Home // INTELLI 2016, The Fifth International Conference on Intelligent Systems and Applications // View article

Concept for Finding Process Models for New Classes of Industrial Production Processes

Authors:
Norbert Link
Jürgen Pollak
Alireza Sarveniazi

Keywords: machine learning; data modeling; hyper-model; process model; welding.

Abstract:
The required knowledge about relations between quantities governing the control and quality estimation of production processes is represented in so-called process models. Such models may relate process parameters and process goals allowing to find appropriate parameter values for given goals. Other models allow the derivation of the process state from observable quantities. Controls based on Markov Decision Processes require a state transition model and a cost function model of subsequent states. The functional relationships between the quantities of a model are usually represented by a dedicated combination of some base functions with given, fixed parameter values. In many cases, this is a linear combination of Kernel functions, where the parameters are determined by fitting known experimental data, such as in Support Vector Regression methods. The process models always refer only to a dedicated process class with given conditions (e.g., parts materials and geometries or machine properties). There are model populations in most industrial process domains, such as laser metal sheet welding, representing several metal alloys in combination with sheet thicknesses and welding equipment. In this paper, we propose novel methods on how to make use of this already existing model knowledge, which is used for the derivation of models of new process classes in the same process domain. For this purpose, the formation of a common model representation is derived from the individual models of the domain. The parameters of the individual models in this common representation form a model space, in which a model of the models can be formed: the hyper-model. General ideas of hyper-model formation are presented and approaches are discussed how dedicated models for specific, new process classes (e.g., with different conditions) can be derived from it.

Pages: 154 to 157

Copyright: Copyright (c) IARIA, 2016

Publication date: November 13, 2016

Published in: conference

ISSN: 2308-4065

ISBN: 978-1-61208-518-0

Location: Barcelona, Spain

Dates: from November 13, 2016 to November 17, 2016