CN114444986B

CN114444986B - Product analysis method, system, device and medium

Info

Publication number: CN114444986B
Application number: CN202210371479.4A
Authority: CN
Inventors: 不公告发明人
Original assignee: Chengdu Shuzhilian Technology Co Ltd
Current assignee: Chengdu Shuzhilian Technology Co Ltd
Priority date: 2022-04-11
Filing date: 2022-04-11
Publication date: 2022-06-03
Anticipated expiration: 2042-04-11
Also published as: CN114444986A

Abstract

The invention discloses a product analysis method, a product analysis system, a product analysis device and a product analysis medium, which relate to the technical field of intelligent manufacturing and artificial intelligence, and comprise the following steps: training by using the cleaned and preprocessed data to obtain a classification prediction model; obtaining a predicted value of each production process parameter in the data to be analyzed by using a classification prediction model; the SHAP value of each production process parameter is calculated and obtained based on the predicted value of each production process parameter, and a plurality of key process link names and a plurality of key production process parameters influencing the product quality can be accurately and quickly obtained based on the SHAP value of each production process parameter.

Description

Product analysis method, system, device and medium

Technical Field

The invention relates to the technical field of intelligent manufacturing and artificial intelligence, in particular to a product analysis method, a product analysis system, a product analysis device and a product analysis medium.

Background

The root cause positioning method of the industrial product defects is applied to the tasks of product quality control and process parameter optimization in the field of industrial manufacturing. In the industrial manufacturing process, the reduction of the reject ratio of products is vital to the quality control of enterprises, and the core of the method lies in accurately and rapidly positioning the root cause of defects in the production and processing process, namely detecting and positioning the key process parameters influencing the product quality, and further improving the production process and improving the product quality by analyzing and adjusting the key process parameters.

In the prior art, correlation between each process parameter and the reject ratio of a product in the production and processing process is mainly analyzed by a multivariate statistical analysis method, the process parameter with the largest weight coefficient is found out based on a multivariate linear regression method, or manual screening is carried out by combining business experts with relevant experience of production and processing. However, in the industrial manufacturing field, the process parameters involved in the production and processing process of the product are often thousands of, the number of the process parameters far exceeds the number of the defective products to be analyzed, and very strong correlation exists among different process parameters, which all seriously affect the accuracy of the multivariate statistical analysis method. The manual screening method highly depends on experienced business experts, has low analysis efficiency and seriously influences the product quality control of enterprises.

In the prior art, there are several related product analysis methods, which are respectively:

prior art 1:

the scheme is from: CN111159645A bad root cause positioning method based on product production record and parameters

The specific scheme is as follows:

(1) reading the production history data of the glass products in the appointed time period, and counting the number of samples of the defective glass products; (2) analyzing whether the bad samples of the devices at the same site are aggregated or not; (3) gradually eliminating parameter rows in four rounds to form an auxiliary table; (4) forming an analysis table according to the feature importance scores given by the models; (5) screening attribute feature columns according to the analysis table to form a result table; (6) and positioning the bad root cause of the product according to the result table.

The scheme has the advantages that: combining the process and experience of historical parameter analysis in the industrial manufacturing process, and fitting the industrial manufacturing business scene; the scheme has the following defects: the processing flow excessively depends on the manual experience of a business expert, and meanwhile, the parameter elimination process is greatly influenced by the threshold value setting. In addition, the parameter elimination processing flow is relatively complicated.

Prior art 2:

the scheme is from: CN113590451A root cause positioning method operation and maintenance service and storage medium

The specific scheme is as follows:

(1) acquiring operation and maintenance data of a service system; (2) constructing a first, a second and a third dependency relationship among the target data characteristics; (3) analyzing to obtain a target relation; (4) performing root cause inference on the candidate abnormal type in the target relation to obtain the abnormal probability of the abnormal type; (5) and determining the root cause type according to the abnormal probability.

The scheme has the advantages that: the abnormal probability of the abnormal type is given through the probability graph model, and the method has strong explanatory property; the scheme has the following defects: (1) the method mainly aims at operation and maintenance data, and the production processing data and the operation and maintenance data in industrial manufacturing have larger difference; (2) when the sample data volume is far less than the parameter dimension, the probability graph model is difficult to construct.

Prior art 3:

the scheme is from: method and equipment for analyzing defects of CN104123298B product

The specific scheme is as follows:

(1) screening and generating a data set consisting of defective products from the recorded product data; (2) determining association rules among different data attributes in the defective product data set based on an association analysis algorithm or a statistical analysis algorithm; (3) and (3) carrying out data screening on the defective product data set according to the association rule among the data attributes in the step (2) to obtain the data set where the defect root is located.

The scheme has the advantages that: through the association rule, the data range of the defect root cause can be quickly reduced under the condition of larger product information amount, so that the defect root cause of the product can be quickly positioned. The scheme has the following defects: the extraction of the association rule is relatively difficult, and particularly for the field of industrial product defect root cause detection, the process flow is complex, the related parameters are numerous, the influence relationship among the parameters is complex, and the accurate association rule is difficult to extract effectively.

Disclosure of Invention

To solve the above problems, the present invention provides a product analysis method, system, device and medium.

To achieve the above object, the present invention provides a product analysis method, comprising:

obtaining quality detection data of a product, and labeling the quality detection data to obtain labeled data;

the method comprises the steps of obtaining original processing data of a product, cleaning the original processing data to obtain first process flow data, and preprocessing the first process flow data to obtain second process flow data;

training the first classification prediction model based on the labeling data and the second process flow data to obtain a second classification prediction model;

obtaining data to be analyzed, inputting the data to be analyzed into the second classification prediction model, and outputting the predicted value of each production process parameter in the data to be analyzed by the second classification prediction model;

calculating to obtain a SHAP value of each production process parameter based on the predicted value of each production process parameter, and obtaining an average SHAP value of each production process parameter in the data to be analyzed based on the absolute value of the SHAP value of each production process parameter;

and sequencing the average SHAP value of each production process parameter in the data to be analyzed to obtain a first sequencing result, and obtaining a plurality of key process link names and a plurality of key production process parameters which influence the product quality based on the first sequencing result.

Compared with the prior art 1, the method does not depend on the business experience criterion, does not set the threshold value of characteristic screening, has stronger flexibility and expansibility, and can quickly and accurately obtain the plurality of key process link names and the plurality of key production process parameters which influence the product quality. Compared with the prior art 2, the method provided by the invention is used for analyzing a specific scene of bad root cause positioning in the industrial manufacturing process, and the processing method provided by the invention has stronger pertinence and effectiveness to the characteristics of small sample amount and high parameter dimension in production and processing data, and does not need a large amount of samples. Compared with the prior art 3, the method does not depend on any association rule, and has stronger flexibility and expansibility.

