WO2021146996A1 - 设备指标优良性等级预测模型训练方法、监控系统和方法 - Google Patents

设备指标优良性等级预测模型训练方法、监控系统和方法 Download PDF

Info

Publication number
WO2021146996A1
WO2021146996A1 PCT/CN2020/073779 CN2020073779W WO2021146996A1 WO 2021146996 A1 WO2021146996 A1 WO 2021146996A1 CN 2020073779 W CN2020073779 W CN 2020073779W WO 2021146996 A1 WO2021146996 A1 WO 2021146996A1
Authority
WO
WIPO (PCT)
Prior art keywords
indicators
equipment
target
dimensionality reduction
data
Prior art date
Application number
PCT/CN2020/073779
Other languages
English (en)
French (fr)
Inventor
王海金
吴建民
冯玉
薛静
Original Assignee
京东方科技集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Priority to CN202080000070.0A priority Critical patent/CN113614758A/zh
Priority to PCT/CN2020/073779 priority patent/WO2021146996A1/zh
Publication of WO2021146996A1 publication Critical patent/WO2021146996A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • the embodiments of the present disclosure relate to the field of display panel manufacturing, and in particular, to a training method of a predictive model of the superiority of equipment indicators, a monitoring system and method of equipment indicators, and a computer-readable medium.
  • the display panel needs to be processed by multiple process equipment in sequence during the production process.
  • the final display panel product inevitably has various defects with a certain probability, and the quality of the product is essentially related to the process equipment, specifically caused by the parameter value of the equipment index.
  • the embodiments of the present disclosure provide a method for training a model for predicting the superiority of equipment indicators, a monitoring system and method for equipment indicators, and a computer-readable medium.
  • a training method for a prediction model of equipment index superiority level including:
  • the machine learning model is trained through the above-mentioned positive sample set and the above-mentioned negative sample set to obtain a prediction model for predicting the goodness level of the target equipment index.
  • the above-mentioned performing dimensionality reduction processing on the above-mentioned historical parameter value according to the association relationship between the above-mentioned multiple device indicators includes:
  • the sample data of the above-mentioned dimensionality reduction variable is determined.
  • the above-mentioned training of the machine learning model by the above-mentioned positive sample set and the above-mentioned negative sample set includes:
  • the machine learning model is trained based on the above-mentioned screening variables and the parameter values corresponding to the above-mentioned screening variables.
  • the aforementioned sample data includes training samples and test samples
  • the foregoing determination of the positive sample set and the negative sample set according to the superiority level of the sample data of the above-mentioned dimensionality reduction variable includes:
  • the above method further includes: evaluating the above prediction model through the above test samples.
  • the above prediction model is evaluated by one or more of the following evaluation indicators, the above evaluation indicators include: accuracy rate, recall rate, and area under the receiver operating characteristic curve AUC.
  • the foregoing obtaining the historical parameter value of each of the foregoing device indicators includes:
  • the above method further includes:
  • the above-mentioned machine learning model is a Bayesian algorithm model, a decision tree algorithm model, or a neural network model.
  • a device indicator monitoring system including: a distributed storage device, and one or more processors, wherein,
  • the foregoing distributed storage device is configured to obtain the first parameter values to be tested about M target device indicators, where M is an integer greater than 1;
  • the one or more processors are configured to monitor the superiority of the M target device indicators based on the prediction model.
  • the above-mentioned one or more processors are further specifically configured as:
  • the second parameter value to be measured is input to the prediction model to monitor the excellence of the M target equipment indicators through the output value of the prediction model.
  • the above-mentioned one or more processors are specifically configured as:
  • the related equipment indicators are combined to obtain N target dimensionality reduction variables
  • the above-mentioned one or more processors are specifically configured as:
  • N Fourier basis eigenvectors based on a linear regression algorithm to obtain N'target screening variables with respect to the above-mentioned superiority level, where N'is less than N;
  • the above-mentioned one or more processors are further configured to:
  • the above-mentioned target dimensionality reduction variable is upgraded to obtain a set of faulty equipment indicators.
  • the above-mentioned one or more processors are further configured to:
  • the above-mentioned system further includes: a display device configured to:
  • the above-mentioned system further includes: an alarm device configured to:
  • a method for monitoring equipment indicators including:
  • the prediction model provided in the embodiment of the training method based on the foregoing equipment superiority level prediction model monitors the superiority of the foregoing M target device indicators.
  • the foregoing monitoring of the superiority of the foregoing M target device indicators based on the foregoing prediction model includes:
  • the second parameter value to be measured is input to the prediction model to monitor the excellence of the M target equipment indicators through the output value of the prediction model.
  • the above-mentioned performing dimensionality reduction processing on the above-mentioned first parameter value to be measured according to the association relationship between the above-mentioned M target device indicators includes:
  • the related equipment indicators are combined to obtain N target dimensionality reduction variables
  • a computer-readable medium on which a computer program is stored, wherein, when the program is executed by a processor, the above-mentioned equipment index goodness level prediction model is realized
  • FIG. 1 is a schematic flowchart of a training method for a predictive model of equipment index superiority level provided by an embodiment of the present disclosure
  • FIG. 2 is a schematic flowchart of a method for acquiring historical parameter values of equipment indicators provided by an embodiment of the disclosure
  • FIG. 3 is a schematic flowchart of a dimensionality reduction processing method provided by an embodiment of the disclosure
  • FIG. 4 is a schematic flowchart of a model training method provided by an embodiment of the disclosure.
  • FIG. 5 is a schematic structural diagram of a monitoring system for equipment indicators provided by an embodiment of the disclosure.
  • FIG. 6 is a schematic structural diagram of an analysis device in a device indicator monitoring system provided by an embodiment of the disclosure.
  • FIG. 7 is a schematic structural diagram of another device indicator monitoring system provided by an embodiment of the disclosure.
  • FIG. 8 is a schematic diagram of a method for monitoring device indicators provided by an embodiment of the disclosure.
  • FIG. 9 is a schematic diagram of a computer-readable medium provided by an embodiment of the disclosure.
  • FIG. 10 is a schematic diagram of an electronic device provided by an embodiment of the disclosure.
  • Each production line includes multiple process sites.
  • Each process site is used to perform certain treatments (such as cleaning, deposition, exposure, etching, etc.) on products (including semi-finished products). Boxing, testing, etc.).
  • each process site usually has multiple process equipment for the same treatment; of course, although the theoretical treatment is the same, the actual treatment effect is not exactly the same due to the different types and conditions of different process equipment. .
  • each process equipment will participate in the production process of some products, but not the production process of all products, that is, each process equipment participates in and only participates in the production process of some products.
  • the embodiments of the present disclosure provide a method for training a model for predicting the superiority of equipment indicators.
  • the equipment index goodness grade prediction model trained by the training method provided by the embodiments of the present disclosure can predict the parameter value of the equipment index, and monitor the predicted parameter value in real time through historical parameter trends, thereby predicting the possibility of occurrence of defects; each of the production lines
  • the equipment index goodness level prediction model trained by the training method provided by the embodiments of the present disclosure can also obtain the optimal trend graph of the parameters of each equipment index, so that when similar products are produced, problems can be found as early as possible to ensure In the production of similar products, the equipment status is the optimal production status, which improves the yield of similar products in batches.
  • defects refers to the quality defects in the product, these defects may cause the product quality to be reduced or even scrapped, and may also cause the product to be reworked or repaired.
  • defects can be divided into different types according to needs. For example, it can be classified according to the direct impact of the bad on the product performance, such as bad bright line, bad dark line, hot spot, etc.; or, it can also be classified according to the specific cause of the bad, such as bad short circuit between the grid line and the data line , Poor alignment, etc.; Or, it can be classified according to the general cause of the defect, such as poor array process, poor color film process, etc.; Or, it can also be classified according to the severity of the defect, such as the defect that leads to scrap, resulting in reduced quality Or, it is not necessary to distinguish the types of defects, that is, as long as there is any "bad” in the product, it is considered as "bad”, otherwise, it is considered as "good”.
  • the above correlation is for "one type” of defects, that is, the same process equipment has different correlations to different types of defects.
  • the "correlation between process equipment and a certain type of defect” refers to the degree of influence that the participation of the process equipment has on the probability of the occurrence of such a defect in the product.
  • the product is a display panel production line.
  • the embodiments of the present disclosure can be used in the production process of display panels (such as liquid crystal display panels, organic light emitting diode display panels, etc.) to determine the correlation between each process equipment of the display panel production line and the poor.
  • display panels such as liquid crystal display panels, organic light emitting diode display panels, etc.
  • the embodiments of the present disclosure can also be applied to other products.
  • an embodiment of the present disclosure provides a method for training a predictive model of equipment index superiority level from step S110 to step S140.
  • step S110 historical parameter values of multiple device indicators are acquired.
  • the above-mentioned equipment indicators are related to the quality of the product (eg, OLED), including: temperature, pressure, humidity, heating time, cooling time, etc.
  • the above-mentioned device index may be the index of multiple devices, or may be the index of a certain device.
  • the historical parameter value of each of the aforementioned equipment indicators includes: a preset temperature value in a historical period, a preset pressure value in a historical period, a preset pressure value in a historical period, and the like.
  • a monitoring model is established separately for each equipment indicator to realize the parameter monitoring of the equipment indicator, and the parameter monitoring of each equipment indicator needs to be manually monitored separately.
  • more equipment indicators will lead to more data in the monitoring model.
  • only certain important parameters are managed, which results in a smaller monitoring range and monitoring accuracy that needs to be improved.
  • this technical solution uses Factory Data Collection (FDC) system data as the big data basis for training the above-mentioned equipment index goodness grade prediction model.
  • FDC Factory Data Collection
  • the FDC system is a system for real-time monitoring and analysis of production equipment production process data (PROCESS data), which can reduce scrap rate and production time and improve product quality and output through real-time monitoring and analysis of data.
  • PROCESS data production equipment production process data
  • This solution determines the samples for training the above classification model based on multiple device indicators and their historical parameter values, which can effectively expand the monitoring range and improve the monitoring accuracy.
  • the embodiment shown in FIG. 2 provides a specific implementation manner for obtaining the historical parameter value of each device indicator in step S110, including:
  • Step S210 load the source data about equipment indicators into a data warehouse (e.g., Hive) for data mining analysis; and, step S220, store the data after the mining analysis in a non-relational database (e.g., Hbase) for data mining analysis. Used to determine the historical parameter value.
  • a data warehouse e.g., Hive
  • Hbase non-relational database
  • the source data of the above device indicators comes from the state variable (State Variable, SV for short) data in the FDC system.
  • state variable State Variable, SV for short
  • SV data is used to represent real-time status parameters of equipment indicators (for example, the current temperature is 80 degrees Celsius).
  • the BLOB (Binary Large Object) field in the SV data can be parsed to generate visualization data.
  • visualized data refers to visually conveying data to users by means of graphical means, so as to convey and communicate information more clearly and effectively.
  • ETL Extraction-Transformation-Loading, that is, the process of data extraction (Extract), transformation (Transform), and loading (Load), which is an important part of building a data warehouse
  • the unified SV data (which can include the above-mentioned visualization data) is integrated and imported into a data warehouse (such as HIVE) to provide an analysis basis for further data decision-making.
  • a data warehouse such as HIVE
  • the data in the data warehouse HIVE is stored in a non-relational database (such as Hbase) through ETL, so as to store massive semi-structured data through Hbase, and realize data retrieval and real-time analysis.
  • a non-relational database such as Hbase
  • ETL electronic book reader
  • data cleaning is also performed when the sample set is determined.
  • the indicator of the equipment to be cleaned whose parameter value is a null value in the data object; obtain the mode value or average value of the attribute value of the indicator of the equipment to be cleaned in other data objects; and determine that the mode value or the average value is all The attribute value of the device indicator to be cleaned in the data object.
  • the parameter value is a null value as the indicator of the equipment to be cleaned.
  • the null value is numeric
  • the missing attribute value is filled according to the average value of the attribute in all other objects;
  • the null value is a numeric value Type, according to the principle of mode in statistics, use the value of the attribute with the most value in all other objects to fill in the missing value, so as to fill in the missing parameter value with the highest possible value. In turn, it helps to improve the prediction accuracy of the trained model.
  • step S120 dimensionality reduction processing is performed on the aforementioned historical parameter values according to the association relationship between the multiple device indicators to obtain sample data about dimensionality reduction variables.
  • a monitoring model is established separately for each equipment indicator to realize the parameter monitoring of the equipment indicator, and the parameter monitoring of each equipment indicator needs to be monitored separately. Since a separate monitoring model is established for each device indicator to realize the parameter monitoring of the device indicator, there is a lack of correlation between each device indicator. When the target monitoring device has a problem, it is difficult to obtain other devices related to the target device parameters. Parameter information leads to low monitoring efficiency.
  • the present technical solution performs dimensionality reduction processing on the relevant historical parameter values according to the correlation between the equipment indicators, so as to obtain sample data about the dimensionality reduction variables.
  • FIG. 3 is a schematic flowchart of a dimensionality reduction processing method provided by an embodiment of the disclosure, and may be used as a specific implementation of step S120.
  • the method shown in this embodiment includes:
  • Step S310 Analyze the association relationship between the multiple equipment indicators through a principal component analysis algorithm
  • Step S320 Perform a combination operation on the associated equipment indicators based on the association relationship to obtain a dimensionality reduction variable
  • Step S330 according to The historical parameter values of the equipment indicators involved in the combination operation determine the sample data of the dimensionality reduction variable.
  • the device indicators with the correlation relationship may be regarded as a set (ie, "device indicator set”).
  • a set of equipment indicators can also be called “Recipe”. Specifically, it refers to a set of equipment indicators related to the completion of a certain function of the product during the product manufacturing process; for example, in the OLED manufacturing process, a set of equipment parameters (such as temperature, pressure, and electrical properties) related to the packaging stage.
