Machine tools are the major consumers of industrial energy, but their energy efficiency remains low, posing a serious challenge to sustainable manufacturing. The current literature predominantly focuses on isolated subsystems or specific operational phases (e.g., cutting parameters), lacking systematic evaluations of how different methodologies interact within the Life Cycle Assessment (LCA) framework. This paper provides a critical synthesis of three core methodologies—modeling methods, system parameter optimization, and machine learning (ML)—across the design/production, usage, and recycling stages. Unlike descriptive reviews, this study highlights the scientific contribution by defining the applicability boundaries and complementary mechanisms of these approaches. The analysis reveals that while modeling lays the theoretical basis for eco-design and remanufacturing assessments, and optimization effectively resolves multi-objective trade-offs, these static methods struggle with the dynamic complexity of real-time operations where ML excels. However, ML is identified to be constrained by high data dependency and poor generalization in heterogeneous environments. Consequently, this review shows that the ‘cross-application’ of modeling methods and machine learning to construct hybrid models is essential for addressing complex nonlinear relationships and achieving accurate energy prediction throughout the entire life cycle. Finally, future directions such as transfer learning and digital twins are proposed to overcome current generalization bottlenecks, providing a theoretical foundation for the industry’s transition from passive energy assessment to active, intelligent energy management.
Ma et al. (Sat,) studied this question.