Preferably, the method further comprises the steps of: for defective products in the data to be analyzed, acquiring a SHAP value of each key production process parameter corresponding to the defective products, sequencing the SHAP values of each key production process parameter corresponding to the defective products to acquire a second sequencing result, and acquiring a first root cause production process parameter causing the largest adverse contribution of the products based on the second sequencing result;

and for the normal products in the data to be analyzed, acquiring the SHAP value of each key production process parameter corresponding to the normal products, sequencing the SHAP values of each key production process parameter corresponding to the normal products to acquire a third sequencing result, and acquiring a second factor production process parameter which has the largest contribution to the product yield based on the third sequencing result.

The method comprises the steps that the SHAP value of each key production process parameter corresponding to a defective product is a negative number, and if the SHAP value of the negative number is directly sorted from large to small in the sorting process to obtain a second sorting result, a plurality of results which are sorted later in the second sorting result are taken as a first factor production process parameter which causes the greatest adverse contribution of the product; if the absolute values are used for sorting from large to small, a plurality of results in the top of the sorting are taken as the first cause production process parameters which cause the greatest adverse contribution of the product. And (3) sequencing the SHAP values of the normal products from large to small according to the SHAP value of each key production process parameter corresponding to the normal products, obtaining a third sequencing result, and taking a plurality of top results in the third sequencing result as a second factor production process parameter which contributes most to the product yield.

After obtaining a plurality of key process link names and a plurality of key production process parameters which affect the product quality, the method needs to judge whether the process links and the process parameters have positive or negative influence on the product quality, and positive or negative SHAP values reflect whether the influence of characteristics on model prediction output is positive promotion or negative weakening.

Preferably, the method further comprises the steps of: and analyzing the influence of different values of the first root cause production process parameter and/or the second root cause production process parameter on the product quality according to the corresponding yield of the product and the SHAP value of the production process parameter when the first root cause production process parameter and/or the second root cause production process parameter have different values.

After the root cause parameters causing the defects of the products to be poor are obtained, analysis and optimization are needed to be carried out according to the values of the poor root cause parameters. The invention can further analyze the influence of different values of the parameters on the product quality for a certain specific process parameter, thereby providing the direction of parameter optimization and assisting parameter tuning.

Preferably, the method divides the quality grade of the product according to the quality detection data of the product to obtain the marking data. And labeling the data to facilitate the model processing.

Preferably, the method for cleaning the original processing data specifically includes:

analyzing the repeated times of the product processing records of the same product in the original processing data, and eliminating the repeated product processing records;

sorting the product processing records according to product numbers and time, and integrating the production process parameters of the same product in the product processing records of different time to form the complete production process parameters of the product;

eliminating the production process parameters of which the product coverage rate exceeds a first preset range in the complete production process parameters;

eliminating the production process parameters with fixed values in the complete production process parameters;

and deleting the product processing record of which the coverage rate of the process parameters exceeds the second preset range in the product processing record.

The purpose of the cleaning of the data is to ensure the accuracy of the subsequent analysis steps, among other things.

Preferably, the preprocessing the first process flow data specifically includes:

calculating a correlation coefficient between any two numerical production process parameters for the numerical production process parameters in the first process flow data, and deleting one of the two numerical production process parameters for which the correlation coefficient exceeds a first threshold value;

and selecting production process parameters for the type in the first process flow data, and coding the type selection production process parameters to generate type selection variables.

Preferably, the correlation coefficient between any two numerical production process parameters is calculated using a pearson correlation coefficient.

Preferably, the type selection production process parameters are subjected to One-Hot coding. Different from numerical process parameters, most of the type selection parameters are number information and cannot be directly used as parameter characteristics analyzed in subsequent steps, so that a single type selection parameter needs to be converted into a plurality of mutually independent type selection variables.

The preprocessing is carried out according to the characteristics of the process flow data, and the feasibility and the accuracy of large-scale process parameter data analysis can be guaranteed.

Preferably, the first classification prediction model is a machine learning model.

Preferably, the calculation method of the SHAP value of the production process parameter is as follows:

wherein,

indicating production process parameters in a sample

The SHAP value of (1); | A Represents a factorial symbol;

is the set of all the production process parameters in the sample, the number of the production process parameters in the sample is

；

Representing the removal of the production process parameters from all the production process parameters of the sample

A post-formed set;

represents from

Wherein the extracted part of the production process parameters form an arbitrary subset with the size of

；

Representing a set of utilizations

The output value of the second classification prediction model corresponding to the production process parameter in (1);

representing simultaneous utilization collections

Production process parameters and production process parameters in

The output value of the second classification prediction model.

Preferably, the first and second liquid crystal materials are,

the following constraints need to be satisfied:

wherein,

representing a sample

The predicted values in the second classification prediction model,

a baseline value representing the second classification prediction model,

representing a sample

All of

The inclusion of individual production process parameters.

Preferably, the influence of different values of the first factor production process parameter and/or the second factor production process parameter on the product quality is analyzed by using a partial dependency graph or a scatter diagram.

The relation between the value of the specific process parameter and the reject ratio of the product can be obtained through a scatter diagram between the value of the specific parameter and the reject ratio of the product, and a reference direction is provided for parameter optimization. The SHAP value reflects the direction of the influence of the feature on the model output ("enhanced output" or "diminished output") and the magnitude of the degree of influence. Therefore, by analyzing the relation between different values of the process parameters and the SHAP value of the parameter characteristics by using the scatter diagram, the influence of the different values of the parameters on the product quality can be analyzed in an auxiliary way.

The present invention also provides a product analysis system, the system comprising:

the labeling unit is used for obtaining quality detection data of a product and labeling the quality detection data to obtain labeled data;

the cleaning and preprocessing unit is used for acquiring original processing data of a product, cleaning the original processing data to acquire first process flow data, and preprocessing the first process flow data to acquire second process flow data;

the training unit is used for training the first classification prediction model based on the labeling data and the second process flow data to obtain a second classification prediction model;

the model processing unit is used for obtaining data to be analyzed, inputting the data to be analyzed into the second classification prediction model, and outputting the predicted value of each production process parameter in the data to be analyzed by the second classification prediction model;

the calculation unit is used for calculating and obtaining the SHAP value of each production process parameter based on the predicted value of each production process parameter, and obtaining the average SHAP value of each production process parameter in the data to be analyzed based on the absolute value of the SHAP value of each production process parameter;

and the analysis unit is used for sequencing on the basis of the average SHAP value of each production process parameter in the data to be analyzed to obtain a first sequencing result, and obtaining a plurality of key process link names and a plurality of key production process parameters which affect the product quality on the basis of the first sequencing result.

The invention also provides a product analysis device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the product analysis method when executing the computer program.

The invention also provides a computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the product analysis method.