  • equipment parameters such as temperature, pressure, and electrical properties
  • the first version of the recipe before a certain packaging technology is improved, it can be called the first version of the recipe, and after the packaging technology is improved, it can be called the second version of the recipe.
  • the difference of parameter values in recipes between different versions is compared to provide a reference basis for optimizing recipes.
  • the quantitative statistical results are provided for each equipment index and different process steps in a set of recipes, so the closely related equipment index (variable) is reduced to a comprehensive index (that is, a small number of new variables), and the comprehensive index is obtained Are pairwise unrelated. Therefore, fewer comprehensive indicators can be used to represent the information contained in a larger number of original equipment indicators.
  • PCA Principal Component Analysis Algorithm
  • Dimensionality reduction is to retain the most important features of high-dimensional data, remove unimportant features, and achieve data enhancement
  • the purpose of processing speed, dimensionality reduction is within a certain range of information loss, which can save us a lot of time and cost) method to preprocess the data, map the high-dimensional data to the low-dimensional space, and make the data more dimensional
  • the variance is the largest, so that fewer data dimensions are used, more characteristics of the original data points are retained, and the internal information of the data is maximized.
  • step S130 the positive sample set and the negative sample set are determined according to the superiority level of the sample data of the aforementioned dimensionality reduction variable.
  • the goodness grade of the sample data is determined according to the goodness grade of the product.
  • the quality of the product is divided into A grade, B grade, C grade and D grade.
  • Product-related sample data can be divided into A-level
  • sample data related to the manufacture of B-level products can be divided into B-level
  • sample data related to the manufacture of C-level products can be divided into C-level and related to manufacturing
  • the sample data related to D-level products can be classified as D-level.
  • the superiority levels of training samples belonging to class A and class B in the sample data of the dimensionality reduction variable can be determined as the positive sample set
  • training samples belonging to class C and class D in the sample data of the dimensionality reduction variable The superiority level of is determined as the negative sample set.
  • the final inspection data of the OLED product is used to determine the positive sample set and the negative sample set. Specifically, if the final inspection data of the OLED product is good, the equipment index parameters for producing the OLED product are taken as the positive sample, so as to determine the positive sample set; if the final inspection data of the OLED product is defective, the equipment index parameter of the OLED product is produced As a negative sample, the set of negative samples is determined. It should be noted that: in order to ensure the accuracy of the forecast, the determination of the above positive and negative sample set is performed for a single model of product. In addition, in order to avoid inaccurate predictions caused by uneven sample distribution, for a single model product, the above-mentioned positive sample set contains at least 10,000 pieces of positive sample information, and the above-mentioned negative sample set contains at least 10,000 pieces of negative sample information.
  • step S140 the machine learning model is trained through the above-mentioned positive sample set and the above-mentioned negative sample set to obtain a classification model for predicting the goodness level of the target equipment index.
  • FIG. 4 is a schematic flowchart of a model training method provided by an embodiment of the disclosure, and may be used as a specific implementation of step S140. Referring to FIG. 4, the method shown in this embodiment includes:
  • Step S410 Sampling the dimensionality reduction variables in the positive sample set and the negative sample set respectively to obtain sampling variables; Step S420: Calculate the Fourier basis feature vector of each of the sampling variables to determine The correlation between the sampling variables; step S430, processing the Fourier basis feature vector based on a linear regression algorithm to obtain screening variables; and, step S440, based on the screening variables and the parameters corresponding to the screening variables Value training machine learning model.
  • the above-mentioned positive sample set is randomly divided into training positive samples and test positive samples
  • the above-mentioned negative sample set is randomly divided into training negative samples and test negative samples.
  • the embodiment shown in the figure trains the machine learning model by training positive samples and training negative samples.
  • the above-mentioned machine learning model is a Bayesian algorithm model, a decision tree algorithm model or a neural network classification model.
  • the eigenvector of the low-degree Fourier basis is calculated for the parameter value corresponding to each sampling variable to determine the correlation between the sampling variables. Further, the eigenvectors of the Fourier basis are processed based on a linear regression algorithm (for example, a lasso algorithm) to obtain screening variables. After a round of screening is completed, screening variables are features that have a greater impact on the quality of the product. Based on at least one round of screening, a machine learning model is trained on the screening features and their corresponding parameters.
  • noise variables that do not have a sufficiently large influence on the goodness level of the product can be removed, which is beneficial to improve the model training speed.
  • the technical solution further includes testing the goodness grade prediction model (denoted as the "model to be tested") by testing the positive samples and testing the negative samples, and using at least one test indicator to test the model to be tested. The test results are verified, and the model that meets the test indicators is predicted for goodness.
  • the goodness grade prediction model (denoted as the "model to be tested")
  • the model performance is evaluated through one or more of the following evaluation indicators, the evaluation indicators include: accuracy, recall, KS (Kolmogorov-Smirnov, Kolmogorov-Smirnov) value, and The receiver operating characteristic curve (Receiver Operating Characteristic curve, referred to as ROC) under the area AUC (a model evaluation index, specifically used to evaluate the predictive value of the model; is the abbreviation of Area Under ROC Curve) the prediction model after the above iterative optimization to evaluate.
  • the evaluation indicators include: accuracy, recall, KS (Kolmogorov-Smirnov, Kolmogorov-Smirnov) value, and The receiver operating characteristic curve (Receiver Operating Characteristic curve, referred to as ROC) under the area AUC (a model evaluation index, specifically used to evaluate the predictive value of the model; is the abbreviation of Area Under ROC Curve) the prediction model after the above iterative optimization to evaluate.
  • the specific method for testing the model to be tested includes:
  • TP is the number of positive classes that are still positive after judging the positive class in the test sample set using the to-be-tested model
  • TN uses the to-be-tested model to determine the number of negative classes in the test sample set that are still negative.
  • FN uses the to-be-tested
  • the model judges the number of positive classes in the test sample set
  • FP uses the model to be tested to judge the number of negative classes in the test sample set.
  • the positive class and the negative class refer to the two categories manually labeled for the first part of the sample, that is, if a sample is manually labeled as belonging to a specific class, the sample belongs to the positive class, and the sample that does not belong to the specific class belongs to the negative class.
  • test result of the model to be tested is calculated based on the true positive TP, true negative TN, false negative FN and false positive FP.
  • test index may be an AUC or KS value. specific:
  • formula 1 and formula 2 are used to determine the false positive rate FPR and the true positive rate TPR,
  • the ROC curve is the characteristic curve of each index obtained, which is used to show the relationship between the indexes, and to further calculate the area under the ROC curve AUC.
  • the ROC curve is the characteristic curve of each index obtained, and is used to show the relationship between the indexes.
  • AUC is the area under the ROC curve. The larger the AUC, the higher the predictive value of the model, and the model to be tested can be tested by the AUC. And when the evaluation result is that the AUC value meets the preset threshold, the obtained model can be used to predict the superiority level.
  • KS (every time a different threshold is selected, we can get a set of FPR and TPR, that is, a point on the ROC curve) curve is the difference between the two curves under each threshold.
  • KS max(TPR-FPR), that is, the maximum value of the difference between TPR and FPR; the KS value can reflect the optimal distinguishing effect of the model, and the threshold taken at this time is generally used as the optimal threshold for defining good and bad users.
  • the value range of the KS value is [0, 1], and the larger the KS value, the higher the accuracy of the model's prediction.
  • KS>0.2 means that the model has better prediction accuracy.
  • the model test index may also be an accuracy rate and a recall rate.
  • the accuracy rate p and the recall rate r are calculated according to formula 3 and formula 4 respectively;
  • the accuracy test result is greater than p'(default value)
  • the accuracy setting condition is satisfied, otherwise the accuracy setting condition is not satisfied
  • the recall test result is greater than r'( The default value) is to meet the recall rate setting conditions, otherwise the recall rate setting conditions are not met.
  • the model to be tested when the test result meets the set condition corresponding to the test index, can be used as a predictive model for predicting the parameter value of the equipment index; when the test result does not meet the set condition , The model to be tested continues to iteratively optimize until the test result of the model to be tested meets the set conditions.
  • the embodiments of the present disclosure provide a monitoring system for equipment indicators.
  • the grid computing of RDBMS divides the problem that requires very huge computing power into many small parts, and then distributes these parts to many computers for separate processing, and finally combines these calculation results.
  • Oracle RAC Real Application Cluster
  • Oracle RAC is the core technology of grid computing supported by the Oracle database, in which all servers can directly access all data in the database.
  • the RDBMS grid computing application system cannot meet user requirements when the amount of data is large. For example, due to the limited expansion space of the hardware, when the data is increased to a large enough order of magnitude, the input/output bottleneck of the hard disk will cause The efficiency of processing data is very low.
  • the Hive tool is a Hadoop-based data warehouse tool that can be used to extract and transform data (Extraction-Transformation-Loading, referred to as: ETL).
  • ETL extraction-Transformation-Loading
  • the Hive tool defines a simple SQL-like query language, and also allows the use of custom MapReduce Mapper and reducer come with complex analysis work that cannot be done by default tools.
  • the Hive tool does not have a special data storage format, nor does it create an index for the data. Users can freely organize the tables in it and process the data in the non-relational database. It can be seen that the parallel processing of distributed file management can meet the storage and processing requirements of massive data. Users can process simple data through SQL queries, and use custom functions for complex processing. Therefore, when analyzing the massive data of the factory, it is necessary to extract the data of the factory database into a distributed file system (distributed storage device 501/701 as described below), specifically:
  • the scattered, disorderly, and non-uniform SV data are integrated and imported into the data warehouse HIVE to provide analysis basis for further data decision-making.
  • the massive structured and full data can be stored, calculated, and mined in the data warehouse HIVE.
  • the data in the data warehouse HIVE is stored in the non-relational database Hbase through ETL, so as to store massive semi-structured data through Hbase, and realize data retrieval and real-time data analysis.
  • the device indicator monitoring system 500 provided by the embodiment of the present disclosure includes a distributed storage device 501 and one or more processors 502.
  • the above-mentioned distributed storage device 501 is configured to respectively obtain the first parameter values to be tested about M target device indicators, where M is an integer greater than 1; it stores production data generated by factory equipment.
  • the aforementioned one or more processors 502 are configured to monitor the excellence of the M target device indicators based on the classification model provided in the aforementioned embodiment. Perform operations that determine relevance.
  • the distributed storage device 501 stores production data from factory equipment.
  • factory equipment refers to any equipment in each factory, which can include process equipment in each process site, or management equipment used to manage the production line in the factory;
  • production data refers to any information related to production, Including which products each process equipment in each production line participates in the production process, and whether each product ultimately has defects and what types of defects exist.
  • the above-mentioned one or more processors (such as CPU) 502 may be in the analysis device 600.
  • the analysis device 600 further includes a memory (such as a hard disk) 601 that can store required programs.
  • the processing 502 and the memory 601 are connected through an I/O interface 602, so as to realize information interaction, so that the processor 502 can be based on
  • the program stored in the memory 601 performs required operations to realize the operation of determining the correlation.
  • the distributed storage device 501 stores relatively complete data (such as a non-relational database), and the distributed storage device 501 includes multiple hardware memories, and different hardware memories are distributed in different physical locations (such as Different factories, or in different production lines), and realize the transfer of information between each other through the network, so that the data is distributed, but logically constitute a database based on big data technology.
  • data such as a non-relational database
  • the distributed storage device 501 includes multiple hardware memories, and different hardware memories are distributed in different physical locations (such as Different factories, or in different production lines), and realize the transfer of information between each other through the network, so that the data is distributed, but logically constitute a database based on big data technology.
  • monitoring system 700 includes: a distributed storage device 701, a source system 702, an analysis device 703 including one or more processors, and a display device 704.
  • the source system 702 contains a large number of raw data of different plant equipment (for example, parameter values related to temperature, parameter values related to pressure, parameter values related to electrical properties, etc.).
  • the foregoing raw data may be processed by a PLC (Programmable Logic Controller, programmable logic controller).
  • the processed data is stored in the CIM PC (a machine that stores equipment-related information), and then the EIS (Equipment International System) classifies and transmits the required data from the CIM PC for further storage
  • YMS Yield Management System
  • FDC system FDC system
  • MES Manufacturing Execution System
  • MDW Manufactory Data Warehouse
  • the system's relational database (such as Oracle, Mysql, etc.). Further, the above-mentioned data is stored in the HADOOP big data platform through ETL, so as to perform related calculations in HADOOP. Exemplarily, the data stored in the HADOOP big data platform can be displayed in the corresponding report with the BO (BUSINESS OBJECT) tool.
  • BO BUSINESS OBJECT
  • OGG (short for Oracle Golden Gate, Oracle's real-time data transmission tool) can be used to transmit the data of the source system in real time to the distributed storage device (such as Hadoop Distributed File System, HDFS Hadoop Distributed File System, HDFS) 701 data lake 71 middle.
  • the distributed storage device such as Hadoop Distributed File System, HDFS Hadoop Distributed File System, HDFS
  • the data in the distributed storage device 701 may be stored in a data warehouse (for example, Hive) or a non-relational database (for example, Hbase) format.
  • the data from the source system 702 is first stored in the data lake; after that, data cleaning, data conversion and other preprocessing can be continued in Hive according to the application theme and scene of the data.
  • different themes such as production history theme, inspection data theme, spot measurement theme, equipment data theme
  • data with different scenarios (such as a sudden bad analysis scenario and a correlation analysis scenario) are stored in the data mart 73.
  • the data in the data mart 73 may be specifically stored in the non-relational database Hbase.
  • the above data mart 73 can be connected to the analysis device 703 and the like through the API interface to realize data interaction with these devices.
  • the data mart 73 transmits the parameter data corresponding to the collected device indicators to one or more processors in the analysis device 703 through the API interface for data processing (for specific implementation, see Figure 2 and Figure 2).
