Swarm algorithms can improve production planning and scheduling

Industrial companies face an enormous challenge when it comes to the highly interconnected nature of their production facilities. These exhibit complex and dynamic behaviour, as can be observed in ants, bees, fish or birds. Inspired by models found in nature, the SWILT project models entire industrial plants as swarms. Three years on, we now have the results of the project: The simulations revealed that the overall performance of a large production planning system can be improved by a percentage in the single digits, which can represent significant financial gains for businesses.

At the University of Klagenfurt, October 2018 saw the launch of the research project SWILT (Swarm Intelligence Layer to Control Autonomous Agents). Four Carinthian partners (Lakeside Labs GmbH, the University of Klagenfurt, Infineon Technologies Austria AG and Novunex GmbH) conducted research on cyber-physical system swarms to connect the real world with the virtual world. Funded by the Austrian Research Promotion Agency (FFG), the project ran for three years and involved a project volume of just under 1 million euros.

“The goal was to improve production planning in an industrial use case by means of swarm algorithms”, Wilfried Elmenreich, the local project leader at the Department of Networked and Embedded Systems explains. He goes on: “In one case study, we modelled production planning in the semiconductor industry as a swarm problem. We applied several nature-inspired swarm algorithms to the problem, including an artificial hormone system, an artificial bee colony algorithm, an algorithm inspired by the foraging behaviour of ants, and an algorithm based on the mechanisms of the social amoebae known as slime moulds. The algorithms were evaluated through simulation in Netlogo, a programmable multi-agent modelling environment particularly suited to swarm experiments.”

The result speaks for itself: The simulations indicate that the overall performance of a large production planning and scheduling system can be improved by a percentage in the single digits, which translates into a significant financial gain for a large factory. Furthermore, the proposed algorithms have the advantage that they require fewer resources than other approaches and can therefore also be used in systems where many units need to be calculated during production.