One or more technical schemes provided by the invention at least have the following technical effects or advantages:

the method creatively applies the model interpretation method based on the SHAP value to the problem of product defect root cause positioning in the field of industrial manufacturing, and provides the most important process parameter characteristics influencing the product quality and the corresponding process links in a targeted manner.

Compared with other characteristic importance analysis methods, the method analyzes the positive and negative influences of important characteristics on the product quality based on the positive and negative of the SHAP values of the characteristics, and according to the SHAP values of products with different quality grades, the method positions and obtains the most main process flow links and specific process parameters which possibly cause the product quality defects, thereby realizing the accurate positioning of the bad root causes

And analyzing the reasonable value of the important process parameter according to the relationship between the SHAP value and the parameter value of different process parameter characteristics, thereby assisting the adjustment and optimization of the bad process parameter.

The analysis method for the bad root cause parameters based on the SHAP values is easy to understand and has strong interpretability.

The method does not depend on a service experience criterion, does not set a threshold value for characteristic screening, does not depend on any association rule, has stronger flexibility and expansibility, can quickly and accurately obtain a plurality of key process link names and a plurality of key production process parameters which influence the product quality, analyzes the specific scene of bad root cause positioning in the industrial manufacturing process, has stronger pertinence and effectiveness to the characteristics of less sample amount and high parameter dimension in production and processing data, and does not need a large amount of samples.

Drawings

The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention;

FIG. 1 is a schematic flow diagram of a product analysis method;

FIG. 2 is a scatter plot of the parameter values versus product reject ratios for different glass panel samples;

FIG. 3 is a scatter plot of the parameter values versus SHAP values for the feature across different glass panel samples;

fig. 4 is a schematic composition diagram of a product analysis system.

Detailed Description

In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments of the present invention and features of the embodiments may be combined with each other without conflicting with each other.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described and thus the scope of the present invention is not limited by the specific embodiments disclosed below.

It should be understood that "system", "device", "unit" and/or "module" as used herein is a method for distinguishing different components, elements, parts, portions or assemblies of different levels. However, other words may be substituted by other expressions if they accomplish the same purpose.

As used in this specification and the appended claims, the terms "a," "an," "the," and/or "the" are not intended to be inclusive in the singular, but rather are intended to be inclusive in the plural, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that steps and elements are included which are explicitly identified, that the steps and elements do not form an exclusive list, and that a method or apparatus may include other steps or elements.

Flow charts are used in this description to illustrate operations performed by a system according to embodiments of the present description. It should be understood that the preceding or following operations are not necessarily performed in the exact order in which they are performed. Rather, the various steps may be processed in reverse order or simultaneously. Meanwhile, other operations may be added to the processes, or a certain step or several steps of operations may be removed from the processes.

Example one

Referring to fig. 1, fig. 1 is a schematic flow chart of a product analysis method, a first embodiment of the present invention provides a product analysis method, including:

and sequencing based on the average SHAP value of each production process parameter in the data to be analyzed to obtain a first sequencing result, and obtaining a plurality of key process link names and a plurality of key production process parameters which influence the product quality based on the first sequencing result.

The product in the method may be a respective product in a craft product, such as a panel, a touch screen, a liquid crystal display, and the like, and the invention is not limited to a specific product, and the product is a panel for example.

For ease of understanding, the present embodiment expresses the production process parameter-related description as a feature description related to machine learning.

The method aims to analyze and position the most main technological parameters (namely, the bad root cause parameters) causing the product to be bad in the product processing and manufacturing process; and analyzing the influence of the value of the specific process parameter on the product quality to assist in the tuning of the process parameter.

The embodiment of the invention provides a root cause positioning method for a defective product in an industrial manufacturing process, and aims to solve the technical problems that in the industrial manufacturing process, a plurality of process parameters are provided, part of parameters are highly related, the product process flow is complex, and the defective root cause of the industrial product is difficult to position.

Where SHAP is an additive interpretation model inspired by Shapley value. Shapley value originates from cooperative game theory.

In order to achieve the above object, the present invention provides a method for detecting and positioning process defect root cause based on a SHAP value, wherein the method comprises:

step 1: respectively labeling and cleaning quality detection data and original processing data of a product to generate labeled data and corresponding complete process flow data;

the product processing data includes, but is not limited to: the production control equipment monitors process state parameters (such as temperature, humidity, pressure and the like in the production equipment) in each process link in the product processing process, intermediate state parameters (such as thickness and surface cleanliness of glass in the panel production intermediate process, quality of pins in the electronic component processing process and the like) in the product monitoring process, and material consumption and time consumption data (such as proportion of different materials, processing time in a certain link and the like) in the processing process.

It should be noted that, because of the existence of a large number of parallel production lines, parallel chambers, etc. in the industrial manufacturing process, even for the same type of product, the production history data and the recorded process parameter types in the manufacturing process may be different. In addition, in the automatic control process, different production monitoring devices record process parameters in different ways. These factors lead to the following conditions that are likely to occur in the raw product processing data: (1) in the process of exporting data by a production monitoring device, due to negligence of related personnel, in an original processing data set, a small number of products with unique numbers can exist a plurality of identical repeated processing records (the repeated records are not deleted during data integration); (2) because of going through a plurality of links in the serial processing process, especially the same process operation is executed by a plurality of processes, in the original processing data, a large number of products with unique numbers may have a plurality of different processing records (the processing data of different links in the serial processing process is not integrated);

(1) dividing the quality grade of the product according to the quality detection data of the product to obtain marking data and corresponding original processing data of the product;

(2) analyzing the repeated times of the same product processing record in the original processing data, and rejecting the completely same repeated processing record;

(3) sequencing the processing records of all products according to product numbers and time, and splicing and integrating parameters in the processing records of the same product at different times to form complete processing parameters of the product;

(4) eliminating technological parameters with serious numerical value loss;

(5) eliminating technological parameters with fixed values;

(6) deleting product processing records with serious numerical value loss;

the present embodiment describes a data cleaning method for the situation that the original processing data is relatively easy to appear and has a large influence on the subsequent analysis step, so as to ensure the accuracy in the subsequent analysis step. In addition, some extra information (such as station numbers in panel production process) may be provided in some product processing data in a specific field, which may assist in the screening of abnormal data and data cleaning, and the present invention is not limited thereto.

