AI-BASED ENERGY-AWARE SCHEDULING AND PROCESS OPTIMIZATION IN ENGINEER-TO-ORDER SMART MANUFACTURING SYSTEMS

Abstract

Engineer-to-Order (ETO) manufacturing represents the most complex production paradigm, characterized by high customization, variable process flows, and deep uncertainty in order specifications. In the era of smart manufacturing, the integration of Artificial Intelligence (AI) for scheduling and process optimization offers a transformative pathway to balance productivity with energy sustainability. This paper proposes a novel framework for Energy-Aware Scheduling (EAS) in ETO environments using a hybrid AI architecture combining Deep Reinforcement Learning (DRL) with a Graph Neural Network (GNN) for state representation. Unlike traditional batch-focused models, our approach accommodates the non-repetitive, project-driven nature of ETO systems. A multi-objective reward function is designed to minimize makespan, reduce total energy consumption (kWh/unit), and flatten peak demand. Simulation results on a realistic ETO smart factory testbed (e.g., custom heavy machinery assembly) demonstrate a 22.4% reduction in energy costs, a 15.7% improvement in schedule adherence, and a 30% decrease in peak power overload events compared to standard heuristic-based scheduling. This research validates that AI-driven energy awareness can be embedded into real-time production control without compromising the flexibility required for ETO manufacturing.