A Hybrid Approach for Surface Roughness Prediction Based on Multi-domain Feature Fusion and Deep Learning

Abstract

The prediction of surface roughness in machining is of great influence on the assembly and reliability of precision equipment. Although the existing data-driven models consider both static and dynamic factors, the multi-domain features of dynamic factors are not effectively integrated, which results in unable to effectively capture the deterioration trend of surface roughness. This paper proposed a hybrid approach composed of a theoretical model and a data-driven model. Specifically, a novel deep network framework is designed to achieve the fusion of time-domain and time-frequency domain features. After that, the end-to-end prediction model of signal-to-surface roughness is established by the knowledge self-mining capability of deep learning. In addition, the transfer learning (TL) technique is also introduced to accelerate the training process of the deep learning network. The proposed approach is applied to surface quality inspection of the milling process and promising experimental results demonstrate the effectiveness of the proposed framework in practical engineering applications.