Step 2: preprocessing the complete process flow data obtained after cleaning;

before data analysis, pre-processing is necessary. It should be noted that, the production and processing data in the industrial manufacturing field often have the following outstanding characteristics: (1) the number of process parameters is large, and the parameter characteristic number is far beyond the sample number; (2) the process parameters in the production and processing data include various types of processing state information (such as pressure and temperature) collected during the processing process, and some subsequent processing data (such as average pressure and temperature variance) aiming at the processing state information. Meanwhile, when the same operation is repeated for a plurality of times, the same operation is recorded as different process parameters in each machining process. These factors result in a very high correlation between process parameters; (3) the technological process is complex, and besides the serial processing process, a parallel processing process also exists in the product processing process. The parallel processing process means that when the same process is completed and the same processing function is realized, one product can be selected to pass through a plurality of parallel production equipment such as parallel production lines, parallel chambers and the like. Therefore, the information of the parallel equipment used by the product in the processing process can also be taken as a parameter and recorded in the production processing data of the product, namely, the parameter is selected for the type of the parallel process (the parameter value is the line of the selected parallel production line, the number of the parallel chamber and the like).

Therefore, the feasibility and the accuracy of large-scale process parameter data analysis can be ensured only by preprocessing the characteristics of the process flow data, and the specific preprocessing comprises the following steps:

(1) the highly cross-correlated process parameters are deleted and recorded.

The highly cross-correlated process parameters will cause severe collinearity among the features in the model and will also severely restrict the running speed of the model.

Therefore, the correlation between all process parameters is first calculated. And only one of the two process parameters with the correlation exceeding the preset value is reserved, and the deleted process parameter is recorded.

It should be noted that the process parameters can be basically divided into numerical process parameters (such as temperature, humidity, pressure, surface cleanliness, average humidity, maximum pressure, temperature variance, etc.) and type selection parameters (such as the line of the parallel production line selected by the link or the process, the number of parallel chambers, etc.), and the cross-correlation in this step is only for the numerical process parameters.

Preferably, the present invention uses Pearson's correlation coefficient to calculate the correlation of numerical process parameters.

(2) A plurality of independent type selection variables are used to replace the original single type selection parameter.

Different from numerical process parameters, most of the type selection parameters are number information and cannot be directly used as parameter characteristics analyzed in subsequent steps, so that a single type selection parameter needs to be converted into a plurality of mutually independent type selection variables.

In this embodiment, the present invention uses an One-Hot coding method to code the type selection parameter and generate the type selection variable, and other coding methods may also be used in practical applications.

Of course, the above steps are only the characteristics of the process parameter data in the industrial manufacturing process, some necessary preprocessing steps are provided, in the actual processing process, some steps can be reduced or other preprocessing steps can be added according to the condition according to the quality and the characteristics of the data, and the invention is not limited.

And step 3: calculating SHAP values of all parameter characteristics according to the predicted values of the classification prediction models;

(1) selecting XGboost and other machine learning models as classification prediction models, and training and testing input marking data and process flow data until the performance index of the models reaches a preset value;

it should be noted that, for the model input data to be analyzed, the better the prediction effect of the prediction model is, which indicates that the "fitting" of the prediction model to the data to be analyzed is better, and in the subsequent steps, the more accurate the calculation of the SHAP value of the process parameter feature in the data to be analyzed by using the prediction value of the model is.

Therefore, for the data to be analyzed, if the performance index of the model is lower than the preset value, the prediction model needs to be retrained until the performance index of the model reaches the preset value for the data to be analyzed.

(2) Calculating SHAP values of different parameter characteristics in each product sample by adopting the following formula:

wherein,

representing features in a sample

The value of the SHAP of (1),

is the set of all features in the sample with dimensions of

(i.e., the number of features is

），

Representing the removed features of all features of the sample

The set of the latter is composed of,

represents from

An arbitrary subset of the extracted features having a size of

。

Representing only utilization sets

The output value of the feature prediction model in (1).

Representing simultaneous utilization collections

Characteristic of (1) and characteristic

The output value of the prediction model, obviously

Embodying features in any of certain feature extraction modes

Marginal benefit to the prediction model output. The formula gives the characteristics in all possible characteristic extraction modes as a whole

The expectation of marginal benefit of the output of the prediction model embodies the characteristics

"degree of contribution" to the output of the prediction model.

Note that SHAP values

The following equation is satisfied:

wherein,

representing a sample

The value of the prediction in the prediction model,

a baseline value representing the model (i.e. the expectation of the model output for all samples),

representing a sample

All of

The inclusion of a feature (i.e., whether a sample feature participates in the modeling of the prediction model), if a sample

If there is no feature, the value of the SHAP of the feature is 0, i.e., the feature does not contribute to the predicted value of the output.

By this formula, the characteristics can be seen

SHAP value of

Is characterized in the sample

Upper, characteristic of

For the contribution degree of the output value of the prediction model, obviously, the value is greater than 0, and the parameter feature is represented to promote the sample

The feature has positive promoting effect on the predicted value output by the model; conversely, the parameter characteristic is shown to reduce the predicted value, namely, the negative correlation, and the characteristic has an inverse attenuation effect on the predicted value output by the model.

Therefore, the SHAP value can represent the influence degree of the characteristic on the model predicted value, and the SHAP value obtained through calculation through the processes is the contribution score of each production process parameter to the product processing quality in each product production and processing process, and is beneficial to screening important characteristics.

And 4, step 4: determining the most important process links and process parameters influencing the product quality according to the sequencing of the SHAP values of the parameter characteristics;

the absolute value of the SHAP value of the parameter characteristic reflects the contribution of the characteristic to the model output, and the importance degree of the characteristic is reflected. By sorting the absolute values of the SHAP values, the most important features affecting the product quality can be determined.

(1) Calculating the average of the absolute values of SHAP values of all the parameter characteristics as the average SHAP value of all the parameter characteristics;

(2) sorting the average SHAP values of the parameter characteristics, and selecting the parameter characteristics arranged at the front preset position as the most important parameter characteristics;

(3) determining the most important process links and process parameters influencing the product quality according to the process links to which the screened important parameter features belong;

and 5: determining the most main process flow links and specific process parameters which possibly cause product quality defects according to SHAP values of various parameter characteristics in different products;

the positive and negative of the characteristic SHAP value reflects that the influence of the characteristic on the model prediction output is forward promotion or reverse weakening, so that the method can further analyze on the basis of obtaining the important parameter characteristic in the step 4, and for each specific product, how the important parameter characteristic influences the product quality.

(1) Screening out parameter characteristics with SHAP values sequenced to exceed a preset bit as a root cause possibly causing product quality defects aiming at defective products;

(2) for normal products, screening out parameter characteristics with SHAP values sequenced beyond the former preset bits as process parameters with better values and great help for improving the product yield in the production and processing data, so that the process parameters can be compared with bad root cause parameters to assist in optimizing the process parameters;

step 6: analyzing the influence of different values of the specific process parameters on the product quality according to the reject ratio and the SHAP value of the specific process parameters with different values, and assisting in adjusting and optimizing the process parameters;

after the root cause parameters causing the defects of the products are obtained, a factory needs to analyze and adjust the values of the bad root cause parameters. The invention can further analyze the influence of different values of the parameters on the product quality for a certain specific process parameter, thereby providing the direction of parameter optimization and assisting parameter tuning.

