全栈式机器学习在钢铁流程智能制造中的应用

Application of full stack machine learning in intelligent manufacturing of steel process

通过实际生产的海量数据与产品质量问题解决方案的积累,针对全流程数据分析方法与思想进行梳理、研发全栈式机器学习平台与模型,形成了一套具有鲁棒性和高精度的钢铁生产全流程多源异构数据分析框架,在钢铁流程生产中进行了应用。提出了工业数据分析的通用方法,针对质量问题的分析流程为:数据获取、数据治理、特征工程、模型选取与应用、问题解决。应用全栈式机器学习平台搜集数据,并应用平台中的算法库进行模型建立,基于mRMR算法进行变量挑选,应用FDA进行特征降维,并应用XGBOOST对数据进行分类,实现了质量问题溯源与监控,解决了线材生产中椭圆度超差的质量问题。

In this paper,through the accumulation of mass data and solutions to product quality problems in actual production,combed the whole process data analysis methods and ideas,developed a full stack machine learning platform and model,and formed a robust and high-precision multi-source heterogeneous data analysis framework for the whole process of steel production.It has been applied in steel process.This paper puts forward a general method of industrial data analysis,aiming at quality problems,the analysis process is:data acquisition,data governance,feature engineering,model selection and application,problem solving.The data are collected by the full stack machine learning platform,the model is established by the algorithm library in the platform,the variable is selected based on mRMR algorithm,the feature dimension is reduced by FDA,and the data is classified by XGBOOST.The quality problem traceability and monitoring are realized,and the quality problem of out of tolerance ellipticity in wire production is solved.