Artificial Intelligence (AI)–Based Surface Quality Prediction Model for Carbon Fiber Reinforced Plastics (CFRP) Milling Process
Jin Woo Kim, Jungsoo Nam, Geun Young Kim, Sang Won Lee
Abstract
The objective of this study is to develop a surface quality prediction model for CFRP milling based on artificial intelligence and enhance the model performance by proposing a novel feature through signal processing and applying it to machine learning algorithms. In this paper, the data-driven approach is applied based on data that is obtained by various sensors such as a two-axis load cell, an acoustic emission (AE) sensor, and an accelerometer, and surface quality is predicted as one of three grades based on surface roughness value. To establish the models, a novel feature, referred to as the tool rotation frequency (TRF) feature, was introduced as well as the conventional wavelet transform (WT) feature. Then, three machine learning algorithms – artificial neural network (ANN), K-nearest neighbor (KNN), and support vector machine (SVM) – are used to construct the surface quality prediction models. Each model is trained with 450 data sets obtained from 45 experimental cases, and their predictive accuracy is evaluated by the k-fold cross validation method. Next, to confirm each model’s robustness, 200 data sets obtained from additional 20 experimental cases designed with untrained conditions are tested. Finally, the monitoring software is developed for visualizing prediction results of the surface grades.