In the embodiment of the present invention, because the process parameters are numerous, in step 6, the important parameter characteristics given in step 4 and step 5, which have a large influence on the product quality, are mainly analyzed.

(1) Analyzing the relation between the value of the specific process parameter characteristic and the product reject ratio by using tools such as a partial dependency graph, a scatter diagram and the like;

through a scatter diagram between the specific parameter value and the product reject ratio, the relation between the specific process parameter value and the product reject ratio can be obtained, and a reference direction is provided for parameter tuning.

(2) Analyzing the relation between the value of the specific process parameter and the SHAP value of the parameter characteristic by using tools such as a partial dependency graph, a scatter diagram and the like;

the SHAP value reflects the direction of the influence of the feature on the model output ("enhanced output" or "diminished output") and the magnitude of the degree of influence. Therefore, by analyzing the relation between different values of the process parameters and the SHAP value of the parameter characteristics by using the scatter diagram, the influence of the different values of the parameters on the product quality can be analyzed in an auxiliary way.

According to the technical scheme, important data cleaning and preprocessing can be performed on the process processing data in industrial manufacturing, the data quality and the feasibility and the reliability of large-scale process processing parameter data analysis are effectively improved, and the most important process parameter characteristics influencing the product quality and the corresponding process links are provided in a targeted manner through the SHAP value. Furthermore, according to the positive and negative properties of the SHAP value and the SHAP values of products with different quality grades, the most main process flow links possibly causing the product quality defects and specific process parameters are obtained through positioning, and accurate positioning of the bad root causes is achieved. The invention can also analyze the reasonable value of the important process parameter through the relationship between different process parameter characteristic SHAP values and parameter values, thereby assisting the adjustment and optimization of the bad process parameter. The analysis method for the bad root cause parameters based on the SHAP values is easy to understand and has strong interpretability.

Example two

On the basis of the first embodiment, in order to more clearly demonstrate the objects, technical solutions and advantages of the present invention, the following takes the problem of poor root cause positioning of defective products in the production and manufacturing of glass panels as an example, and the present invention is explained in detail with reference to the accompanying drawings and specific embodiments.

It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As shown in fig. 1, which is a flowchart of a preferred embodiment of the method for locating the root cause of the process defect according to the present invention, the sequence of some steps in the flowchart can be adjusted and some steps can be omitted according to different requirements.

During the production of glass panels, fluctuations in the values of parameters set by the processing equipment can lead to the production of poor glass. The processing parameters of the glass panel have 2 outstanding characteristics: (1) there is a high correlation between the processing parameters. Because a production process flow is often divided into a plurality of different process sections which are closely connected, the abnormal processing of one process section often affects subsequent process sections, for example, after a certain glass panel has abnormal processing on etching equipment, the surface of a product can have flaws, the glass can consume more time on photoetching equipment subsequently, and thus, the processing parameters recorded on the etching equipment and the photoetching equipment can be abnormal. (2) The number of processing parameters far exceeds the number of bad samples. The parameters of the equipment related to the whole process of the glass panel are as much as more than 5 ten thousand, and obviously, the magnitude order of the parameter characteristics far exceeds the magnitude order of a poor sample;

in the traditional regression analysis method, based on the assumption that the number of bad samples is enough and the processing states of different processing devices are relatively independent, the effective positioning of the bad root cause in the scene can not be met obviously. The XGboost classification model is used for 'fitting' product data, and bad root causes which may cause product defects can be effectively positioned on the basis of the SHAP value of the process parameter characteristics.

The following is an analysis of the actual data of a certain type of glass panel with a certain defect in production, and the detailed steps of the method are explained:

step S10: respectively labeling and cleaning quality detection data and original processing data of a product to generate labeled data and corresponding complete process flow data;

the defect detection data of the glass panel includes a product number, a type of the defect, a severity of the defect (e.g., information on the number of defects generated or a defect rate per glass panel), and the like. According to the severity of the defect, the quality grades of the products are divided according to the business rules of the product quality control of different manufacturers, such as the grades are divided into 'excellent, good, defective and the like', in the embodiment, classification and labeling are carried out according to 'qualified products' and 'defective products', the 'defective products' are defined as positive samples, the 'qualified products' are defined as negative samples, and labeling data are obtained.

Meanwhile, only the processing records of the products with the quality detection data are screened from the original processing data of the products.

Obviously, in other embodiments, the product quality grades may be classified into other multiple grade categories, and the positive and negative samples may be redefined, which is not limited by the present invention.

the number of repetitions of the processing record was analyzed based on the product number of the glass panel. If the identical processing records (the processing time, the process parameters and the like are the same) appear on a certain glass panel, the redundant repeated records are removed. If all products have the same repetition number in a certain process stage, it indicates that all products in the process link have undergone multiple links (or procedures) of the serial processing process, and the repeated data belong to processing data of different links (procedures) of the serial processing process and need to be integrated in step (3).

Preferably, the number of repetitions of the processing record may be analyzed using a repetition value analysis function of a pivot table or a data analysis tool (e.g., SPSS, SAS);

(3) and simultaneously sequencing the processing records of all the products according to the product numbers and time, splicing and integrating the parameters in all the processing records of the same product at different times to obtain the complete processing parameters of the product, and forming a wide table with the product numbers of the glass panels as indexes and the parameter names as attribute values. In particular, the same process parameters may exist in the processing record data of different stages and different sites in the processing process, and must be distinguished when the parameters are integrated, for example, the parameters may be distinguished in the modes of "ABLine 1" and "ABLine 2", or site numbers such as "12100 _ ABLine" are added to the parameter names; preferably, in order to facilitate later positioning of the process flow stage to which the process parameter belongs according to the process parameter name, a process flow stage name, such as "a 136995_ PHOTO _1H400_ CLN", may be added to the parameter name, that is, the process parameter with the number "a 136995" in the PHOTO process with the sub-unit "CLN" at the site "1H 400";

(4) eliminating technological parameters with serious numerical value loss;

counting the number of records with each technological parameter value not being empty one by one aiming at all the technological parameters, and dividing the number by the total number of records to obtain the product coverage rate of the technological parameters; and if the product coverage rate of a certain parameter is lower than a preset value, rejecting the process parameter.

Preferably, the product coverage rate of the process parameters can be set to be between 0.8 and 0.9.

(5) Eliminating technological parameters with fixed values;

if the values of a certain process parameter for all the glass panels are the same, the process parameter is not different among different glass panels, and the process parameter is removed.

(6) Deleting product processing records with serious numerical value loss;

counting the number of the technological parameters with non-empty values in each glass panel processing record one by one according to all the processing records, and dividing the number by the total number of the technological parameters in the data set to obtain the coverage rate of the technological parameters in the processing record; and if the coverage rate of the process parameters in the product processing record is lower than a preset value, rejecting the product processing record to form complete process flow data of the product.

