Structural Integrity Diagnosis for Valve Based on Deep Convolution Neural Network

Abstract

Valve is a critical device to control fluid flow as an in-plant installation. However, much accidents due to an out-of-control valve have been reported in the past. One reason that the valve will become out-of-control is a Fixation-damage due to corrosion. It is important to evaluate the structural integrity of valve installed in plants because a fixation-damage causes serious accidents. For the purpose of developing a structural integrity monitoring system of valve, this study attempts to classify structural integrity of valve from vibration image data by using deep convolution neural network. First attempt is made to investigate the relationship between vibration characteristics of the valve with fixed-damage due to corrosion and torque to open/close the valve. Second attempt is made to make training data sets from the torque and the image data of vibration frequency of valve and train deep convolution neural network. Finally, it is demonstrated whether the structural integrity of the valve can be classified by using the trained deep convolution neural network.