전이학습 기반 가공동력 예측 모델링 방법

Abstract

Machining power is a critical indicator for energy-efficient machining because it influences energy consumed during machine tool’s operations. Previous studies have derived predictive models that figured out the relationship between process parameters and machining power and help decide process parameters contributory to energy reduction. These studies mainly use machine learning approaches, which rely on learning datasets. However, real fields cannot always provide learning datasets due to the difficulty of data collection and thus such traditional approaches cannot work in a data scarce environment. The present work proposes a transfer learning-driven approach of predictive modeling for machining power. The proposed approach can create machining power prediction models in the data scarce environment through knowledge transfer of prior machining contexts to the target machining context. The present work includes a case study to demonstrate the validity of the proposed approach. The case study shows that machining power prediction models for titanium material of which machining power has been unlabeled can be created from those models for steel and aluminum materials of which machining power was labeled.