Prediction of surface roughness in turning using vibration features selected by largest Lyapunov exponent based ICEEMDAN decomposition

The prediction of surface roughness during turning using indirect methods helps improve the quality of production. This study proposes an intelligent surface roughness prediction scheme during turning of forged EN8 steel using ELM. The training and testing datasets are developed using Taguchi L25 and L8 orthogonal arrays respectively. The vibration signal acquired from machining is decomposed into modes using ICEEMDAN, followed by selection of the sensitive mode using LLE. Cutting parameters have been combined with features of the sensitive mode to form the feature matrix. Z-score normalization is used to normalize the matrix which is fed to ELM for surface roughness prediction. The SMA is used to optimize the kernel parameters of the ELM. The results obtained using arbitrary kernel parmeters have been compared with the optimized model. The mean absolute percentage error obtained using the proposed model is 4.957% representing better accuracy than the unoptimized results.