Machine learning-assisted life cycle assessment for sustainable and environmentally responsible manufacturing

Abstract

Abstract. Life Cycle Assessment (LCA) represents the de facto standard of measuring the environmental footprint of manufactured products throughout their full value chain, but its usability in industrial decision-making is hampered by its prohibitive computational demands, the large amount of inventory data required, and its lack of real-time process-level capabilities. This paper proposes a comprehensive Machine Learning-Assisted LCA (ML-LCA) framework that addresses these barriers through five tightly integrated components: (1) an eXtreme Gradient Boosting (XGBoost) regressor for rapid Global Warming Potential (GWP) prediction achieving R² = 0.974 and RMSE = 3.8 kg CO₂-eq per functional unit; (2) a Long Short-Term Memory (LSTM) network for dynamic Life Cycle Inventory (LCI) modeling that captures temporal manufacturing variability; (3) SHapley Additive exPlanations (SHAP) for interpretable identification of environmental hotspots; (4) a Non-dominated Sorting Genetic Algorithm III (NSGA-III) multi-objective eco-design optimizer; and (5) an ISO 14040/44-compliant uncertainty quantification module achieving 94% confidence interval coverage. When applied to the case study of aluminum alloy precision machining as a representative gate-to-gate case study, the proposed framework is able to cut down GWP by 27.6, acidification potential by 31.9 and eutrophication potential by 24.7 compared to conventional process baselines and compresses LCA computation time down to less than 0.02 seconds per functional unit. The findings make ML-LCA a technically sound, industrially viable roadmap to a manufacturing process that is environmentally responsible.