiple data parameters in the multiple feature data categories based on the target parameter factor prediction model to determine multiple target parameter factors;

[0105] The weight determination module 804 is used to determine the weights of the multiple target parameter factors based on the weight training model, and determine the abnormal factor that causes the abnormal product to be abnormal from the multiple target parameter factors according to the weights.

[0106] Figure 2 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 10 can be a mobile phone, a tablet computer, a smart wearable device, an augmented reality (AR) / virtual reality (VR) device, a laptop, a netbook, an energy storage device, a power distribution device, a vehicle-mounted device, a self-moving device, and other electronic devices. The electronic device can also be a manufacturing device, such as a product testing device, etc. The embodiment of the present application does not impose any restrictions on the specific type of the electronic device.

[0107] like Figure 2 As shown, the electronic device 10 may include a communication module 101, a memory 102, a processor 103, an input / output (I / O) interface 104 and a bus 105. The processor 103 is coupled to the communication module 101, the memory 102 and the I / O interface 104 through the bus 105.

[0108] The communication module 101 may include a wired communication module and / or a wireless communication module. The wired communication module may provide one or more wired communication solutions such as universal serial bus (USB), controller area network bus (CAN, Controller Area Network). The wireless communication module may provide one or more wireless communication solutions such as wireless fidelity (wireless fidelity, Wi-Fi), bluetooth (bluetooth, BT), mobile communication network, frequency modulation (frequency modulation, FM), near field communication technology (near field communication, NFC), infrared technology (infrared, IR), etc.

[0109] The memory 102 may include one or more random access memories (RAM) and one or more non-volatile memories (NVM). The random access memory can be directly read and written by the processor 103, and can be used to store executable programs (such as machine instructions) of the operating system or other running programs, and can also be used to store user and application data. The random access memory may include static random-access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), etc.

[0110] The non-volatile memory may also store executable programs and user and application data, etc., and may be loaded into the random access memory in advance for direct reading and writing by the processor 103. The non-volatile memory may include a disk storage device and a flash memory.

[0111] The memory 102 is used to store one or more computer programs. The one or more computer programs are configured to be executed by the processor 103. The one or more computer programs include multiple instructions, and when the multiple instructions are executed by the processor 103, the abnormal factor determination method executed on the electronic device 10 can be implemented.

[0112] In other embodiments, the electronic device 10 further includes an external memory interface for connecting to an external memory to expand the storage capacity of the electronic device 10 .

[0113] The processor 103 may include one or more processing units, for example, the processor 103 may include an application processor (AP), a modem processor, a graphics processor (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Different processing units may be independent devices or integrated into one or more processors.

[0114] The processor 103 provides computing and control capabilities. For example, the processor 103 is used to execute a computer program stored in the memory 102 to implement the above-mentioned abnormal factor determination method.

[0115] The I / O interface 104 is used to provide a channel for user input or output. For example, the I / O interface 104 can be used to connect various input and output devices, such as a mouse, keyboard, touch device, display screen, etc., so that the user can enter information or visualize information.

[0116] The bus 105 is at least used to provide a channel for mutual communication among the communication module 101 , the memory 102 , the processor 103 , and the I / O interface 104 in the electronic device 10 .

[0117] It is to be understood that the structure illustrated in the embodiment of the present application does not constitute a specific limitation on the electronic device 10. In other embodiments of the present application, the electronic device 10 may include more or fewer components than shown in the figure, or combine some components, or split some components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0118] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the methods in the above-mentioned embodiments of the present application.

[0119] The computer-readable storage medium may be an internal memory of the electronic device described in the above embodiment, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device.

[0120] In some embodiments, the computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the electronic device, etc.

[0121] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0122] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be determined to be beyond the scope of this application.

[0123] In the embodiments provided in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0124] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0125] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for determining an abnormal factor, characterized in that: The method comprises: Acquire product data of a plurality of sample products, wherein the product data of each of the sample products includes a plurality of data categories, each of the data categories includes a plurality of data parameters, and the plurality of sample products include abnormal products and normal products; Performing feature screening on the multiple data categories of the multiple sample products to obtain multiple feature data categories; Based on the target parameter factor prediction model, performing difference analysis on the multiple data parameters in the multiple feature data categories to determine multiple target parameter factors; Based on the weight training model, the weights of the multiple target parameter factors are determined, and the abnormal factor causing the abnormal product to be abnormal is determined from the multiple target parameter factors according to the weights.

