Automated machine learning for predictive quality in production

作     者：Jonathan Krauß Bruno Machado Pacheco Hanno Maximilian Zang Robert Heinrich Schmitt 

作者机构：Fraunhofer Institute for Production Technology IPT Steinbachstr. 17 Aachen 52074 Germany Laboratory for Machine Tools WZL RWTH Aachen University Steinbachstr. 19 Aachen 52074 Germany 

出 版 物：《Procedia CIRP》 (CIRP会议集)

年 卷 期：2020年第93卷

页      面：443-448页

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 

主　　题：Predictive Quality Machine Learning Data Science Automated ML AutoML Benchmarking Artificial Intelligence Data Integration Data Preprocessing Hyperparameter Tuning 

摘      要：Applications that leverage the benefits of applying machine learning (ML) in production have been successfully realized. A fundamental hurdle to scale ML-based projects is the necessity of expertise from manufacturing and data science. One possible solution lies in automating the ML pipeline: integration, preparation, modeling and model deployment. This paper shows the possibilities and limits of applying AutoML in production, including a benchmarking of available systems. Furthermore, AutoML is compared to manual implementation in a predictive quality use case: AutoML still requires programming knowledge and is outperformed by manual implementation - but sufficient results are available in a shorter timespan.