  • the embodiment shown in 3) obtains a sample set; further, in one or more processors in the analysis device 703, model training is performed by training positive samples and training negative samples in the sample set (for specific implementation, see Figure 4 ⁇ ).
  • the model evaluation is performed through the test positive samples and the test negative samples in the sample set (for specific implementation, please refer to the foregoing model test embodiment).
  • the training samples of the prediction model described above are trained based on dimensionality reduction variables obtained by performing dimensionality reduction processing on multiple equipment indicators, it is necessary to predict the superiority level of equipment indicators through the prediction model.
  • the target device indicator ie, the device to be tested indicator
  • the analysis device 703 to perform data processing (ie, dimensionality reduction processing) on the target device indicator through one or more processors.
  • the parameter value corresponding to each target device indicator of M is recorded as the "first parameter value to be measured", and the N target dimensionality reduction variables obtained after the dimensionality reduction processing of the M target device indicators correspond to The parameter value is recorded as "the second parameter value to be measured”.
  • one or more processors in the analysis device 703 are further configured to: perform dimensionality reduction processing on the first parameter value to be measured according to the association relationship between the M target device indicators to obtain information about the N target device indicators.
  • the second parameter value to be measured of the dimensional variable, where N is less than a positive integer of M; the second parameter value to be measured is input to the prediction model to monitor the M target devices through the output value of the prediction model The goodness of the index.
  • this technical solution performs dimensionality reduction processing according to the correlation between the target equipment indicators, and uses a small number of comprehensive indicators to represent the information contained in a large number of original equipment indicators.
  • the process of one or more processors in the analysis device 703 specifically performing dimensionality reduction processing on the above M target device indicators includes: analyzing the association relationship between the M target device indicators through a principal component analysis algorithm; based on the association relationship The associated equipment indicators are combined to obtain N target dimensionality reduction variables; according to the first to-be-tested parameters of the target equipment indicators involved in the combined operation, the second to-be-tested parameters of the target dimensionality reduction variables are determined.
  • the display device 704 can be used to display an "interactive interface", the interactive interface can include a sub-interface that displays analysis results (such as goodness levels), and is used to control the product's failure cause analysis system to perform required tasks (such as tasks). Setting) and the sub-interface for controlling each process equipment (such as setting its process parameters).
  • analysis results such as goodness levels
  • Setting and the sub-interface for controlling each process equipment (such as setting its process parameters).
  • the user can provide relevant parameters for the algorithm design in the analysis device 703 through the interactive interface, and the user can also select a model (ie, model selection) for predicting the superiority level of the device index through the interactive interface, and so on.
  • the process of predicting the superiority level of the target dimensionality reduction variable and its corresponding second parameter to be measured through the trained prediction model is as follows, that is, one or more processors in the analysis device 703 are The specific configuration is: input the second parameter values to be measured about the N target dimensionality reduction variables into the prediction model, and respectively calculate the Fourier basis eigenvectors corresponding to the N target dimensionality reduction variables; based on linear regression
  • the algorithm processes the feature vectors of the N Fourier bases to obtain N'target screening variables with respect to the superiority level, where N'is less than N; the parameter values of the target screening variables are obtained based on the prediction model Respectively belong to the prediction probabilities of each superiority level; the superiority level corresponding to the maximum value of the prediction probability is determined as the output value of the prediction model.
  • a specific implementation manner for determining a set of failure indicators is as follows: if the output value of the prediction model is lower than the pre-set level value, one or more processors in the analysis device 703 It is also configured to: by performing a dimensionality increase process corresponding to the dimensionality reduction process on the target screening variable, a set of equipment indicators related to the target screening variable (for example, it can be expressed as RecipeX). Further, a group of equipment indicators included in the equipment indicator set RecipeX is determined as a faulty equipment indicator set. Furthermore, a correlation analysis is performed on the set of equipment indicators contained in RecipeX (that is, the correlation analysis theme shown in Figure 7) to determine the cause of the failure.
  • the quality level of the product is divided into A level, B level, C level and D level
  • B level can be taken as the above-mentioned preset level value. Therefore, if the output value of the prediction model is C-level and D-level, the equipment index set M related to the target screening variable is obtained to determine that the equipment index set M is combined into a faulty equipment index set.
  • the above-mentioned preset level value may also be level C. Then, when the output value of the prediction model is level D, the failure equipment index set is determined. That is to say, the above-mentioned faulty equipment index set is related to the pre-divided product quality level and the predicted value of the model.
  • the excellent parameter range corresponding to each device index in the index set is further determined. Specifically, since the current parameter value of at least one equipment index in the faulty equipment index set is outside the corresponding excellent parameter range, the current parameter value of the faulty equipment indicator can be adjusted according to the excellent parameter range, so that there is a fault. The equipment index value of the equipment will be restored to the range of the corresponding excellent parameters in time. Then reduce the defect rate of the product.
  • one or more processors in the analysis device 703 are further configured to determine the excellent parameter value interval of the target device indicator to be monitored based on the prediction model.
  • any one of the foregoing M target device indicators is used as the target device indicator to be monitored in this embodiment.
  • the good parameter value range of the target equipment index to be monitored is determined according to the trained prediction model. Referring to FIG. 8, the abscissa represents M target device indicators, and the ordinate is used to represent the good parameter value interval of each target device indicator. Therefore, the current parameter value of the target device indicator to be monitored can be monitored in real time through the excellent parameter value interval.
  • the parameter value of Q 1 is obtained in real time.
  • the parameter value of the target device indicator Q 1 to be monitored is within the corresponding good parameter value range, it indicates that the monitoring result of the above Q1 is good, and it also indicates that the current parameter value of Q 1 will not cause a defective product.
  • the situation of another target device index Q 2 is the same as the above-mentioned target device index Q 1 to be monitored, and a detailed description of the target device index Q 2 is omitted.
  • the parameter value of Q M is acquired in real time.
  • This embodiment Exemplarily displays the current parameter value of Q M as a triangle to remind the staff to take improvement measures in time.
  • this technical solution provides the history of abnormal parameters in the entire life cycle process through big data prediction, which can confirm the impact of equipment parameters on products in a targeted manner.
  • real-time monitoring is achieved through the trend line of variable parameters, which is conducive to determining the early warning of the parameter value in a timely manner and reducing the defective rate of the product.
  • the display device 704 has a display function for displaying analysis results (such as correlation) calculated by the analysis device.
  • the display device also has a display function, which is used to display the analysis result (such as the goodness level, etc.) calculated by the analysis device.
  • the display device may include one or more displays, including one or more terminals with display functions, so that the analysis device can send the correlation obtained by its analysis to the display device, and the display device will then display it.
  • the display interface of the display device 704 is displayed as shown in Figure 8 above: specifically, it displays the excellent parameter value interval of the target device indicator to be monitored, where the excellent parameter value interval refers to the parameter of the target device indicator to be monitored The value does not lead to the parameter value range of the defective product; and the display interface also displays the current parameter value of the target device indicator to be monitored obtained in real time.
  • the current parameter value is within the corresponding good parameter value range, it is displayed as Dot (can be changed according to actual needs, for example, can be displayed as a red dot), when the current parameter value is outside the corresponding good parameter value range, displayed as a triangular point (can be changed according to actual needs, for example, can be displayed as Green dot). Therefore, the image of the good parameter value interval and current parameter value of any target equipment index is displayed as the staff, so that the staff can monitor the above-mentioned first-dimensional variables and take corresponding measures in time for bad early warning.
  • the aforementioned monitoring system 700 further includes: an alarm device.
  • the alarm device is configured to: when the difference between the current parameter value of the target device index to be monitored and the boundary value of the excellent parameter value interval is less than a preset value, an alarm (such as a buzzer, a voice alarm light) ). In order to remind the staff that there is a bad warning, it is convenient for the staff to take timely corresponding measures to the bad warning in time.
  • FDC system data is first used as the big data basis for product superiority level prediction and analysis.
  • distributed storage devices can efficiently realize the original data of multiple factory equipment through big data. Collection and preliminary processing.
  • One or more processors (which can be arranged in the analysis device) can obtain the required data from the distributed storage device. Based on the trained prediction model, it is possible to determine the respective good parameter value ranges of the M target equipment indicators to monitor each target equipment indicator in real time, predict the possibility of failure and the alarm function, and the failure prediction can be discovered through the management of all parameters
  • the bad prediction can be automatically generated and provided to the FDC system to help the FDC system to reasonably confirm the range of relevant equipment indicators, thereby saving labor and time and detecting problems as soon as possible.
  • the state of the equipment is the optimal production state, which improves the yield rate of the production of similar products in batches.
  • the present technical solution also provides a computer-readable medium 900, which can adopt a portable compact disk read-only memory (CD-ROM) and includes program code, and can run on a terminal device, such as a personal computer.
  • a computer-readable medium 900 can adopt a portable compact disk read-only memory (CD-ROM) and includes program code, and can run on a terminal device, such as a personal computer.
  • the program product of the present disclosure is not limited thereto.
  • the readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, device, or device.
  • the above-mentioned program product can adopt any combination of one or more readable media.
  • the readable medium may be a readable signal medium or a readable storage medium.
  • the readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above.
  • readable storage media include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (Read -Only Memory, ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, Or any suitable combination of the above.
  • the computer-readable signal medium may include a data signal propagated in baseband or as a part of a carrier wave, and readable program code is carried therein. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the readable signal medium may also be any readable medium other than a readable storage medium, and the readable medium may send, propagate, or transmit a program for use by or in combination with the instruction execution system, apparatus, or device.
  • the program code contained on the readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.
  • the program code for performing the operations of the present disclosure can be written in any combination of one or more programming languages.
  • the above-mentioned programming languages include object-oriented programming languages—such as Java, C++, etc., as well as conventional procedural programs. Design language-such as "C" language or similar programming language.
  • the program code can be executed entirely on the user's computing device, partly on the user's device, executed as an independent software package, partly on the user's computing device and partly executed on the remote computing device, or entirely on the remote computing device or server Executed on.
  • the remote computing device can be connected to the user computing device through any kind of network, including Local Area Network (LAN) or Wide Area Network (WAN), or it can be connected to the outside.
  • Computing equipment for example, using an Internet service provider to connect via the Internet).
  • an electronic device capable of implementing the above method is also provided.
  • the electronic device 1000 according to this embodiment of the present disclosure will be described below with reference to FIG. 10.
  • the electronic device 1000 shown in FIG. 10 is only an example, and should not bring any limitation to the function and scope of use of the embodiments of the present disclosure.
  • the electronic device 1000 is represented in the form of a general-purpose computing device.
  • the components of the electronic device 1000 may include, but are not limited to: the aforementioned at least one processing unit 1010, the aforementioned at least one storage unit 1020, and a bus 1030 connecting different system components (including the storage unit 1020 and the processing unit 1010).
  • the foregoing storage unit stores program codes, and the foregoing program codes can be executed by the foregoing processing unit 1010, so that the foregoing processing unit 1010 executes the method steps described in the foregoing embodiments of this specification.
  • the storage unit 1020 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 10201 and/or a cache storage unit 10202, and may further include a read-only memory unit.
  • RAM random access memory
  • ROM Read-Only Memory
  • the storage unit 1020 may also include a program/utility tool 10204 having a set of (at least one) program module 10205.
  • program module 10205 includes but is not limited to: an operating system, one or more application programs, other program modules, and program data, Each of these examples or some combination may include the implementation of a network environment.
  • the bus 1030 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local area using any bus structure among multiple bus structures. bus.
  • the electronic device 1000 can also communicate with one or more external devices 1100 (such as keyboards, pointing devices, Bluetooth devices, etc.), and can also communicate with one or more devices that enable a user to interact with the electronic device 1000, and/or communicate with Any device (such as a router, modem, etc.) that enables the electronic device 1000 to communicate with one or more other computing devices.
  • This communication can be performed through an input/output (Input/Output, I/O) interface 1050.
  • I/O interface 1050 is connected to the display unit 1040 to transmit the content to be displayed to the display unit 1040 through the I/O interface 1050 for the user to view.
  • the electronic device 1000 may also communicate with one or more networks (for example, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through the network adapter 1060.
  • networks for example, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet
  • the network adapter 1060 communicates with other modules of the electronic device 1000 through the bus 1030.
  • other hardware and/or software modules can be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives And data backup storage system, etc.
  • the example embodiments described here can be implemented by software, or can be implemented by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk, etc.) or on the network , Including several instructions to make a computing device (which may be a personal computer, a server, a terminal device, or a network device, etc.) execute the method according to the embodiments of the present disclosure.
  • a computing device which may be a personal computer, a server, a terminal device, or a network device, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Factory Administration (AREA)