2. The abnormal factor determination method according to claim 1, characterized in that: The target parameter factor prediction model is based on which a difference analysis is performed on a plurality of data parameters in the plurality of feature data categories to determine a plurality of target parameter factors, including: Inputting the data parameters of the plurality of feature data categories into the target parameter factor prediction model to obtain a prediction value corresponding to each input feature data category; The difference between the data parameter of each input feature data category and the corresponding predicted value is determined, and the feature data category corresponding to the difference greater than or equal to the preset difference is selected from the multiple feature data categories as the target parameter factor.

3. The abnormal factor determination method according to claim 2, characterized in that: The method further comprises: The parameter factor prediction model is trained using the product data of the normal product to obtain the target parameter factor prediction model.

4. The abnormal factor determination method according to claim 1, characterized in that: The multiple data categories include numerical categories and classification categories, and the feature screening of the multiple data categories of the multiple sample products includes: If the number of abnormal products in the plurality of sample products is less than or equal to a preset number, feature screening is performed on the numerical category; or, If the number of abnormal products in the plurality of sample products is greater than the preset number, feature screening is performed on the numerical category, and feature screening is performed on the classification category.

5. The abnormal factor determination method according to claim 4, characterized in that: The feature screening of the numerical category includes: Determine a probability density function corresponding to each of the numerical categories, wherein the probability density function is used to indicate the distribution probability of each data parameter of each of the numerical categories; According to the probability density function, the distribution probability of the data parameters of any numerical category of the abnormal product is determined. If the distribution probability of the data parameters of any numerical category is less than a preset probability threshold, the any numerical category is used as the characteristic data category.

6. The abnormal factor determination method according to claim 4, characterized in that: The classification category includes a first category, and the feature screening of the classification category includes: Determine a first proportion of the number of abnormal products corresponding to the first subcategory in the first category to the total number of the sample products; Determine a second proportion of the number of abnormal products corresponding to the second subcategory in the first category in the total number of the sample products; If the deviation between the first proportion and the second proportion is greater than or equal to a preset deviation threshold, the first category is used as the feature data category.

7. The abnormal factor determination method according to any one of claims 1 to 5, characterized in that: The multiple data categories include numerical categories and classification categories, wherein the numerical categories are used to characterize processing parameters or test parameters in the production process of the sample product; and the classification categories are used to characterize at least one of the processing process type, processing fixture type, processing equipment type and detection equipment type in the production process of the sample product.

8. The abnormal factor determination method according to claim 1, characterized in that: The multiple data categories include classification categories and numerical categories, the data parameters of the classification categories include classification values ​​and null values, and the data parameters of the numerical categories include numerical values ​​and null values; Before performing feature screening on the multiple data categories of the multiple sample products, the method further includes preprocessing the product data, and the preprocessing includes: Cleaning the product data of null values ​​according to the proportion of null values ​​in the product data; and / or, The product data is filled with null values ​​according to the category of each data category in the plurality of data categories.

9. The abnormal factor determination method according to claim 8, characterized in that: The performing null value cleaning on the product data according to the proportion of null values ​​in the product data comprises: Determine the null value ratio of each data category in all product data, and delete data categories corresponding to the null value ratio exceeding a preset threshold from the product data.

10. The abnormal factor determination method according to claim 8, characterized in that: The filling of the product data with null values ​​according to the category of each data category in the multiple data categories comprises: If the data category is a classification category, determining the mode of the data parameters of each classification category in the product data, and using the mode to update the null values ​​in the corresponding classification category; and / or If the data category is a numerical category, determine the statistics of the data parameters of each numerical category in the product data, and use the statistics to update the null values ​​in the corresponding numerical category.

11. The abnormal factor determination method according to claim 1, characterized in that: Before determining the plurality of target parameter factors in the plurality of feature data categories based on the target parameter factor prediction model, the method further comprises: The data parameters in the classification category and the data parameters in the numerical category in the feature data category are numerically encoded respectively to be converted into numerical variables respectively, wherein the format of the data parameters in the classification category includes at least one of text, symbols, and character strings.

12. The abnormal factor determination method according to claim 1, characterized in that: The target parameter factor prediction model includes an autoencoder model trained based on an unsupervised algorithm.

13. The abnormal factor determination method according to claim 1, characterized in that: The algorithms used to determine the weight of each target feature variable in the plurality of target feature variables include: Shapley plus and explanation algorithms.

14. An electronic device, characterized in that: The electronic device includes a memory and at least one processor, wherein the memory stores at least one instruction, and when the at least one instruction is executed by the at least one processor, the abnormal factor determination method according to any one of claims 1 to 13 is implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the abnormal factor determination method according to any one of claims 1 to 13 is implemented.

Images