Ensemble transfer learning for cutting energy consumption prediction of aviation parts towards green manufacturing

Cutting energy consumption prediction gives decision supports for the energy-saving operation to realise green manufacturing. However, there are challenges when predicting aviation parts due to the machining features causing tool wear and expensive data labelling. Consequently, this paper builds a prediction model and emphasises training with limited experimental data by proposing an ensemble transfer learning approach. The approach incorporates transfer learning, i.e., TrAdaBoost-R2 (TR) algorithm, and calibration, i.e., Bayesian and Markov chain Monte Carlo calibration (MCMC). Firstly, a cutting energy consumption prediction model considering tool wear is formulated with cutting and tool parameters as the inputs. Secondly, a dataset including experiment and simulation data for training is constructed, where TR is used to identify the valuable data from the simulation model calibrated by MCMC. Then random forest regression (RFR) is introduced as a base learner to train the prediction model on the hybrid dataset. Finally, a case study of the aluminium alloy 7075 parts milling process shows that the proposed method is accurate in cutting energy consumption prediction. Compared with RFR and TR-RFR, the proposed method's coefficient of determination (R2) increases by 11.60% and 3.55%, indicating high goodness of fit under the same small samples of the experiment. Therefore, the proposed method could help determine the most efficient process plan without excessive time, materials and energy, significantly contributing to green manufacturing.