Abstract

一种设备指标优良性等级预测模型的训练方法，包括：获取多个设备指标的历史参数值（S110）；根据所述多个设备指标之间的关联关系，对所述历史参数值进行降维处理，得到关于降维变量的样本数据（S120）；根据所述降维变量的样本数据的优良性等级，确定正样本集合和负样本集合（S130）；通过所述正样本集合和所述负样本集合训练机器学习模型，得到用于预测目标设备指标的优良性等级的预测模型（S140）。

Description

设备指标优良性等级预测模型训练方法、监控系统和方法 技术领域
本公开实施例涉及显示面板制造领域，特别涉及一种设备指标优良性等级预测模型的训练方法、设备指标的监控系统和方法、计算机可读介质。
背景技术
显然，显示面板在生产过程中，需依次经过多个工艺设备的处理。同时，最终的显示面板产品不可避免的有一定概率存在各种不良，而产品的优良性等级的本质上与工艺设备相关，具体是根据设备指标的参数值引起的。
因此，确定工艺设备的设备指标与产品的优良性等级的相关性，对定位不良、调整生产流程等有重要意义。
发明内容
本公开实施例提供一种设备指标优良性等级预测模型的训练方法、设备指标的监控系统和方法、计算机可读介质。
在本公开的一种实施例中，提供了一种设备指标优良性等级预测模型的训练方法，包括：
获取多个设备指标的历史参数值；
根据上述多个设备指标之间的关联关系，对上述历史参数值进行降维处理，得到关于降维变量的样本数据；
根据上述降维变量的样本数据的优良性等级，确定正样本集合和负样本集合；
通过上述正样本集合和上述负样本集合训练机器学习模型，得到用于预测目标设备指标的优良性等级的预测模型。
在本公开的一种实施例中，上述根据上述多个设备指标之间的关联关系，对上述历史参数值进行降维处理，包括：
通过主成分分析算法分析上述多个设备指标之间的关联关系；
基于上述关联关系将相关联的设备指标进行组合操作，得到降维变量；
根据上述组合操作涉及的设备指标的历史参数值，确定上述降维变量的样本数据。
在本公开的一种实施例中，上述通过上述正样本集合和上述负样本集合训练机器学习模型，包括：
分别对上述正样本集合中和上述负样本集合中的降维变量进行采样，得到采样变量；
计算每个上述采样变量的傅里叶基的特征向量，以确定上述采样变量之间的相关性；
基于线性回归算法处理上述傅里叶基的特征向量，得到筛选变量；
基于上述筛选变量和上述筛选变量对应的参数值训练机器学习模型。
在本公开的一种实施例中，上述样本数据包括训练样本和测试样本；
上述根据上述降维变量的样本数据的优良性等级，确定正样本集合和负样本集合，包括：
根据上述降维变量的样本数据中训练样本的优良性等级，确定正样本集合和负样本集合；
上述方法还包括：通过上述测试样本评估上述预测模型。
在本公开的一种实施例中，通过以下评估指标中的一种或多种评估上述预测模型，上述评估指标包括：准确率、召回率和接收者操作特征曲线下面积AUC。
在本公开的一种实施例中，上述获取每个上述设备指标的历史参数值，包括：
将关于设备指标的源数据加载至数据仓库以进行数据挖掘分析；
将上述挖掘分析之后的数据存储至非关系型数据库以供确定上述历史参数值。
在本公开的一种实施例中，上述方法还包括：
获取在数据对象中参数值为空值的待清理设备指标；
获取在其他数据对象中待清理设备指标的属性值的众数值或平均值；
确定上述众数值或上述平均值为上述数据对象中待清理设备指标的属性值。
在本公开的一种实施例中，上述机器学习模型为贝叶斯算法模型、决策树算法模型或神经网络模型。
在本公开的一种实施例中，提供了一种设备指标的监控系统，包括：分布式存储设备，以及一个或多个处理器，其中，
上述分布式存储设备，被配置为分别获取关于M个目标设备指标的第一待测参数值，其中，M为大于1的整数；
上述一个或多个处理器，被配置为基于上述预测模型监控上述M个目标设备指标的优良性。
在本公开的一种实施例中，上述一个或多个处理器还被具体配置为：
根据上述M个目标设备指标之间的关联关系，对上述第一待测参数值进行降维处理，得到关于N个目标降维变量的第二待测参数值，其中，N小于M的正整数；
将上述第二待测参数值输入上述预测模型，以通过上述预测模型的输出值监控上述M个目标设备指标的优良性。
在本公开的一种实施例中，上述一个或多个处理器被具体配置为：
通过主成分分析算法分析上述M个目标设备指标之间的关联关系；
基于上述关联关系将相关联的设备指标进行组合操作，得到N个目标降维变量；
根据上述组合操作涉及的目标设备指标的第一待测参数，确定上述目标降维变量的第二待测参数。
在本公开的一种实施例中，上述一个或多个处理器被具体配置为：
将上述关于N个目标降维变量的第二待测参数值输入上述预测模型，分别计算上述N个目标降维变量对应的傅里叶基的特征向量；
基于线性回归算法处理上述N个傅里叶基的特征向量，得到关于上述优良性等级的N’个目标筛选变量，其中，N’小于N；
基于上述预测模型得到上述目标筛选变量的参数值分别属于各个优良性等级的预测概率；
确定预测概率最大值对应的优良性等级为上述预测模型的输出值。
在本公开的一种实施例中，上述一个或多个处理器还被配置为：
若上述预测模型的输出值为上述优良性等级低于预设等级值较低，则，则对上述目标降维变量进行升维处理得到故障设备指标集合。
在本公开的一种实施例中，上述一个或多个处理器还被配置为：
基于上述预测模型确定待监控目标设备指标的优良参数值区间，其中，上述待监控目标设备指标为上述M个目标设备指标中的任意一个；
根据上述优良参数值区间实时监控上述待监控目标设备指标的当前参数值。
在本公开的一种实施例中，上述系统还包括：显示设备，被配置为：
显示关于待监控目标设备指标的优良参数值区间，以及显示实时获取的上述待监控目标设备指标的当前参数值。
在本公开的一种实施例中，上述系统还包括：报警设备，被配置为：
当上述待监控目标设备指标的当前参数值与上述优良参数值区间的边界值的差值小于预设值时，发出警报。
在本公开的一种实施例中，提供了一种设备指标的监控方法，包括：
分别获取关于M个目标设备指标的第一待测参数值，其中，M为大于1的整数；
基于上述设备优良性等级预测模型的训练方法实施例提供的预测 模型监控上述M个目标设备指标的优良性。
在本公开的一种实施例中，上述基于上述预测模型监控上述M个目标设备指标的优良性，包括：
根据上述M个目标设备指标之间的关联关系，对上述第一待测参数值进行降维处理，得到关于N个目标降维变量的第二待测参数值，其中，N小于M的正整数；
将上述第二待测参数值输入上述预测模型，以通过上述预测模型的输出值监控上述M个目标设备指标的优良性。
在本公开的一种实施例中，上述根据上述M个目标设备指标之间的关联关系，对上述第一待测参数值进行降维处理，包括：
通过主成分分析算法分析上述M个目标设备指标之间的关联关系；
基于上述关联关系将相关联的设备指标进行组合操作，得到N个目标降维变量；
根据上述组合操作涉及的目标设备指标的第一待测参数，确定上述目标降维变量的第二待测参数。
在本公开的一种实施例中，提供了一种计算机可读介质，其上存储有计算机程序，其中，所述程序被处理器执行时，实现如上述所述的设备指标优良性等级预测模型的训练方法，以及，实现如上所述的设备指标的监控方法。
附图说明
附图用来提供对本公开实施例的进一步理解，并且构成说明书的一部分，与本公开实施例一起用于解释本公开，并不构成对本公开的限制。通过参考附图对详细示例实施例进行描述，以上和其它特征和优点对本领域技术人员将变得更加显而易见，在附图中：
图1为本公开实施例提供的一种设备指标优良性等级预测模型的训练方法的流程示意图；
图2为本公开实施例提供的一种设备指标的历史参数值的获取方法的流程示意图；
图3为本公开实施例提供的一种降维处理方法的流程示意图；
图4为本公开实施例提供的一种模型训练方法的流程示意图；
图5为本公开实施例提供的一种设备指标的监控系统的结构示意图；
图6为本公开实施例提供的一种设备指标的监控系统中分析设备的结构示意图；
图7为本公开实施例提供的另一种设备指标的监控系统的结构示意图；
图8为本公开实施例提供的一种设备指标的监控方式的示意图；
图9为本公开实施例提供的一种计算机可读介质的示意图；
图10为本公开实施例提供的一种电子设备的示意图。
具体实施方式
为使本领域的技术人员更好地理解本公开实施例的技术方案，下面结合附图对本公开实施例提供的产品不良成因分析的系统和方法、计算机可读介质进行详细描述。
在下文中将参考附图更充分地描述本公开实施例，但是所示的实施例可以以不同形式来体现，且不应当被解释为限于本公开阐述的实施例。反之，提供这些实施例的目的在于使本公开透彻和完整，并将使本领域技术人员充分理解本公开的范围。
本公开实施例可借助本公开的理想示意图而参考平面图和/或截面图进行描述。因此，可根据制造技术和/或容限来修改示例图示。
在不冲突的情况下，本公开各实施例及实施例中的各特征可相互组合。
本公开所使用的术语仅用于描述特定实施例，且不意欲限制本公开。如本公开所使用的术语“和/或”包括一个或多个相关列举条目的任何和所有组合。如本公开所使用的单数形式“一个”和“该”也意欲包括复数形式，除非上下文另外清楚指出。如本公开所使用的术语“包括”、“由……制成”，指定存在所述特征、整体、步骤、操作、元件和/或组件，但不排除存在或添加一个或多个其它特征、整体、步骤、操作、元件、组件和/或其群组。
除非另外限定，否则本公开所用的所有术语(包括技术和科学术语)的含义与本领域普通技术人员通常理解的含义相同。还将理解，诸如那些在常用字典中限定的那些术语应当被解释为具有与其在相关技术以及本公开的背景下的含义一致的含义，且将不解释为具有理想化或过度形式上的含义，除非本公开明确如此限定。
本公开实施例不限于附图中所示的实施例，而是包括基于制造工艺而形成的配置的修改。因此，附图中例示的区具有示意性属性，并且图中所示区的形状例示了元件的区的具体形状，但并不是旨在限制性的。
许多产品(如显示面板)都是通过生产线生产的，每条生产线包括多个工艺站点，每个工艺站点用于对产品(包括半成品)进行一定的处理(如清洗、沉积、曝光、刻蚀、对盒、检测等)。同时，每个工艺站点通常有多个用于进行同样处理的工艺设备；当然，虽然理论上进行的处理相同，但不同工艺设备由于型号、状态等的不同，故实际的处理效果并不完全相同。
其中，每个产品的生产过程需要经过多个工艺站点，且不同产品在生产过程中经过的工艺站点可能不同；而经过同一工艺站点的产品也可能由其中的不同工艺设备处理。因此，在一条生产线中，每个工艺设备都会参与部分产品的生产过程，但不是参与全部产品的生产过程，即每个工艺设备都参与且仅参与部分产品的生产过程。
第一方面，本公开实施例提供一种设备指标优良性等级预测模型的训练方法。
通过本公开实施例提供的训练方法训练得到的设备指标优良性等级预测模型能预测设备指标的参数值，并通过历史参数趋势实时监控预测参数值，进而预测不良发生的可能性；生产线中的各工艺设备与产品的各种不良的相关性，即确定产品不良的成因，以进行定位不良、调整生产流程等。
进一步地，通过本公开实施例提供的训练方法训练得到的设备指标优良性等级预测模型，还能够得到每个设备指标的参数的最优趋势图，在同类产品生产时，能够尽早发现问题，保证同类产品的生产时，设备状态都是最优的生产状态，批量性的提升同类产品生产的良率。
其中，“不良”是指产品中的质量缺陷，这些缺陷可能导致产品品质降低甚至报废，也可能导致产品需要进行返工或修复。
