Integration of modeling and solution approaches for flow shop scheduling in the chemical industry

The chemical industry, positioned at the beginning of the production chain, plays a fundamental role in the global economy. Optimizing chemical processes is crucial for improving both economic efficiency and sustainability. However, process optimization, particularly for batch processes in multipurpose or multiproduct flow shops, presents significant challenges due to constraints such as limited secondary resources, intermediate storage, sequence-dependent changeovers, and lot-sizing decisions. These interdependent factors make scheduling an NP-hard problem, necessitating advanced modeling and solution approaches. This dissertation contributes to this field by developing new mixed-integer linear programming (MILP) models and integrating discrete-event simulation (DES) techniques to enhance scheduling accuracy and computational efficiency.

First, a MILP model for hybrid flow shops is introduced using a time-bucket formulation to incorporate real-world constraints effectively. Applied to an agrochemical production case study, this model demonstrates a 17% reduction in machine hours and improved on-time delivery. However, computational experiments reveal large optimality gaps due to loose lower bounds. To address this, Chapter 3 enhances MILP formulations by introducing strengthening inequalities that tighten linear programming relaxations, leading to improved lower bounds and faster solutions.

To further refine scheduling, Chapter 4 integrates MILP with DES using a Benders decomposition approach. This hybrid algorithm outperforms genetic algorithms in terms of solution time and computational efficiency and proves effective in pharmaceutical production scheduling. Chapter 5 builds on this by refining the Benders-DES algorithm through three modifications—warm-up phases, smoothening cuts, and re-initialization—all of which enhance convergence speed and solution robustness.

Finally, Chapter 6 extends the methodology to distributed flow shops by integrating MILP, DES, and neural network models. Despite a remaining optimality gap of 29%, this approach offers a promising framework for optimizing heterogeneous production environments in the chemical industry.

Overall, this dissertation advances process scheduling methodologies by improving MILP models and bridging them with DES techniques. The proposed approaches provide innovative workflows that enhance scheduling efficiency while reducing computational complexity, offering practical solutions for industrial applications.