드릴 가공 시스템의 무인화를 위한 이상감지 기법 연구
Research on Anomaly Detection Techniques for Unmanned Drilling Systems
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Although air vent hole machining is a core process in the automotive mold & die industry requiring high precision and repeatability, the detection of tool breakage and wear relies heavily on the qualitative judgment of operators, presenting a significant barrier to complete automation. To address this, this study proposes a vibration and acoustic-based add-on module and a machine learning-based closed-loop control system that can be deployed on legacy drilling equipment without structural modification. Using heterogeneous sensor data sampled at 6400 Hz, 84 statistical features reflecting physical characteristics were extracted. An anomaly detection model was constructed by combining RobustScaler and the Random Forest algorithm, achieving 93.54% accuracy and demonstrating robustness against industrial noise. Furthermore, 'Pierce' (penetration detection) and 'Wear' (wear/breakage response) algorithms were implemented based on real-time state detection, verifying the effectiveness of an intelligent unmanned process capable of Automatic Tool Change (ATC) and process resumption upon tool failure.
Abstract
Although air vent hole machining is a core process in the automotive mold & die industry requiring high precision and repeatability, the detection of tool breakage and wear relies heavily on the qualitative judgment of operators, presenting a significant barrier to complete automation. To address this, this study proposes a vibration and acoustic-based add-on module and a machine learning-based closed-loop control system that can be deployed on legacy drilling equipment without structural modification. Using heterogeneous sensor data sampled at 6400 Hz, 84 statistical features reflecting physical characteristics were extracted. An anomaly detection model was constructed by combining RobustScaler and the Random Forest algorithm, achieving 93.54% accuracy and demonstrating robustness against industrial noise. Furthermore, 'Pierce' (penetration detection) and 'Wear' (wear/breakage response) algorithms were implemented based on real-time state detection, verifying the effectiveness of an intelligent unmanned process capable of Automatic Tool Change (ATC) and process resumption upon tool failure.