Approach to adapt manufacturing process parameters systematically based on machine learning algorithms ?

摘要

The market environment for manufacturers of high precision products is highly turbulent. On the one hand, international low-wages competitors from Asian countries pose a challenge from a monetary point of view. On the other hand, customers expect individual solutions within short innovation cycles. Both facts cause manufacturers to be flexible regarding customer requirements very briefly. Opportunities for action are changes concerning product design and/or the adaptation of manufacturing process parameters. This paper shows an approach for an optimised adaptation of manufacturing process parameters based on machine learning algorithms to require customer demands. The goal is to determine parameter settings that enable production of products in a required target area regarding product characteristics like the geometry or the surface topography. The essential steps of the approach are as follows: 1) Definition of product specification and specification areas, 2) Manufacturing of samples with varied process parameters based on design of experiments (DoE), 3) Application of machine learning algorithms, 4) Detection and interpretation of rules and relations, 5) Filtering of reasonable relations between input and output parameter sets, 6) Classification of parameter sets, which leads based on detected rules - to product variants. The shown research work includes the case study ”cutlery grinding process”. A modified product with adapted surface characteristics (e.g. roughness and gloss) is required. The new, adapted product characteristics are matched with specification areas of the origin product specification. Based on machine learning rules, the grinding parameter sets are detectable, which lead to the required, new product characteristics.