其中，不良可根据需要分为不同种类。例如，可根据不良对产品性能的直接影响进行分类，如亮线不良、暗线不良、萤火虫不良(hot spot)等；或者，也可根据不良的具体成因进行分类，如栅线与数据线短路不良、对位不良等；或者，也可根据不良的大体成因进行分类，如阵列工艺不良、彩膜工艺不良等；或者，也可根据不良的严重程度进行分类，如导致报废的不良、导致降低品质的不良等；或者，也可不区分不良的种类，即只要产品存在任何“不良”，即认为其有“不良”，反之则认为其为“良”。
当然，以上相关性是针对“一种”不良而言的，即同一工艺设备对不同种的不良有不同的相关性。其中，“工艺设备与某种不良的相关性”是指，该工艺设备的参与，对产品是否发生该种不良的概率的影响程度。
在一些实施例中，产品为显示面板生产线。本公开实施例可用于在显示面板(如液晶显示面板、有机发光二极管显示面板等)的生产过程中，确定显示面板生产线的各工艺设备与不良的相关性。当然，本公开实施例也可用于其它产品。
参考图1，本公开实施例提供的一种设备指标优良性等级预测模型的训练方法步骤S110-步骤S140。
在步骤S110中，获取多个设备指标的历史参数值。
其中，上述设备指标与产品(如，OLED)品质相关，包括：温度、压力、湿度、加热时间、冷却时间等。上述设备指标可以是多个设备的指标，也可以是某一个设备的指标。上述每个述设备指标的历史参数值包括：预设的历史时长内温度值、预设的历史时长内压力值、预设的历史时长内压力值等。
在示例性的实施例中，相关技术中，针对每个设备指标单独建立监控模型以实现对该设备指标的参数监控，进而各个设备指标的参数监控需人工单独监控。然而，设备指标较多将会导致监控模型的数据较多，为了节省人力物力，则只针对某些重要参数进行管理，从而导致监控范围较小且监控准确度有待提高。
为了克服上述问题，本技术方案以工厂数据采集(Factory DataCollection，简称：FDC)系统数据作为训练上述设备指标优良性等级预测模型的大数据基础。示例性的，FDC系统为实时监控和分析产设备在产品生产过程数据(PROCESS data)的系统，能够通过实时监控及分析数据来减少报废率和生产时间并提高产品的质量及产量。从而，收集到设备的各个指标及其参数。本方案基于多个设备指标及其历史参数值来确定训练上述分类模型的样本，可以有效扩大监控范围且提升监控准确度。
在示例性的实施例中，图2所示实施例提供了上述步骤S110中获取每个设备指标的历史参数值的具体实施方式，包括：
步骤S210，将关于设备指标的源数据加载至数据仓库(如，Hive)以进行数据挖掘分析；以及，步骤S220，将所述挖掘分析之后的数据存储至非关系型数据库(如，Hbase)以供确定所述历史参数值。
具体的，上述设备指标的源数据来源于FDC系统中的状态变量(State Variable，简称：SV)数据。其中，SV数据用于表示设备指标的实时状态参数(如，当前温度为80摄氏度)。本实施例中，可以对SV数据中BLOB(BinaryLargeObject，二进制大对象)字段进行解析，以生成可视化数据。其中，可视化数据是指通过借助图形化手段的方式在视觉方面想用户传达数据，从而更加清晰有效地传达与沟通信息。
本实施例中，还可以通过ETL(Extraction-Transformation-Loading，即数据抽取(Extract)、转换(Transform)、装载(Load)的过程，是构建数据仓库的重要环节)将分散，零乱，标准不统一的SV数据(其中可以包括上述可视化数据)整合到一起，并将其导入到数据仓库(如，HIVE)，为进一步地数据决策提供分析依据。从而可以在数据仓库对海量结构化全量数据进行存储、运算、挖掘分析。
进一步地，再通过ETL将数据仓库HIVE中的数据存储到非关系型数据库(如，Hbase)中，以通过Hbase来存储海量半结构化数据、以及实现对数据的检索和实时分析。从而可以分别确定上述各个设备指标的历史参数值。
在示例性的实施例中，为了提升训练后模型的预测准确度，在确定样 本集时，还进行数据清理。具体的：
获取在数据对象中参数值为空值的待清理设备指标；获取在其他数据对象中待清理设备指标的属性值的众数值或平均值；以及，确定所述众数值或所述平均值为所述数据对象中待清理设备指标的属性值。
示例性的，对存储于上述非关系型数据库Hbase中的数据进行清洗。将参数值为空值作为待清理设备指标，具体的，如果空值是数值型的，就根据该属性在其他所有对象的取值的平均值来填充该缺失的属性值；如果空值是数值型的，就根据统计学中的众数原理，用该属性在其他所有对象的取值次数最多的值来补齐该缺失的值，从而以最大概率可能的取值来补充缺失的参数值，进而有利于提升训练后模型的预测准确度。
继续参考图1，在步骤S120中，根据所述多个设备指标之间的关联关系，对上述历史参数值进行降维处理，得到关于降维变量的样本数据。
在示例性的实施例中，相关技术中，针对每个设备指标单独建立监控模型以实现对该设备指标的参数监控，进而各个设备指标的参数监控需单独监控。由于针对每个设备指标单独建立监控模型以实现对该设备指标的参数监控，则各个设备指标之间缺乏关联性，当目标监控设备发生问题时，则难以无法获取与目标设备参数相关的其他设备参数信息，导致监控效率低。
为了克服上述问题，本技术方案根据设备指标之间的关联关系，对相关历史参数值进行降维处理，以得到关于降维变量的样本数据。
具体地，图3为本公开实施例提供的一种降维处理方法的流程示意图，可以作为步骤S120的一种具体实施方式。参考图3，该实施例所示方法包括：
步骤S310，通过主成分分析算法分析所述多个设备指标之间的关联关系；步骤S320，基于所述关联关系将相关联的设备指标进行组合操作，得到降维变量；以及，步骤S330，根据所述组合操作涉及的设备指标的历史参数值，确定所述降维变量的样本数据。
在示例性的实施例中，由于设备指标之间存在一定的相关关系，则可以将具有相关关系的设备指标作为一个集合(即“设备指标集合”)。一组 设备指标集合还可以称为“Recipe”。具体指在产品制造过程中与完成产品某一功能相关的一组设备指标集合；例如，在OLED制造过程中，与封装阶段相关的一组设备参数(如，温度、压强和电性)。同时，随着技术改进一组Recipe的参数值也是不断变化的。那么，以上述与封装阶段相关的Recipe为例，某次封装技术改进之前，可以将其称为第一版本的Recipe，在该次封装技术改进之后，可以将其称为第二版本的Recipe。本实施例中，通过比较出不同版本间Recipe中参数值的差异，为优化Recipe提供参考根据。
进一步地，对一组Recipe中各个设备指标及不同工艺步骤提供量化统计结果，因此将关系紧密的设备指标(变量)降维处理为综合指标(即数量较少的新变量)，得到的综合指标是两两不相关的。从而，便可以用较少的综合指标来代表数量较多的原设备指标所包含的信息。
示例性的，利用PCA(主成分分析算法，就是一种对高维度特征数据预处理方法。降维是将高维度的数据保留下最重要的一些特征，去除不重要的特征，从而实现提升数据处理速度的目的，降维在一定的信息损失范围内，可以为我们节省大量的时间和成本)方法将数据进行预处理，将高维的数据映射到低维的空间，在维度上使数据的方差最大，以此使用较少的数据维度，保留较多的原数据点的特性，最大化保持数据的内在信息。
进一步地，在步骤S130中，根据上述降维变量的样本数据的优良性等级，确定正样本集合和负样本集合。
在示例性的实施例中，根据产品的优良性等级确定样本数据的优良性等级，例如，该产品的质量被划分为A级、B级、C级和D级，对应的，与制造出A级产品相关的样本数据可以被划分为A级、与制造出B级产品相关的样本数据可以被划分为B级、与制造出C级产品相关的样本数据可以被划分为C级以及与制造出D级产品相关的样本数据可以被划分为D级。进一步地，可以将降维变量的样本数据中属于A级和属于B级的训练样本的优良性等级确定为正样本集合，将降维变量的样本数据中属于C级和属于D级的训练样本的优良性等级确定为负样本集合。
示例性的，以OLED产品的终检数据确定正样本集合和负样本集合。 具体的，若OLED产品的终检数据良品，则生产该OLED产品的设备指标参数作为正样本，从而确定正样本集合；若OLED产品的终检数据不良品，则生产该OLED产品的设备指标参数作为负样本，从而确定负样本集合。需要说明的是：为了确保预测准确度，确定上述正负样本集合时是针对单一型号的产品进行的。另外，为了避免样本分布不均匀，导致的预测不准确，针对单一型号的产品，上述正样本集合至少包含1万条正样本信息，上述负样本集合至少包含1万条负样本信息。
进一步地，在步骤S140中，通过上述正样本集合和上述负样本集合训练机器学习模型，得到用于预测目标设备指标的优良性等级的分类模型。
在示例性的实施例中，图4为本公开实施例提供的一种模型训练方法的流程示意图，可以作为步骤S140的一种具体实施方式。参考图4，该实施例所示方法包括：
步骤S410，分别对所述正样本集合中和所述负样本集合中的降维变量进行采样，得到采样变量；步骤S420，计算每个所述采样变量的傅里叶基的特征向量，以确定所述采样变量之间的相关性；步骤S430，基于线性回归算法处理所述傅里叶基的特征向量，得到筛选变量；以及，步骤S440，基于所述筛选变量和所述筛选变量对应的参数值训练机器学习模型。
示例性的，将上述正样本集合随机分为训练正样本和测试正样本，以及将上述负样本集合随机分为训练负样本和测试负样本。该图所示实施例通过训练正样本和训练负样本训练机器学习模型。其中，上述机器学习模型为贝叶斯算法模型、决策树算法模型或神经网络分类模型。
示例性的，在训练样本对应的参数空间中，随机对设备指标对应的降维变量进行采样，得到采样变量。然后，对每个采样变量对应的参数值计算低度数傅里叶基的特征向量，以确定采样变量之间的相关性。进一步地，基于线性回归算法(如，拉索(lasso)算法)处理上述傅里叶基的特征向量，得到筛选变量。完成一轮筛选，筛选变量是对产品优良性等级影响较大的特征。基于至少一轮筛选，将筛选特征及其对应的参数训练机器学习模型。通过图4所示的技术方案，可以去除对产品优良性等级影响不足够大的噪声变量，有利于提升模型训练速度。
在示例性的实施例中，本技术方案还包括通过测试正样本和测试负样本评估优良性等级预测模型(记作“待测试模型”)进行测试，并且使用至少一种测试指标对待测试模型的测试结果进行验证，并将符合测试指标的模型进行优良性等级预测。
具体的，通过以下评估指标中的一种或多种评估模型性能，所述评估指标包括：准确率、召回率、KS(Kolmogorov-Smirnov，柯尔莫戈洛夫-斯米尔诺夫)值和接收者操作特征曲线(Receiver Operating Characteristic curve，简称：ROC)下面积AUC(一种模型评估指标，具体用于评估模型的预测价值；是Area Under ROC Curve的简称)对上述迭代优化后的预测模型进行评估。
在示例性的实施例中，具体对待测试模型进行测试的方法包括：
首先，根据测试样本的描述特征输入至待测试模型，模型的输出数据得到以下：真阳性TP，真阴性TN，伪阴性FN和伪阳性FP。其中，TP是利用待测试模型对测试样本集中正类进行判断后属于仍是正类的数目，TN利用待测试模型对测试样本集负类进行判断后属于仍是负类的数目，FN利用待测试模型对测试样本集中正类进行判断后属于是负类的数目，FP利用待测试模型对测试样本集负类进行判断后属于是正类的数目。正类和负类是指人工对第一部分样本标注的两种类别，即人工标注某个样本属于特定的类，则该样本属于正类，不属于该特定类的样本则属于负类。
其次，根据真阳性TP，真阴性TN，伪阴性FN和伪阳性FP计算待测试模型的测试结果。
在示例性的实施例中，测试指标可以为AUC或KS值。具体的：
在示例性的实施例中，利用公式一和公式二确定伪阳性率FPR和真阳性率TPR，
FPR＝FP/(FP+TN)公式一