Preferably, the coverage rate of the process parameters in the product processing record can be preset to be 0.7-0.8;

it should be noted that the order between step (4) and step (6) may be interchanged. Preferably, because the number of parameters in the industrial manufacturing field is large and often far exceeds the number of samples in the data set (i.e. the number of records of product processing data), the product processing data should be kept as much as possible, the process parameter characteristics with serious numerical value loss are deleted first, and then the corresponding product processing data are deleted, i.e. the step (4) is executed first and then the step (6) is executed.

Step S20: and (4) preprocessing the complete process flow data obtained after cleaning.

(1) The highly cross-correlated process parameters are deleted and recorded.

Calculating correlation coefficients between each numerical type process parameter and all other numerical type process parameters one by one aiming at all numerical type process parameters; and if the correlation coefficient of any two numerical process parameters exceeds a preset value, selecting one of the two numerical process parameters to delete, and correspondingly recording the deleted process parameters into a 'variable list highly cross-correlated with the reserved process parameters'.

In the production and processing data of a certain glass panel, the number of the parameters with the correlation degree of more than 0.9 in the original 2.6-thousand process parameters is up to 1.9-thousand, the 'collinearity' among the parameters can be obviously reduced by deleting the process parameters with the high cross correlation, the dimension can be rapidly reduced, and the speed of the subsequent model analysis and calculation is improved.

Preferably, the correlation coefficient between the two numerical process parameters is calculated using a pearson correlation coefficient.

(2) Using multiple independent type selection variables instead of original single type selection parameter

Since there are parallel production lines, machines, chambers, and the like in the production history of the glass panel, there are a large number of type selection parameters (e.g., the line type of the parallel production line, the number of the chamber, the number of the production unit sub-module, and the like) in the processing data, and these pieces of information cannot be directly used as features for subsequent analysis, and need to be replaced with a plurality of independent type selection variables. Preferably, One-Hot encoding is adopted to generate a type variable, wherein a parameter value is 0 or 1, that is, whether the current type is selected (whether the parallel production line or the parallel chamber is selected, etc.) can be represented.

For example: a certain process parameter in the original processing data is 'Line', the process parameter is recorded in a parallel production Line selected in the current process stage product processing, the value of the process parameter is 'AB' or 'CD', the character type value cannot be directly input into a subsequent analysis model, and two type selection variables 'ABline' and 'CDline' need to be generated to replace the process parameter 'Line'. If the original parameter "Line" takes a value of "AB", the values of the generated type selection variables "ABLine" and "CDLine" are 1 and 0, respectively; on the contrary, if the original parameter "Line" takes a value of "CD", the values of the generated type selection variables "ABLine" and "CDLine" are 0 and 1, respectively.

Step S30: and calculating SHAP values of the parameter characteristics according to the predicted values of the classification prediction model.

and (3) taking the labeling data and the process flow data obtained by the processing of the steps S10 and S20 as input data of a classification prediction model, dividing the input data into a training set, a verification set and a test set, taking production processing parameters in the process flow data as input characteristics of the model, taking the prediction of the product quality grade as an output result of the model, taking the quality grade label of the product in the labeling data as a true value, and realizing the prediction of the product quality grade classification by using the prediction model, wherein the prediction effect of the model is a performance index when the model is used for classification prediction. And continuously adjusting parameters of the optimized prediction model, and obtaining the prediction model for product quality classification after the performance indexes of the model reach preset values on the test set.

The performance index of the model and the preset value can be configured by self-defining, for example, the performance index of the model is set as the accuracy of classification or an AUC value, and the preset value is set as 0.8.

The predictive models may include, but are not limited to: an XGboost model, a random forest model, a LightGBM model, a neural network model, an SVM model, a logistic regression model, and the like. Preferably, compared with a regression model, the prediction model selection tree model (such as XGBOOST, random forest and the like) has a better effect on the process parameter data in the industrial manufacturing field;

in the present embodiment, the predicted classification of the product quality grade ("defective product" or "normal product") is used as the output result of the model (the defective product is regarded as a positive sample, and the normal product is regarded as a negative sample). Of course, in other embodiments, other quality measurement information of the product may be set as the output of the predictive model, including but not limited to: the reject ratio of the product, the number of defective products, the positions of the defects of the product and the like.

(2) SHAP values of different parameter characteristics in each product sample are obtained through calculation

Since the classification prediction model XGBoost adopted in this embodiment is a tree model, a TreeShap method provided by the shield library is directly used, and the shield value of each feature parameter in each sample is obtained by performing fast calculation based on the product quality grade classification prediction model obtained by training.

For example: if 5000 samples are input into the glass panel process flow data of the model and each sample has 6000 parameter characteristics, the SHAP value needs to be calculated once for each characteristic of each glass panel sample to obtain 30000000 SHAP values.

Step S40: determining the most important process links and process parameters influencing the product quality according to the sequencing of the SHAP values of the parameter characteristics;

(1) calculating the average of the absolute values of the SHAP values of all the parameter characteristics on all the product samples according to all the parameter characteristics to obtain the average SHAP value (absolute value) of each parameter characteristic;

for example: the glass panel process parameter data set has 6000 parameter characteristics, and 5000 samples in the data set correspond to 5000 different SHAP values of each corresponding parameter. For each parameter characteristic, taking the absolute value of each SHAP value of the parameter characteristic on 5000 samples, and then calculating the average of the SHAP absolute values of the 5000 samples to obtain the average SHAP value (absolute value) of the parameter characteristic;

(2) sorting the average SHAP value (absolute value) of each parameter characteristic from high to low aiming at all parameter characteristics;

for example, for the processing data of the glass panel blue picture mura defect, the average snap value (absolute value) is ranked from high to low (top 20 features).

(3) Selecting the parameter characteristics of the previous preset positions as the most important parameter characteristics influencing the product quality;

the preset bits can be configured by self-definition, such as top 20, top 50, and the like.

For example, on the same data set, according to the sorting result of the average SHAP value (absolute value) in step (2), the top 10-ranked parameter features are obtained, and then comparison is performed, the dimension of comparison may be the parameter features, and the average SHAP value (absolute value) of each parameter feature is taken as the importance of each parameter feature.

According to the determined most important parameter characteristics, the name of the inspection parameter and the process flow are compared, and the corresponding process flow links are obtained through positioning, so that the process links and the process parameters which influence the product quality most importantly are determined.

Particularly, because a large number of parameter characteristics with high cross correlation are deleted in the preprocessing stage, in order to avoid missing real bad root causes, parameters with high cross correlation with the screened process parameters need to be listed in all subsequent root cause positioning analysis links.

