A generalizable bilevel reinforcement learning model to solve large-scale unrelated parallel machine scheduling problem with sequence-dependent setups in real-time

Our research addresses the challenges in solving large-scale parallel machine scheduling, an important combinatorial optimization problem in computer science and operations research. With applications ranging from manufacturing to healthcare and supercomputing, our goal is to provide a real-time solution for instances exceeding 1,000 jobs. In this research, we explore the application of parallel machine scheduling to solve production scheduling in the manufacturing industry. Our innovative approach involves a two-level decomposition using two encoder-decoder deep neural networks trained through reinforcement learning. This model ensures scalability by training on small-scale instances and generalizing the learned policies for large-scale instances. The project’s impact is profound for manufacturing industries, promising optimal schedules within short timeframes. The anticipated outcome of this research for manufacturing industries is the achievement of superior production schedules, which can directly translate into a substantial increase in revenue, significant time savings in the preparation of future schedules, and an annual reduction in carbon emissions due to more efficient energy consumption. Furthermore, the development of a real-time production scheduling solution will contribute to the realization of agile manufacturing, empowering quick responses to unforeseen events such as machine breakdowns, job reworks, and urgent orders.

Faculty Supervisor:

Homayoun Najjaran

Student:

Partner:

Universität Bielefeld

Discipline:

Engineering

Sector:

Artificial Intelligence; Advanced Manufacturing; Other

University:

University of Victoria

Program: