Enhancing industrial scheduling through machine learning: A synergistic approach with predictive modeling and clustering

Abstract

In this study, a novel solution framework is developed integrating Machine Learning (ML) techniques with Mixed-Integer Linear Programming (MILP) to address the optimization of production scheduling in manufacturing industries characterized by multiple products, shared resources and parallel production lines. The dynamic nature of these industries often requires rapid schedule adjustments within minutes to address unexpected events. At the same time, manufacturing facilities face frequent changeover operations to prevent cross-contamination. Inefficient product allocation to packing lines often leads to significant downtime and material waste, especially when introducing new products with unrecorded changeover times. To overcome these challenges, the proposed framework first compiles a representation space in which distances correspond to changeover times. This enables the employment of constrained clustering to group production orders according to the available packing lines, minimizing changeover times within each cluster. Then, the derived allocation is used to restrict the solution space of an MILP-based scheduling model to reduce its computational complexity. Furthermore, to tackle the issue of unavailable changeover data, a predictive ML model is trained to predict unknown changeover times for new or existing products. An evaluation study based on a construction materials plant is conducted to test the applicability of the framework. It is concluded that the proposed approach achieves accurate solutions rapidly, reduces downtime and facilitates the smooth integration of new products into the production process. In addition, it is applicable to a wide range of industries, and it enables the extension to an online scheduling framework due to its computational speed.