Improving the Prediction Accuracy of Surface Roughness in the Boring Process using Integrated Machine Learning Methods

Precision boring is essential for achieving high-quality holes, yet attaining the desired surface roughness remains a challenge due to the confined machining environment and the instability of long, slender boring bars. Although simple analytical models exist, they often fall short in accurately predicting surface roughness in boring operations. In machining applications, conducting extensive experiments is limited by high costs and time constraints; thus, small sampling problems occur in the prediction models. Therefore, it is necessary to establish high-accuracy surface roughness prediction models with optimized input features. In this study, an integrated machine learning-based methodology is proposed to overcome these challenges. The approach includes ranking input features by importance, incorporating physics-based features, expanding the feature space, and generating virtual samples to enrich the training data. The results indicate that the proposed model improves prediction accuracy by 27%. The most significant enhancement was observed when virtual samples were added after expanding the feature set with physics-based parameters. These findings contribute to more precise and cost-effective planning in boring operations.