영상 기반 딥러닝을 통한 배관 진동 주파수 가시화 기술

Abstract

Pipe systems in industries function similar to blood vessels in the human body. Pipe vibration is a natural phenomenon caused by external motors and fluid flow in the pipe. However, any unfavorable factors, such as in-wall collision by loose parts or unusual fluid flow, can significantly affect the vibration, which results in abnormal vibration patterns when compared to those during regular operation. For this reason, pipe vibration frequency is one of the important parameters to monitor in structural health monitoring. Therefore, a monitoring system that measures the vibration frequency of each pipe area helps to detect these anomalies early. In this study, a multi-kernel neural network was applied to visualize the vibration frequency of pipe areas using a multi-kernel neural network, by analyzing the characteristics of pixel-wise color variations in video data. The results showed that the vibration areas can be visualized using the color that corresponds to the frequency. The proposed model can be utilized for anomaly detection based on pipe vibration monitoring.