개선된 Deep Residual Learning 및 물리 모델 데이터를 이용한 잡음 조건에서의 자동차 샤시 결함 진단

초록

For autonomous vehicles, technology that monitors the state of the vehicle and detects a failure using sensor data is receiving increasing attention. The purpose of this study is to determine the type and location of faults of a vehicle chassis system under noisy conditions using acceleration data and deep learning. Because there is a limit in the acquisition of specific defect data from a real vehicle, normal and defect data were obtained using a vehicle physics model that considers various vehicle speeds, vehicle-to-vehicle variations and road changes. We proposed DNI-ResNet (DenseNet Inspired ResNet), which applied the advantages of DenseNet to ResNet, and used it to determine the type and location of defects occurring in the rubber of the vehicle chassis system. Additionally, the domain adaptation ability of the proposed method was verified with various vehicle speed and new types of defects.

키워드