SENECA

A Self-Learning Decision Support System for Order Sequencing and Machine Allocation Planning
(BMBF, 2020-2022)

The research project SENECA pursued the development of a self-learning decision support system for real-time order sequencing and machine scheduling. The research question was how machine learning (ML) methods could be applied in order to calculate admissible solutions with sufficient quality for order sequencing and machine allocation problems in real time. Various ML methods were investigated with regard to their applicability for order sequencing and machine allocation planning. Due to the highly dynamic nature of modern production systems and the resulting planning uncertainty, it was expected that production sequence planning in particular would benefit from ML-based, real-time capable and adaptive decision support systems. ML algorithms were so far primarily used for regression and classification problems. Their direct use for calculating optimization problems had hardly been researched and industrial applications were not yet known. The technical objective of the project was to develop a software and hardware prototype that supports decision-makers in production planning and control. The technical challenges related in particular to aspects of production and application-specific design. On the one hand, a high level of user-friendliness was important. This implied, among other things, that humans are always the final decision-making authority. The system had be able to continuously improve itself with human expertise. On the other hand, the assistance system needed to be designed in such a way that the real-time capability of the solution processes was fully utilized. Proposed order sequences and machine assignments needed to be transferable from production planning to production control at short notice.

Having identified suitable machine learning methods for solving the optimization problem of order sequencing and machine scheduling planning, three algorithms were selected for implementation in a demonstrator. A server was set up, in which both the trained AI models and a genetic algorithm — as a comparative benchmark — were implemented. A demonstrator was developed in the form of a client with a graphical user interface. This demonstrator was finally integrated into a real production environment at TECTRON WORBIS GmbH and tested there.

official project website

final project report

Project Partners