Approach of Automated ML Algorithm Selection for the Realization of Intelligent Production

For achieving the ambitious objectives of intelligent production, the artificial intelligence through its machine learning algorithms represents one of the most promising technology. The employment of machine learning algorithms for the optimization of complex production processes faces the big challenge of selecting the suitable machine learning method which fits the optimization parameter objectives. This paper introduces our approach for automated selection of ML algorithms to be used for optimization of a specific production process. The approach and the corresponding method consists of the following main blocks: (a) production process definition, (b) ML performance, (c) selector constructor and (d) assessment and incremental improvement of selector performance. The first component defines a typical production process or domain based on a well-established set of features, e.g. product quality inspection through process features as accuracy, material characteristics, etc. The ML performance construct contains precise defined performance of well clustered ML algorithms based on established benchmarking. The third construct, the Selector, automatically realizes a perfect mapping between the production process features and the performance of the ML algorithms. The logic of this automated selection represents the innovation of our work. The last component assesses the selector algorithm based on a set of specific KPIs for each production domain or process. The incremental improvement of the selector is approached as well, closing the loop between all method components. The developed approach and method have as foundations our work on identifying critical production processes/domains as core of realizing the intelligent production and laborious developed collection of ML algorithms, based on their performance data. These foundations and a motivation scenario are presented inside our paper to highlight the relevance of our research work.