Multi-Objective Optimization of Smart Sustainable Manufacturing Using Hybrid Machine Learning and Metaheuristic Algorithms: A Comparative Computational Study

Abstract

Manufacturers must increasingly optimize throughput, energy, emissions, quality, reliability, and worker-centered constraints at the same time. Machine learning (ML) predicts complex system behavior, while metaheuristics search large, discontinuous, and multi-objective decision spaces; their hybridization can therefore connect perception with action. This article synthesizes exactly 20 recent and authoritative sources - 11 peer-reviewed studies, five institutional or standards-based sources, and four official company cases - and develops a reproducible analytical demonstrator. The computational study represents a synthetic cell with 30 jobs, six heterogeneous machines, three eligible machines per job, three operating modes, time-varying carbon intensity, and machine-health effects. A scalar genetic algorithm (GA), NSGA-II, and an Extra-Trees-assisted NSGA-II were compared over eight independent runs under an equal budget of 396 exact simulator evaluations. The hybrid screened four times as many candidate offspring while retaining 15% random exploration and exactly re-evaluating every selected candidate. Its mean weighted compromise score was 0.977, compared with 0.990 for plain NSGA-II and 0.989 for GA; the paired hybrid-versus-NSGA-II score difference was significant (Wilcoxon p = 0.039). Relative to energy-aware list dispatch, the selected hybrid operating point reduced electricity by 6.13%, modeled CO2 by 5.42%, and quality/reliability risk by 10.47%, while increasing risk-adjusted makespan by 5.10%. These are model results, not plant claims. The evidence and demonstrator jointly indicate that hybrid methods are most valuable when evaluations are expensive, objectives conflict, and decisions must adapt to changing production conditions. Their responsible industrial use requires exact feasibility checks, model-drift monitoring, lifecycle-aware sustainability accounting, human approval for consequential actions, and governance aligned with manufacturing standards.