An adaptive modelling approach using a novel modified AOA/SVR for prediction of drilling-induced delamination in CFRP/Ti stacks

Delamination is the most critical damage in drilling the CFRP/Ti stacks under the impact of drilling parameters and tool structure, which makes the traditional theoretical or empirical models not have enough accuracy and be time-consuming due to the multi variables, while the machine learning model would suffer the unsuitable hyper-parameters and have a bad accuracy and generalization ability. This paper proposed an adaptive modelling approach to predict the delamination while drilling the CFRP/Ti stacks. This approach adapted the original arithmetic optimization algorithm (AOA) by adding a random disturbance phase to update the penalty coefficient C and the kernel coefficient γ of the support vector regression (SVR) automatically. In the meanwhile, the approach made use of the energy of the 5 stages in drilling the CFRP/Ti stacks and predicted the delamination damage both at the entrance and exit. The modified AOA optimized the training mean squared error(MSE) in predicting the entrance and exit delamination by 10.27 % and 33.63 %, while the accuracy of the proposed model can reach 96.7 % and 97.17 % respectively. The model got validated, and had a comprehensive ability containing the accuracy and generalization ability.