TPR＝TP/(TP+FN)公式二
进一步地，以FPR为横坐标，TPR为纵坐标，绘制ROC曲线。其中，ROC曲线是获得的各指标的特征曲线，用于展示各指标之间的关系，并进一步计算出ROC曲线下面积AUC。ROC曲线是获得的各指标的特征曲线， 用于展示各指标之间的关系，AUC即ROC曲线下面积，AUC越大，则模型的预测价值越高，进而可通过AUC对待测试模型进行测试。并可以在评估结果为AUC值满足预设阈值时，便可以将得到的模型用于预测优良性等级。
KS(每次选取一个不同的阈值，我们就可以得到一组FPR和TPR，即ROC曲线上的一点)曲线则是两条曲线的在每一个阈值下的差值。KS＝max(TPR-FPR)，即为TPR与FPR的差的最大值；KS值可以反映模型的最优区分效果，此时所取的阈值一般作为定义好坏用户的最优阈值。KS值的取值范围是[0，1]，且KS值越大，说明模型的预测准确准确度越高。示例性的，KS>0.2即可认为模型有比较好的预测准确度。
在示例性的实施例中，模型测试指标还可以为准确率、召回率。具体的，根据公式三和公式四分别计算准确率p和召回率r；
p＝TP/(TP+FP)公式三
r＝TP/(TP+FN)公式四
假如，测试指标对应的设定条件为：准确率测试结果大于p’(预设值)则为满足准确率设定条件，否则不满足准确率设定条件，以及召回率测试结果大于r’(预设值)则为满足召回率设定条件，否则不满足召回率设定条件。
在示例性的实施例中，在测试结果满足测试指标对应的设定条件的情况下，则待测试模型可以作为用于预测设备指标的参数值的预测模型；在测试结果不满足设定条件时，则上述待测试模型继续迭代优化直至所述待测试模型的测试结果满足设定条件。
第二方面，本公开实施例提供一种设备指标的监控系统。
由于涉及本技术方案涉及多个工厂的多个工厂设备，故关于设备指标以及参数值的原始数据的数据量是很大的。例如，所有工厂设备每天产生的原始数据可能有几百G，每小时产生的数据也可能有几十G。对海量结构化数据实现存储与计算主要有两种方案：RDBMS关系型数据库管理(Relational Database Management System,RDBMS)的网格计算方案；分布式文件管理系统(Distributed File System,DFS)的大数据方案。
其中，RDBMS的网格计算是把需要非常巨大的计算能力的问题分成许多小部分，然后把这些部分分配给许多计算机分别处理，最后把这些计算结果综合起来。例如，作为一种具体例子，Oracle RAC(真正应用集群)是Oracle数据库支持的网格计算的核心技术，其中所有服务器都可直接访问数据库中的所有数据。但是，RDBMS的网格计算的应用系统在数据量很大时无法满足用户要求，例如，由于硬件的扩展空间有限，故数据增加到足够大的数量级后，会因为硬盘的输入/输出的瓶颈使得处理数据的效率非常低。
分布式文件管理为基础的大数据技术，则允许采用多个廉价硬件设备构建大型集群，以对海量数据进行处理。如Hive工具是基于Hadoop的数据仓库工具，可用来进行数据提取转化加载(Extraction-Transformation-Loading，简称：ETL)，Hive工具定义了简单的类SQL查询语言，同时也允许通过自定义的MapReduce的mapper和reducer来默认工具无法完成的复杂的分析工作。Hive工具没有专门的数据存储格式，也没有为数据建立索引，用户可以自由的组织其中的表，对非关系型数据库中的数据进行处理。可见，分布式文件管理的并行处理可满足海量数据的存储和处理要求，用户可通过SQL查询处理简单数据，而复杂处理时可采用自定义函数来实现。因此，在对工厂的海量数据分析时，需要将工厂数据库的数据抽取到分布式文件系统(如下文所述的分布式存储设备501/701)中，具体地：
通过ETL将分散，零乱，标准不统一的SV数据整合到一起，并将其导入到数据仓库HIVE，为进一步地数据决策提供分析依据。从而可以在数据仓库HIVE对海量结构化全量数据进行存储、运算、挖掘分析。进一步地，再通过ETL将数据仓库HIVE中的数据存储到非关系型数据库Hbase中，以通过Hbase来存储海量半结构化数据、以及实现数据检索以及实时数据分析。
从而，一方面不会对原始数据造成破坏，另一方面提高了数据分析效率。
参照图5，本公开实施例提供的设备指标的监控系统500包括分布式 存储设备501，以及一个或多个处理器502。
上述分布式存储设备501，被配置为分别获取关于M个目标设备指标的第一待测参数值，其中，M为大于1的整数；为存储工厂设备产生的生产数据。上述一个或多个处理器502，被配置为基于上述实施例提供的分类模型监控所述M个目标设备指标的优良性。执行确定相关性的操作。
其中，上述分布式存储设备501中存储有来自工厂设备的生产数据。其中，工厂设备是指各工厂中的任何设备，其可包括各工艺站点中的工艺设备，也可包括工厂中用于管理生产线的管理设备等；而生产数据是指与生产相关的任何信息，包括各生产线中的每个工艺设备参与了哪些产品的生产过程，以及每个产品最终是否存在不良以及存在何种不良。
参照图6，上述一个或多个处理器(如CPU)502可处于分析设备600中。示例性的，分析设备600还包括可具有存储有所需程序的存储器(如硬盘)601，处理502与存储器601通过I/O接口602连接，从而能实现信息交互，由此处理器502可根据存储器601中存储的程序进行所需运算，以实现确定相关性的操作。
其中，分布式存储设备501中存储有相对完整的数据(如一个非关系型数据库)，而且，分布式存储设备501包括多个硬件的存储器，且不同的硬件存储器分布在不同物理位置(如在不同工厂，或在不同生产线)，并通过网络实现相互之间信息的传递，从而其数据是分布式关系的，但在逻辑上构成一个基于大数据技术的数据库。
参照图7所示的监控系统700，包括：分布式存储设备701、源系统702、包含一个或多个处理器的分析设备703以及显示设备704。
其中，源系统702中包含大量不同工厂设备的原始数据(如，关于温度的参数值、关于压力的参数值、关于电性的参数值等)。示例性的，可以将上述原始数经过PLC(Programmable Logic Controller，可编程逻辑控制器)处理。进一步地，将处理后的数据存储到CIM PC(存储设备相关信息的机器)，然后，EIS(Equipment international System，设备国际化系统)从CIM PC中将按需要的数据进行分类传输，以进一步存储在多个生产制造系统中，如YMS(Yield Management System，良率管理系统)存储 良率相关数据、FDC系统、MES(Manufacturing Execution System，制造执行系统)、MDW(ManufactoryData Warehouse，制造数据仓库)等系统的关系型数据库(如Oracle、Mysql等)中。进一步地，将上述数据通过ETL的方式存储到HADOOP大数据平台中，以在HADOOP进行相关计算。示例性的，存储至HADOOP大数据平台的数据可以再以BO(BUSINESS OBJECT)工具进行相应报表展示。
进一步地，为了便于后续分析设备703的数据读取，同时降低对工厂设备和生产制造系统的负载。可以利用OGG(Oracle Golden Gate的简称，是ORACLE的实时传输数据的工具)将源系统的数据实时传输至分布式存储设备(如Hadoop Distributed File System，HDFSHadoop Distributed File System，HDFS)701的数据湖71中。
分布式存储设备701中的数据可采用数据仓库(如，Hive)或非关系型数据库(如，Hbase)格式存储。示例性的，来自源系统702的数据先存储在数据湖中；之后，可继续在Hive中按照数据的应用主题、场景等进行数据清洗、数据转换等预处理。示例性的，在数据仓库72中通过ETL处理等方式得到具有不同主题(如生产履历主题、检测数据主题、点位测量主题、设备数据主题)。进一步地，将具有不同场景(如突发不良分析场景、相关性分析场景)的数据存储至数据集市73。示例性的，数据集市73中的数据可具体存储于非关系型数据库Hbase中。以上数据集市73可再通过API接口与分析设备703等连接，以实现与这些设备间的数据交互。
示例性的，数据集市73将采集到的设备指标对应的参数数据通过API接口被传输至分析设备703中一个或多个处理器中，以进行数据处理(具体实施方式可参见图2和图3所示实施例)得到样本集；进一步地，在分析设备703中一个或多个处理器中，通过样本集中的训练正样本和训练负样本进行模型训练(具体实施方式可参见图4所示的实施例)。另外，在分析设备703中一个或多个处理器中，通过样本集中的测试正样本和测试负样本进行模型评价(具体实施方式可参见上述模型测试的实施例)。
在示例性的实施例中，鉴于上述预测模型的训练样本是基于多个设备 指标进行降维处理得到的降维变量进行训练的，因此，在通过预测模型预测设备指标的优良性等级时，需将分布式存储设备701中获取到的目标设备指标(即待测试设备指标)传输至分析设备703，以通过一个或多个处理器对目标设备指标进行数据处理(即，降维处理)。
具体的，本实施例中，将M各目标设备指标对应的参数值记作“第一待测参数值”，将M个目标设备指标进行降维处理之后得到的N个目标降维变量对应的参数值记作“第二待测参数值”。则分析设备703中一个或多个处理器还被配置为：根据所述M个目标设备指标之间的关联关系，对所述第一待测参数值进行降维处理，得到关于N个目标降维变量的第二待测参数值，其中，N小于M的正整数；将所述第二待测参数值输入所述预测模型，以通过所述预测模型的输出值监控所述M个目标设备指标的优良性。
可见，本技术方案根据目标设备指标之间的关联关系进行降维处理，通过数量较少的综合指标来代表数量较多的原设备指标所包含的信息。
分析设备703中一个或多个处理器具体对上述M个目标设备指标进行降维处理的过程包括：通过主成分分析算法分析所述M个目标设备指标之间的关联关系；基于所述关联关系将相关联的设备指标进行组合操作，得到N个目标降维变量；根据所述组合操作涉及的目标设备指标的第一待测参数，确定所述目标降维变量的第二待测参数。
需要说明的是，由于分析设备703中一个或多个处理器进行上述降维处理过程与图3所示实施例相同，因此在此不再赘述。
参考图7，显示设备704可用于显示“交互界面”，该交互界面可包括显示分析结果(如优良性等级)的子界面，用于控制该产品不良成因分析的系统进行所需工作(如任务设定)的子界面，以及对各工艺设备进行控制(如设置其工艺参数)的子界面等。也就是说，通过该显示设备的“交互界面”，可实现用户与产品不良成因分析的系统的完全交互(控制和接收结果)。示例性的，用户可以通过交互界面提供用于分析设备703中算法设计的相关参数，用户还可以通过交互界面选择用于预测设备指标的优良性等级的模型(即，模型选择)等等。
在示例性的实施例中，通过训练后的预测模型对上述目标降维变量及其对应的第二待测参数值的优良性等级预测过程如下，即分析设备703中一个或多个处理器被具体配置为：将所述关于N个目标降维变量的第二待测参数值输入所述预测模型，分别计算所述N个目标降维变量对应的傅里叶基的特征向量；基于线性回归算法处理所述N个傅里叶基的特征向量，得到关于所述优良性等级的N’个目标筛选变量，其中，N’小于N；基于所述预测模型得到所述目标筛选变量的参数值分别属于各个优良性等级的预测概率；确定预测概率最大值对应的优良性等级为所述预测模型的输出值。