For example, on the same data set, for the blue picture color spot defect of the glass panel, according to the important parameter characteristics screened in the step (3), by locating the process link where the important parameter characteristics are located, the names of the first 20 most important process parameters which have an influence on the product quality, the located process segment, the affiliated production subunit and the like can be obtained, and other parameters highly related to the process segment can be obtained, so that the related process parameters can be accurately and comprehensively located in the complex process flow.

Of course, in other embodiments, on the basis of the feature SHAP value, the feature importance given by the prediction model itself or the feature importance calculated by the method such as permatation may be combined, and the feature importance values obtained by different methods are comprehensively considered to give a comprehensive ranking, which is not limited by the present invention.

Step S50: and analyzing the influence of different characteristics on the product quality according to the SHAP value of each parameter characteristic in different products.

(1) And sorting the SHAP values of the parameter characteristics in all the defective product samples (positive samples) from high to low according to the quality grade labels of the products in the labeling data. Selecting the parameter characteristics arranged at the previous preset positions as the characteristics which have the largest positive influence (promote 'bad') on the model predicted value, namely the most main process parameters (bad root causes) causing the product to be bad;

For example, on the same data set, for the blue-frame mura of the glass panel, the top 20 screened parameter features that may cause the mura are sorted from high to low by ranking the SHAP values of the parameter features in all the samples (positive samples) of the "defective products". For example, in the defect sample, the values of the parameter characteristics such as characteristic No. 3889, 3893, 2706, 1849 and the like are not reasonable, so that the defect occurrence is highlighted, and it is likely that the defect occurrence is a cause of the defect, and important attention is required.

(2) Similar to the step (1), sorting SHAP values of all parameter characteristics in all 'normal product' samples (negative samples) from low to high; selecting the parameter characteristics arranged in the previous preset position as the characteristics which have the maximum negative influence (weaken bad) on the model predicted value, namely the process parameter characteristics which cause high product yield, namely obtaining the process parameters which have good values in the production process and greatly help to improve the product yield, so that the parameter can be compared with the bad root cause parameters to assist in optimizing the process parameters; likewise, the preset bits can be configured by self-definition, such as 20 th top, 50 th top, and the like.

For example, on the same data set, for the swap values of each parameter feature in all "normal product" samples (negative samples), the first 20 parameter features with better values and higher yield are obtained by screening according to the low-to-high order.

Obviously, in a normal sample, the values of the parameter characteristics such as No. 3889, No. 1948, No. 5851 and the like are reasonable, which has a positive influence on the quality of the product, and the reasonable values of the parameters are worthy of intensive study.

Particularly, the characteristic 3889 has the greatest influence on both defective products and normal products, and it is very important to embody the reasonable value of the parameter. The influence sequence of the parameter 1948 in normal products is obviously higher than that of defective products, which shows that the parameter has a more prominent effect on improving yield.

It should be noted that in the preferred embodiment of the present invention, the "defective product" is regarded as a positive sample, and thus the positive influence on the model prediction value is "defect and bad" of the "promoted and strengthened" product. In other embodiments, if the "normal product" is regarded as a positive sample, the parameter feature having the largest negative influence on the model prediction value is the bad root cause of the product defect.

Step S60: analyzing the influence of different values of the specific process parameters on the product quality according to the reject ratio and the SHAP value of the specific process parameters with different values, and assisting in adjusting and optimizing the process parameters;

for example: aiming at the parameter 1849A 138130_ WET _1C501 screened from the data set, fig. 2 is a scatter diagram between the parameter value and the reject ratio of the product on different glass panel samples, the abscissa in fig. 2 is the value of the parameter A138130_ WET _1C501, and the ordinate is the reject ratio of the product with the blue picture rejection, so that it can be obviously seen that when the parameter value is in the sections at two ends (such as 8.06-8.12 and 8.27-8.29), the reject ratio of the product with the blue picture rejection is low. When the parameter value is in the middle range section (such as 8.16-8.21), the reject ratio of the product is high. Thereby providing a reference for root cause parameter tuning.

It should be noted that, since the product defect is affected by a plurality of parameters, the relationship between the product defect rate and a specific parameter shown by the scatter diagram is not obvious and can only be used as a reference.

(2) Analyzing the relation between the value of the specific process parameter characteristic and the SHAP value of the parameter characteristic by using tools such as a partial dependency graph, a scatter diagram and the like;

the value of the parameter is very important for the quality of the product, if the value is reasonable, the SHAP value is a large positive value, and the parameter characteristic brings forward promotion effect (namely 'strengthening badness' in the embodiment) to the output; conversely, the SHAP value is a small negative value, which brings about a "negative attenuation effect" (in this embodiment, referred to as "poor attenuation"). Therefore, the influence of the important parameter value on the product quality can be judged by analyzing the SHAP value corresponding to the specific important parameter value.

For example: the parameter "a 138130_ WET _1C 501" screened from the data set is 1849, fig. 3 is a scatter diagram between the value of the parameter and the value of the shield of the characteristic on different glass panel samples, the abscissa in fig. 3 is the value of the parameter "a 138130_ WET _1C 501", and the ordinate is the shield value of the parameter characteristic. It can be seen that when the value of the parameter is small (e.g. less than 8.20), the SHAP value of the parameter is almost negative, and for the bad product, the negative attenuation effect is brought; when the value of the parameter is large (for example, greater than or equal to 8.20), the SHAP value of the parameter is almost a positive value, and the adverse effect of the product is a positive enhancement effect. Through the analysis process of the SHAP value, the adjustment and optimization of the defect root cause parameters can be assisted.

Obviously, for the analysis of the characteristic parameter values, the relationship between the relatively definite parameter values and the product quality is easier to obtain based on the SHAP value than based on the product reject ratio.

In this embodiment, the method for analyzing the influence of different values of the parameter characteristics on the characteristic SHAP value includes, but is not limited to: drawing a scatter diagram or a partial dependency graph, calculating a correlation coefficient, constructing a linear regression model and the like. Preferably, the relationship between the SHAP value and different values of the parameter characteristics can be intuitively and simply analyzed in a scatter diagram or partial dependency diagram mode.

EXAMPLE III

Referring to fig. 4, fig. 4 is a schematic diagram of a product analysis system, where the system includes:

Example four

The fourth embodiment of the present invention provides an apparatus for analyzing user energy consumption behavior, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the method for analyzing user energy consumption behavior when executing the computer program.

EXAMPLE five

An embodiment five of the present invention provides a computer-readable storage medium, where a computer program is stored, and when the computer program is executed by a processor, the steps of the user energy consumption behavior analysis method are implemented.

