Prozessführungskomponenten für die Integration von Machine Learning Modellen in die operative Prozessführung

Grothoff, Julian Alexander; Epple, Ulrich (Thesis advisor); Abel, Dirk (Thesis advisor); Kleinert, Tobias Theodor (Thesis advisor)

Als Manuskript gedruckt. - Düsseldorf : VDI Verlag GmbH (2022)
Book, Dissertation / PhD Thesis

In: Fortschrittberichte VDI : Reihe 8, Meß-, Steuerungs- und Regelungstechnik 1277
Page(s)/Article-Nr.: XI, 164 Seiten : Illustrationen, Diagramme

Dissertation, RWTH Aachen University, 2022

Abstract

A rising number of application examples utilizing machine learning appear in development and research environments. Some of the trained models are build for the direct feed-forward or feed-back control of technical processes. However, challenges arise when integrating them into industrial systems. In most cases, solutions are individually adapted to the process control system and are therefore not interoperable. This thesis describes the approach of encapsulating solutions as services of standardized components for process control. A design pattern shows how the challenges are addressed, enabling stepwise integration as well as uniform orchestration. For validation, a simulation of a transport process was created in Unity and a neural network was trained using PyTorch. Subsequently, the solution was integrated into the process control system ACPLT/RTE. This work is therefore aimed equally at engineers and scientists in the fields of automation and artificial intelligence.

Institutions

• Division of Materials Science and Engineering [520000]
• Chair of Information and Automation Systems for Process and Material Technology [526610]