Increased environmental awareness is driving the manufacturing industry towards novel ways of energy reduction to become sustainable yet stay competitive. Climate and enviromental challenges put high priority on incorporating aspects of sustainability into both strategic and operational levels, such as production scheduling, in the manufacturing industry. Considering energy as a parameter when planning makes an already existing highly complex task of production scheduling even more multifaceted. The focus in this study is on inverse scheduling, defined as the problem of finding the number of jobs and duration times to meet a fixed input capacity. The purpose of this study was to investigate how scheduling can be formulated, within the environment of discrete event simulation coupled with reinforcement learning, to meet production demands while simultaneously minimize makespan and reduce energy. The study applied the method of modeling a production robot cell with its uncertainties, using discrete event simulation combined with deep reinforcement learning and trained agents. The researched scheduling approach derived solutions that take into consideration the performance measures of energy use. The method was applied and tested in a simulation environment with data from a real robot production cell. The study revealed opportunities for novel approaches of studying and reducing energy in the manufacturing industry. Results demonstrated a move towards a more holistic approach for production scheduling, which includes energy usage, that can aid decision-making and facilitate increased sustainability in production. We propose a conceptual framework for scheduling for minimizing energy use applying discrete event simulation as an environment for deep reinforcement learning.
The work was carried out at the Production Technology Centre at University West, Sweden supported by the Swedish Governmental Agency for Innovation Systems (Vinnova) under the project SmoothIT and by the KK Foundation under the project Artificial and Human Intelligence through Learning (AHIL). Their support is gratefully acknowledged. Assistance provided by Master's students Maria Vincenta Vivo and Mohammadali Zakeriharandi was greatly appreciated.