Data-driven advisory system for industrial manufacturing. Application to the aluminum extrusion process

Abstract

The article presents a methodology for developing systems supporting optimal decision-making in the selection of manufacturing process parameters and tooling, from the point of view of minimizing the tendency for the occurrence of product defects. The developed advisory system is based on two essential pillars: searching a database extracted from the company's database and a model obtained from the same data; the latter is used to extract the working database from the original enterprise database and set up the search algorithm. The paper first discusses the characteristics of typical data recorded in a manufacturing process and presents the data used in the present work. Next, various general concepts of building an advisory system are presented, and the choice of a system that combines a database approach with modeling based on machine learning is justified. The fundamentals of the system dedicated to the aluminum extrusion process are presented, followed by the system implementation, including examples of its functioning. The developed system not only allows decisions to be made based on recorded cases and hidden dependencies detected by models in the production process but also helps identify the causes of problems. It can also help to draw attention to critical parameters of a product from the point of view of its manufacturing.