The processor may be a Central Processing Unit (CPU), or other general-purpose processor, a digital signal processor (digital signal processor), an Application Specific Integrated Circuit (Application Specific Integrated Circuit), an off-the-shelf programmable gate array (field programmable gate array) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory may be used to store the computer programs and/or modules, and the processor may implement various functions of the user-enabled behavior analysis apparatus in the invention by executing or executing data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (such as a sound playing function, an image playing function, etc.), and the like. Further, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, at least one magnetic disk storage device, a flash memory device, or other volatile solid state storage device.

The user energy consumption behavior analysis means, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, all or part of the flow in the method of implementing the embodiments of the present invention may also be stored in a computer readable storage medium through a computer program, and when the computer program is executed by a processor, the computer program may implement the steps of the above-described method embodiments. Wherein the computer program comprises computer program code, an object code form, an executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying said computer program code, a recording medium, a usb-disk, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, a point carrier signal, a telecommunications signal, a software distribution medium, etc. It should be noted that the computer readable medium may contain content that is appropriately increased or decreased as required by legislation and patent practice in the jurisdiction.

While the invention has been described with respect to the basic concepts, it will be apparent to those skilled in the art that the foregoing detailed disclosure is only by way of example and not intended to limit the invention. Various modifications, improvements and adaptations to the present description may occur to those skilled in the art, although not explicitly described herein. Such modifications, improvements and adaptations are proposed in the present specification and thus fall within the spirit and scope of the exemplary embodiments of the present specification.

Also, the description uses specific words to describe embodiments of the description. Reference throughout this specification to "one embodiment," "an embodiment," and/or "some embodiments" means that a particular feature, structure, or characteristic described in connection with at least one embodiment of the specification is included. Therefore, it is emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, some features, structures, or characteristics of one or more embodiments of the specification may be combined as appropriate.

Moreover, those skilled in the art will appreciate that aspects of the present description may be illustrated and described in terms of several patentable categories or situations, including any new and useful combination of processes, machines, manufacture, or materials, or any new and useful modification thereof. Accordingly, aspects of this description may be performed entirely by hardware, entirely by software (including firmware, resident software, micro-code, etc.), or by a combination of hardware and software. The above hardware or software may be referred to as "data block," module, "" engine, "" unit, "" component, "or" system. Furthermore, aspects of the present description may be represented as a computer product, including computer readable program code, embodied in one or more computer readable media.

The computer storage medium may comprise a propagated data signal with the computer program code embodied therewith, for example, on baseband or as part of a carrier wave. The propagated signal may take any of a variety of forms, including electromagnetic, optical, etc., or any suitable combination. A computer storage medium may be any computer-readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code located on a computer storage medium may be propagated over any suitable medium, including radio, cable, fiber optic cable, RF, or the like, or any combination of the preceding.

Computer program code required for the operation of various portions of this specification may be written in any one or more of a variety of programming languages, including an object oriented programming language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C + +, C #, VB.NET, Python, and the like, a conventional programming language such as C, Visual Basic, Fortran 2003, Perl, COBOL 2002, PHP, ABAP, a dynamic programming language such as Python, Ruby, and Groovy, or other programming languages, and the like. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any network format, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet), or in a cloud computing environment, or as a service, such as a software as a service (SaaS).

Additionally, the order in which the elements and sequences of the process are recited in the specification, the use of alphanumeric characters, or other designations, is not intended to limit the order in which the processes and methods of the specification occur, unless otherwise specified in the claims. While various presently contemplated embodiments of the invention have been discussed in the foregoing disclosure by way of example, it is to be understood that such detail is solely for that purpose and that the appended claims are not limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications and equivalent arrangements that are within the spirit and scope of the embodiments herein. For example, although the system components described above may be implemented by hardware devices, they may also be implemented by software-only solutions, such as installing the described system on an existing server or mobile device.

Similarly, it should be noted that in the preceding description of embodiments of the present specification, various features are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure aiding in the understanding of one or more of the embodiments. This method of disclosure, however, is not intended to imply that more features than are expressly recited in a claim. Indeed, the embodiments may be characterized as having less than all of the features of a single embodiment disclosed above.

For each patent, patent application publication, and other material, such as articles, books, specifications, publications, documents, etc., cited in this specification, the entire contents of each are hereby incorporated by reference into this specification. Except where the application history document does not conform to or conflict with the contents of the present specification, it is to be understood that the application history document, as used herein in the present specification or appended claims, is intended to define the broadest scope of the present specification (whether presently or later in the specification) rather than the broadest scope of the present specification. It is to be understood that the descriptions, definitions and/or uses of terms in the accompanying materials of this specification shall control if they are inconsistent or contrary to the descriptions and/or uses of terms in this specification.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims

1. A method of product analysis, the method comprising:

2. The product analysis method of claim 1, further comprising the steps of: for defective products in the data to be analyzed, acquiring a SHAP value of each key production process parameter corresponding to the defective products, sequencing the SHAP values of each key production process parameter corresponding to the defective products to acquire a second sequencing result, and acquiring a first root cause production process parameter causing the largest adverse contribution of the products based on the second sequencing result;

3. The product analysis method of claim 2, further comprising the steps of: and analyzing the influence of different values of the first root cause production process parameter and/or the second root cause production process parameter on the product quality according to the corresponding yield of the product and the SHAP value of the production process parameter when the first root cause production process parameter and/or the second root cause production process parameter have different values.

4. The product analysis method according to claim 1, wherein the labeling data is obtained by classifying quality grades of products based on quality detection data of the products.

5. The product analysis method according to claim 1, wherein the cleaning of the raw process data specifically comprises:

6. The product analysis method of claim 1, wherein preprocessing the first process flow data specifically comprises:

7. The product analysis method of claim 6, wherein the correlation coefficient between any two numerical production process parameters is calculated using a Pearson correlation coefficient.

8. The product analysis method of claim 6, wherein the type-selective production process parameters are One-Hot encoded.

9. The product analysis method of claim 1, wherein the first classification prediction model is a machine learning model.

10. The product analysis method of claim 1, wherein the SHAP value of the production process parameter is calculated by:

wherein,

indicating production process parameters in a sample

The SHAP value of (1); | A Represents a factorial symbol;

；

A post-formed set;

represents from

；

Representing a set of utilizations

representing simultaneous utilization collections

Production process parameters and production process parameters in

The output value of the second classification prediction model.

11. The product analysis method of claim 10,

the following constraints need to be satisfied:

wherein,

representing a sample

The predicted values in the second classification prediction model,

a baseline value representing the second classification prediction model,

representing a sample

All of

The inclusion of individual production process parameters.

12. The product analysis method according to claim 3, wherein the influence of different values of the first root cause production process parameter and/or the second root cause production process parameter on the product quality is analyzed by using a partial dependency graph or a scatter diagram.

13. A product analysis system, characterized in that the system comprises:

14. Product analysis apparatus comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the product analysis method according to any of claims 1 to 12 when executing the computer program.

15. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the product analysis method according to any one of claims 1 to 12.