Canny Edge검출 기법과 합성곱 신경망을 이용한 심층학습 기반 터보차저 진동 분석

초록

This study proposes a novel method for analyzing turbocharger vibration utilizing a combination of the convolutional neural networks (CNN) technique and the Canny edge detection technique. The conventional 1D machine-learning techniques exhibit poor accuracies due to the irregularly changing rotating speeds of the turbine and compressor, which are inherent to turbochargers. To address the issue, the adoption of a 2D CNN accompanied by the Canny edge detection technique has been proposed here. First, the experimental vibration data of the turbocharger have been created and labeled. To augment the limited number of experimental data, the data augmentation technique is applied, resulting in a sufficient amount of data. Subsequently, various preprocessing and machine- learning techniques have been applied for the analysis. By comparing the accuracy and loss data of these analyses, the 2D CNN accompanied by the Canny edge detection technique has been confirmed as the most effective approach for analyzing turbocharger vibration.

키워드