We consider flexible flow shop scheduling problems with sequence-dependent setup times and uncertain processing and setup times. Traditional paradigms, such as predictive solution approaches, generate static schedules. Dynamic scheduling, in contrast, adapts decisions based on the current state, o"ering flexibility, but has historically been limited by relying on dispatching rules or simple heuristics. This study re-evaluates dynamic scheduling, focusing on its advantages over static plans utilizing state-of-the-art algorithms. We model the dynamic scheduling problem as a Markov decision process. Due to the curse of dimensionality, exact solution methods are intractable. We propose using policy function approximation methods like Genetic Programming and Deep Reinforcement Learning. We compare the performance of these dynamic scheduling approaches against static schedules generated by Constraint Programming. Through extensive sensitivity analysis, we explore the impact of uncertainty levels, the number of parallel machines, the di"erences in parallel machines, and the ratio between setup and processing times on the benefits of dynamic scheduling. Our results indicate that problems with high uncertainty favor dynamic scheduling, while these benefits diminish with an increasing setup ratio or increasing di"erences between parallel machines.     «

We consider flexible flow shop scheduling problems with sequence-dependent setup times and uncertain processing and setup times. Traditional paradigms, such as predictive solution approaches, generate static schedules. Dynamic scheduling, in contrast, adapts decisions based on the current state, o"ering flexibility, but has historically been limited by relying on dispatching rules or simple heuristics. This study re-evaluates dynamic scheduling, focusing on its advantages over static plan...     »