在示例性的实施例中，一种确定故障指标集合的具体实施方式如下：若预测模型的输出值为所述优良性等级低于预设等级值，则分析设备703中一个或多个处理器还被配置为：通过对上述目标筛选变量进行与上述降维处理对应的升维处理，可以获取该目标筛选变量相关的设备指标集合(如可以表示为RecipeX)。进一步地，将该设备指标集合RecipeX所包含的一组设备指标确定为故障设备指标集合。更进一步地，对RecipeX所包含的该组设备指标进行相关性分析(即图7所示出的相关性分析主题)，以确定不良产生的原因。示例性的，若产品的质量等级分为A级、B级、C级和D级，可以取B级作为上述预设等级值。从而，若预测模型的输出值为C级和D级，则获取上述目标筛选变量相关的设备指标集合M，以确定该设备指标集M合为故障设备指标集合。当然，上述预设等级值还可以是C级，那么，预测模型的输出值为D级时确定故障设备指标集合。也就是说，上述故障设备指标集合与事先划分的产品质量等级以及模型的预测值相关。
在示例性的实施例中，上述确定故障设备指标集合之后，进一步确定指标集合中各个设备指标分别对应的优良参数范围。具体的，由于故障设备指标集合中存在至少一个设备指标的当前参数值处于对应的优良参数范围之外，因此，可以根据该优良参数范围对故障设备指标的当前参数值进行调整，以使得存在故障的设备指标值及时恢复至对应的优良参数范围内。进而降低产品的不良率。
需要说明的是，由于分析设备703中一个或多个处理器进行上述数据处理过程与图4所示实施例相同，因此在此不再赘述。
在示例性的实施例中，分析设备703中一个或多个处理器还被配置为：基于所述预测模型确定待监控目标设备指标的优良参数值区间。其中，将上述M个目标设备指标中的任意一个中的任意一个设备指标作为本实施例中的待监控目标设备指标。具体的，根据训练后的预测模型确定待监控目标设备指标的优良参数值区间。参考图8，横坐标表示M个目标设备指标，纵坐标用来表示各个目标设备指标的优良参数值区间。从而，可以通过优良参数值区间来实时监控上述待监控目标设备指标的当前参数值。
参考图8，在示例性的实施例，对于待监控目标设备指标Q 1，实时获取Q 1的参数值。当待监控目标设备指标Q 1的参数值处于对应的优良参数值区间内时，说明对上述Q1的监控结果为良，也说明Q 1的当前参数值不会导致产生不良产品，本实施例中示例性的将Q 1的当前参数值显示为圆点。另外，另一目标设备指标Q 2的情况与上述待监控目标设备指标Q 1相同，对一目标设备指标Q 2不再赘述。另一示例性的实施例中，对于待监控目标设备指标Q M，实时获取Q M的参数值。当待监控目标设备指标Q M的参数值处于对应的优良参数值区间之外时，说明对上述Q M的监控结果为不良，也说明Q M的当前参数值会导致产生不良产品，本实施例中示例性的将Q M的当前参数值显示为三角形，以提醒工作人员及时采取改善措施。
可见，本技术方案通过大数据预测提供的整个生命周期过程中异常参数的履历，可以有针对性的确认设备参数对于产品的影响。并且，通过变量参数的趋势线实现实时监控，有利于及时确定参数值不良预警，降低产品的不良率。
参考图7，显示设备704具有显示功能，用于将分析设备计算得到分析结果(如相关性)显示出来。
在一些实施例中，显示设备还具有显示功能，用于将分析设备计算得到分析结果(如优良性等级等)显示出来。其中，显示设备可包括一个或 多个显示器，包括一个或多个具有显示功能的终端，从而分析设备可将其分析得到的相关性发送给显示设备，显示设备再将其显示出来。
本实施例中，显示设备704的展示界面如上述图8所示显示：具体的，显示关于上述待监控目标设备指标的优良参数值区间，其中，优良参数值区间指待监控目标设备指标的参数值不会导致产生不良产品的参数值范围；以及展示界面还显示实时获取的所述待监控目标设备指标的当前参数值，其中，当前参数值处于对应的优良参数值区间之内时，显示为圆点(可以根据实际需要进行改变，例如，可以显示为红色点)，当前参数值处于对应的优良参数值区间之外时，显示为三角点(可以根据实际需要进行改变，例如，可以显示为绿色点)。从而，将任一目标设备指标的优良参数值区间和当前参数值的形象的展示为工作人员，从而便于工作人员对上述第一维变量进行实施监控，以及对不良预警及时采取对应措施。
在示例性的实施例中，上述监控系统700还包括：报警设备。具体的，该报警设备被配置为：当上述待监控目标设备指标的当前参数值与上述优良参数值区间的边界值的差值小于预设值时，发出警报(如蜂鸣声、语音警报灯)。以提醒工作人员出现不良预警，进而便于工作人员及时对不良预警及时采取对应措施。
本公开提供的实施例中，首先以FDC系统数据作为产品优良性等级预测分析的大数据基础，进一步地，分布式存储设备可通过大数据方式高效率的实现对多个工厂设备的原始数据的收集和初步处理。一个或多个处理器(可布置于分析设备中)则可从分布式存储设备获取所需的数据。基于训练后的预测模型，可以确定M个目标设备指标分别的优良参数值区间，以实时监控每个目标设备指标，预测不良发生的可能性及报警功能，不良预测通过对所有参数进行管理能够发掘一些之前未受关注的重要参数给FDC系统，提高FDC系统的管控范围，进而较为准确地定位不良、调整生产流程等。并且不良预测可以自动生成模提供给FDC系统，帮助FDC系统合理确认相关设备指标的范围，从而节约人工耗时，能够尽早发现问题。在保证同类产品的生产时，设备状态都是最优的生产状态，批量性的提升同类产品生产的良率。
参考图9所示，本技术方案还提供一种计算机可读介质900，其可以采用便携式紧凑盘只读存储器(CD-ROM)并包括程序代码，并可以在终端设备，例如个人电脑上运行。然而，本公开的程序产品不限于此，在本文件中，可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
上述程序产品可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以为但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式盘、硬盘、随机存取存储器(Random Access Memory，RAM)、只读存储器(Read-Only Memory，ROM)、可擦式可编程只读存储器(erasable programmable read-only memory，EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了可读程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。可读信号介质还可以是可读存储介质以外的任何可读介质，该可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于无线、有线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言的任意组合来编写用于执行本公开操作的程序代码，上述程序设计语言包括面向对象的程序设计语言—诸如Java、C++等，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分 在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。在涉及远程计算设备的情形中，远程计算设备可以通过任意种类的网络，包括局域网(Local Area Network，LAN)或广域网(Wide Area Network，WAN)，连接到用户计算设备，或者，可以连接到外部计算设备(例如利用因特网服务提供商来通过因特网连接)。
此外，在本公开的示例性实施例中，还提供了一种能够实现上述方法的电子设备。
所属技术领域的技术人员能够理解，本公开的各个方面可以实现为系统、方法或程序产品。因此，本公开的各个方面可以具体实现为以下形式，即：完全的硬件实施方式、完全的软件实施方式(包括固件、微代码等)，或硬件和软件方面结合的实施方式，这里可以统称为“电路”、“模块”或“系统”。
下面参照图10来描述根据本公开的这种实施方式的电子设备1000。图10显示的电子设备1000仅仅是一个示例，不应对本公开实施例的功能和使用范围带来任何限制。
如图10所示，电子设备1000以通用计算设备的形式表现。电子设备1000的组件可以包括但不限于：上述至少一个处理单元1010、上述至少一个存储单元1020、连接不同系统组件(包括存储单元1020和处理单元1010)的总线1030。
其中，上述存储单元存储有程序代码，上述程序代码可以被上述处理单元1010执行，使得上述处理单元1010执行本说明书上述实施例中描述的方法步骤。
存储单元1020可以包括易失性存储单元形式的可读介质，例如：随机存取存储单元(Random Access Memory，RAM)10201和/或高速缓存存储单元10202，还可以进一步包括只读存储单元只读存储器(Read-Only Memory，ROM)10203。
存储单元1020还可以包括具有一组(至少一个)程序模块10205的程序/实用工具10204，这样的程序模块10205包括但不限于： 操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。
总线1030可以为表示几类总线结构中的一种或多种，包括存储单元总线或者存储单元控制器、外围总线、图形加速端口、处理单元或者使用多种总线结构中的任意总线结构的局域总线。
电子设备1000也可以与一个或多个外部设备1100(例如键盘、指向设备、蓝牙设备等)通信，还可与一个或者多个使得用户能与该电子设备1000交互的设备通信，和/或与使得该电子设备1000能与一个或多个其它计算设备进行通信的任何设备(例如路由器、调制解调器等等)通信。这种通信可以通过输入/输出(Input/Output，I/O)接口1050进行。进一步地，I/O接口1050与显示单元1040连接，以通过I/O接口1050将待显示内容传输至显示单元1040，以供用户查看。
并且，电子设备1000还可以通过网络适配器1060与一个或者多个网络(例如局域网(Local Area Network，LAN)，广域网(Wide Area Network，WAN)和/或公共网络，例如因特网)通信。如图所示，网络适配器1060通过总线1030与电子设备1000的其它模块通信。应当明白，尽管图中未示出，可以结合电子设备1000使用其它硬件和/或软件模块，包括但不限于：微代码、设备驱动器、冗余处理单元、外部磁盘驱动阵列、RAID系统、磁带驱动器以及数据备份存储系统等。
通过以上的实施方式的描述，本领域的技术人员易于理解，这里描述的示例实施方式可以通过软件实现，也可以通过软件结合必要的硬件的方式来实现。因此，根据本公开实施方式的技术方案可以以软件产品的形式体现出来，该软件产品可以存储在一个非易失性存储介质(可以是CD-ROM，U盘，移动硬盘等)中或网络上，包括若干指令以使得一台计算设备(可以是个人计算机、服务器、终端装置、或者网络设备等)执行根据本公开实施方式的方法。
此外，上述附图仅是根据本公开示例性实施例的方法所包括的处 理的示意性说明，而不是限制目的。易于理解，上述附图所示的处理并不表明或限制这些处理的时间顺序。另外，也易于理解，这些处理可以是例如在多个模块中同步或异步执行的。
本领域技术人员在考虑说明书及实践这里公开的内容后，将容易想到本公开的其他实施例。本申请旨在涵盖本公开的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本公开的一般性原理并包括本公开未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本公开的真正范围和精神由权利要求指出。

