Integration of improved meta-heuristic and machine learning for optimizing energy efficiency in additive manufacturing process

Additive manufacturing (AM) has been a vital element of smart manufacturing. The high energy intensity or environmental sustainability issue of AM, however, has posed a great challenge to the future massive application, particularly laser-based direct energy deposition (L-DED). This study aims to determine the optimal processing parameters for energy-saving without compromising the geometrical appearance. A quantification model for energy efficiency at the process level was established, with two energy efficiency indicators of AM process. Then, a meta-heuristic Mayfly algorithm, augmented with Bayesian technique and mutation strategies, was proposed to improve hyperparameters in a typical machine learning model (XGBoost). Based on the full factorial L-DED experiments, this study compared the improved XGBoost with four types of XGBoost derivatives via four algorithm evaluation metrics. Non-dominated sorting genetic algorithm II was adopted to optimize the processing parameters subject to the constraints of geometrical appearance. Results indicated that the proposed algorithm outperformed other XGBoost derivatives in terms of prediction accuracy and convergence rate. The energy efficiency could be improved by 76.35 J/g or 6.78 % on average while ensuring the geometry of the deposited layers. This study could help enhance energy-efficient additive manufacturing via proper processing parameters selection and facilitate the sustainability in AM domain.