Claims (20)

  1. 一种设备指标优良性等级预测模型的训练方法，其中，所述方法包括：
    获取多个设备指标的历史参数值；
    根据所述多个设备指标之间的关联关系，对所述历史参数值进行降维处理，得到关于降维变量的样本数据；
    根据所述降维变量的样本数据的优良性等级，确定正样本集合和负样本集合；
    通过所述正样本集合和所述负样本集合训练机器学习模型，得到用于预测目标设备指标的优良性等级的预测模型。
  2. 根据权利要求1所述的方法，其中，所述根据所述多个设备指标之间的关联关系，对所述历史参数值进行降维处理，包括：
    通过主成分分析算法分析所述多个设备指标之间的关联关系；
    基于所述关联关系将相关联的设备指标进行组合操作，得到降维变量；
    根据所述组合操作涉及的设备指标的历史参数值，确定所述降维变量的样本数据。
  3. 根据权利要求1所述的方法，其中，所述通过所述正样本集合和所述负样本集合训练机器学习模型，包括：
    分别对所述正样本集合中和所述负样本集合中的降维变量进行采样，得到采样变量；
    计算每个所述采样变量的傅里叶基的特征向量，以确定所述采样变量之间的相关性；
    基于线性回归算法处理所述傅里叶基的特征向量，得到筛选变量；
    基于所述筛选变量和所述筛选变量对应的参数值训练机器学习模型。
  4. 根据权利要求1至3中任意一项所述的方法，其中，所述样本数据包括训练样本和测试样本；
    所述根据所述降维变量的样本数据的优良性等级，确定正样本集合 和负样本集合，包括：
    根据所述降维变量的样本数据中训练样本的优良性等级，确定正样本集合和负样本集合；
    所述方法还包括：通过所述测试样本评估所述预测模型。
  5. 根据权利要求4所述的方法，其中，通过以下评估指标中的一种或多种评估所述预测模型，所述评估指标包括：准确率、召回率和接收者操作特征曲线下面积AUC。
  6. 根据权利要求1至3中任意一项所述的方法，其中，所述获取每个所述设备指标的历史参数值，包括：
    将关于设备指标的源数据加载至数据仓库以进行数据挖掘分析；
    将所述挖掘分析之后的数据存储至非关系型数据库以供确定所述历史参数值。
  7. 根据权利要求6述的方法，其中，所述方法还包括：
    获取在数据对象中参数值为空值的待清理设备指标；
    获取在其他数据对象中待清理设备指标的属性值的众数值或平均值；
    确定所述众数值或所述平均值为所述数据对象中待清理设备指标的属性值。
  8. 根据权利要求1至3中任意一项所述的方法，其中，所述机器学习模型为贝叶斯算法模型、决策树算法模型或神经网络模型。
  9. 一种设备指标的监控系统，其中，所述系统包括：分布式存储设备，以及一个或多个处理器，其中，
    所述分布式存储设备，被配置为分别获取关于M个目标设备指标的第一待测参数值，其中，M为大于1的整数；
    所述一个或多个处理器，被配置为基于所述权利要求1至8中任意一项所述的预测模型监控所述M个目标设备指标的优良性。
  10. 根据权利要求9所述的系统，其中，所述一个或多个处理器还被具体配置为：
    根据所述M个目标设备指标之间的关联关系，对所述第一待测参数值进行降维处理，得到关于N个目标降维变量的第二待测参数值，其中， N小于M的正整数；
    将所述第二待测参数值输入所述预测模型，以通过所述预测模型的输出值监控所述M个目标设备指标的优良性。
  11. 根据权利要求10所述的系统，其中，所述一个或多个处理器被具体配置为：
    通过主成分分析算法分析所述M个目标设备指标之间的关联关系；
    基于所述关联关系将相关联的设备指标进行组合操作，得到N个目标降维变量；
    根据所述组合操作涉及的目标设备指标的第一待测参数，确定所述目标降维变量的第二待测参数。
  12. 根据权利要求11所述的系统，其中，所述一个或多个处理器被具体配置为：
    将所述关于N个目标降维变量的第二待测参数值输入所述预测模型，分别计算所述N个目标降维变量对应的傅里叶基的特征向量；
    基于线性回归算法处理所述N个傅里叶基的特征向量，得到关于所述优良性等级的N’个目标筛选变量，其中，N’小于N；
    基于所述预测模型得到所述目标筛选变量的参数值分别属于各个优良性等级的预测概率；
    确定预测概率最大值对应的优良性等级为所述预测模型的输出值。
  13. 根据权利要求12所述的系统，其中，所述一个或多个处理器还被配置为：
    若所述预测模型的输出值为所述优良性等级低于预设等级值，则对所述目标降维变量进行升维处理得到故障设备指标集合。
  14. 根据权利要求9所述的系统，其中，所述一个或多个处理器还被配置为：
    基于所述预测模型确定待监控目标设备指标的优良参数值区间，其中，所述待监控目标设备指标为所述M个目标设备指标中的任意一个；
    根据所述优良参数值区间实时监控所述待监控目标设备指标的当前参数值。
  15. 根据权利要求14所述的系统，其中，所述系统还包括：显示设备，被配置为：
    显示关于待监控目标设备指标的优良参数值区间，以及显示实时获取的所述待监控目标设备指标的当前参数值。
  16. 根据权利要求14所述的系统，其中，所述系统还包括：报警设备，被配置为：
    当所述待监控目标设备指标的当前参数值与所述优良参数值区间的边界值的差值小于预设值时，发出警报。
  17. 一种设备指标的监控方法，其中，所述方法包括：
    分别获取关于M个目标设备指标的第一待测参数值，其中，M为大于1的整数；
    基于所述权利要求1至8中任意一项所述的预测模型监控所述M个目标设备指标的优良性。
  18. 根据权利要求17所述的方法，其中，所述基于所述预测模型监控所述M个目标设备指标的优良性，包括：
    根据所述M个目标设备指标之间的关联关系，对所述第一待测参数值进行降维处理，得到关于N个目标降维变量的第二待测参数值，其中，N小于M的正整数；
    将所述第二待测参数值输入所述预测模型，以通过所述预测模型的输出值监控所述M个目标设备指标的优良性。
  19. 根据权利要求18所述的方法，其中，所述根据所述M个目标设备指标之间的关联关系，对所述第一待测参数值进行降维处理，包括：
    通过主成分分析算法分析所述M个目标设备指标之间的关联关系；
    基于所述关联关系将相关联的设备指标进行组合操作，得到N个目标降维变量；
    根据所述组合操作涉及的目标设备指标的第一待测参数，确定所述目标降维变量的第二待测参数。
  20. 一种计算机可读介质，其上存储有计算机程序，其中，所述程序被处理器执行时，实现如权利要求1至8中任意一项所述的设备指标 优良性等级预测模型的训练方法，或实现如权利要求18至20中任意一项所述的设备指标的监控方法。
PCT/CN2020/073779 2020-01-22 2020-01-22 设备指标优良性等级预测模型训练方法、监控系统和方法 WO2021146996A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202080000070.0A CN113614758A (zh) 2020-01-22 2020-01-22 设备指标优良性等级预测模型训练方法、监控系统和方法
PCT/CN2020/073779 WO2021146996A1 (zh) 2020-01-22 2020-01-22 设备指标优良性等级预测模型训练方法、监控系统和方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/073779 WO2021146996A1 (zh) 2020-01-22 2020-01-22 设备指标优良性等级预测模型训练方法、监控系统和方法

Publications (1)

Publication Number Publication Date
WO2021146996A1 true WO2021146996A1 (zh) 2021-07-29

Family

ID=76991967

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/073779 WO2021146996A1 (zh) 2020-01-22 2020-01-22 设备指标优良性等级预测模型训练方法、监控系统和方法

Country Status (2)

Country Link
CN (1) CN113614758A (zh)
WO (1) WO2021146996A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113673866A (zh) * 2021-08-20 2021-11-19 上海寻梦信息技术有限公司 农作物决策方法、模型训练方法以及相关设备
CN114418011A (zh) * 2022-01-21 2022-04-29 京东方科技集团股份有限公司 一种产品不良成因分析的方法、设备及系统、存储介质
CN114720582A (zh) * 2021-11-26 2022-07-08 韩山师范学院 一种不同陈化年份老香黄的综合评价方法
CN115293282A (zh) * 2022-08-18 2022-11-04 昆山润石智能科技有限公司 制程问题分析方法、设备及存储介质
CN115660513A (zh) * 2022-12-29 2023-01-31 四川省水利科学研究院 一种基于水利工程渡槽变形的监测方法及系统
CN116484269A (zh) * 2023-06-25 2023-07-25 深圳市彤兴电子有限公司 显示屏模组的参数处理方法、装置、设备及存储介质
CN118138749A (zh) * 2024-04-30 2024-06-04 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） 一种多维度信息系统性能测试与分析工具及方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113538020B (zh) * 2021-07-05 2024-03-26 深圳索信达数据技术有限公司 获取客群特征关联度方法、装置、存储介质和电子装置
WO2023130304A1 (zh) * 2022-01-06 2023-07-13 京东方科技集团股份有限公司 数据处理方法、系统、计算机可读存储介质
CN116629707B (zh) * 2023-07-20 2023-10-20 合肥喆塔科技有限公司 基于分布式并行计算的fdc溯因分析方法及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063747A (zh) * 2014-06-26 2014-09-24 上海交通大学 一种分布式系统中的性能异常预测方法及系统
CN108022001A (zh) * 2017-09-20 2018-05-11 河海大学 基于pca和分位数回归森林的短期负荷概率密度预测方法
CN109165160A (zh) * 2018-08-28 2019-01-08 北京理工大学 基于核主成分分析算法的软件缺陷预测模型设计方法
CN109492193A (zh) * 2018-12-28 2019-03-19 同济大学 基于深度机器学习模型的异常网络数据生成与预测方法
US10424388B2 (en) * 2016-10-18 2019-09-24 Samsung Electronics Co., Ltd. Storage device, memory system, and read voltage decision method thereof

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107391760B (zh) * 2017-08-25 2018-05-25 平安科技（深圳）有限公司 用户兴趣识别方法、装置及计算机可读存储介质
CN107704495B (zh) * 2017-08-25 2018-08-10 平安科技（深圳）有限公司 主题分类器的训练方法、装置及计算机可读存储介质
CN108805202B (zh) * 2018-06-11 2021-07-30 上海新增鼎数据科技有限公司 一种用于电解槽故障预警的机器学习方法及其应用
CN110674842A (zh) * 2019-08-26 2020-01-10 明阳智慧能源集团股份公司 一种风电机组主轴轴承故障预测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063747A (zh) * 2014-06-26 2014-09-24 上海交通大学 一种分布式系统中的性能异常预测方法及系统
US10424388B2 (en) * 2016-10-18 2019-09-24 Samsung Electronics Co., Ltd. Storage device, memory system, and read voltage decision method thereof
CN108022001A (zh) * 2017-09-20 2018-05-11 河海大学 基于pca和分位数回归森林的短期负荷概率密度预测方法
CN109165160A (zh) * 2018-08-28 2019-01-08 北京理工大学 基于核主成分分析算法的软件缺陷预测模型设计方法
CN109492193A (zh) * 2018-12-28 2019-03-19 同济大学 基于深度机器学习模型的异常网络数据生成与预测方法

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113673866A (zh) * 2021-08-20 2021-11-19 上海寻梦信息技术有限公司 农作物决策方法、模型训练方法以及相关设备
CN114720582A (zh) * 2021-11-26 2022-07-08 韩山师范学院 一种不同陈化年份老香黄的综合评价方法
CN114720582B (zh) * 2021-11-26 2023-10-20 韩山师范学院 一种不同陈化年份老香黄的综合评价方法
CN114418011A (zh) * 2022-01-21 2022-04-29 京东方科技集团股份有限公司 一种产品不良成因分析的方法、设备及系统、存储介质
CN115293282A (zh) * 2022-08-18 2022-11-04 昆山润石智能科技有限公司 制程问题分析方法、设备及存储介质
CN115293282B (zh) * 2022-08-18 2023-08-29 昆山润石智能科技有限公司 制程问题分析方法、设备及存储介质
CN115660513A (zh) * 2022-12-29 2023-01-31 四川省水利科学研究院 一种基于水利工程渡槽变形的监测方法及系统
CN116484269A (zh) * 2023-06-25 2023-07-25 深圳市彤兴电子有限公司 显示屏模组的参数处理方法、装置、设备及存储介质
CN116484269B (zh) * 2023-06-25 2023-09-01 深圳市彤兴电子有限公司 显示屏模组的参数处理方法、装置、设备及存储介质
CN118138749A (zh) * 2024-04-30 2024-06-04 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） 一种多维度信息系统性能测试与分析工具及方法

Also Published As

Publication number Publication date
CN113614758A (zh) 2021-11-05

Similar Documents

Publication Publication Date Title
WO2021146996A1 (zh) 设备指标优良性等级预测模型训练方法、监控系统和方法
US20140006338A1 (en) Big data analytics system
US11409962B2 (en) System and method for automated insight curation and alerting
US7805639B2 (en) Tool to report the status and drill-down of an application in an automated manufacturing environment
CN117176560B (zh) 一种基于物联网的监测设备监管系统及其方法
CN117393076B (zh) 耐热环氧树脂生产过程智能监测方法及系统
US20230075005A1 (en) Intelligent asset anomaly prediction via feature generation
CN113360722B (zh) 一种基于多维数据图谱的故障根因定位方法及系统
US10120372B2 (en) Event processing based system for manufacturing yield improvement
US20210364999A1 (en) System and method for analyzing cause of product defect, computer readable medium
CN110806730A (zh) 大数据运维平台、服务器及存储介质
WO2021142622A1 (zh) 确定不良原因的方法、电子设备、存储介质及系统
US10810072B2 (en) Systems and methods for monitoring input data and model scores in analytical models for IOT (internet of things) devices
CN113408659A (zh) 一种基于数据挖掘的建筑能耗集成分析方法
CN114550336B (zh) 设备巡检方法、装置、计算机设备和存储介质
CN116955434A (zh) 一种工业装备全生命周期管理及多维度能效分析系统
US20220374402A1 (en) Contextualized time series database and/or multi-tenant server system deployment
CN116992346A (zh) 一种基于人工智能大数据分析的企业生产数据处理系统
CN117670378A (zh) 一种基于大数据的食品安全监控方法及系统
EP3745321A1 (en) An operating envelope recommendation system with guaranteed probabilistic coverage
US11727021B2 (en) Process control tool for processing big and wide data
WO2023273461A1 (zh) 机器人运营状态监控系统及方法
CN109740941A (zh) 军品数据处理系统
CN114625763A (zh) 用于数据库的信息分析方法、装置、电子设备和可读介质
Tan et al. Research on Fault Prediction Model Based on 5G Data Center

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20915866

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20915866

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20915866

Country of ref document: EP

Kind code of ref document: A1