Damage Detection With an Ultrasound Array and Deep Convolutional Neural Network Fusion

Abstract

Diagnostic methods for power transmission facilities are important for energy security because the growth of defects in power facilities increases the risk of blackouts in an entire power grid. However, damage in power transmission facilities is difficult to detect because cracks or defects are minuscule and are challenging to determine. One interesting phenomenon caused by damage in power transmission facilities is ultrasound emissions on a damaged surface. However, measuring ultrasound emissions to detect defects is limited by the severity of the surrounding noise. To overcome this limitation, this study proposes a new method for damage detection by fusing ultrasound measurements with recorded optical images. The proposed method consists of two phases. The first phase preprocesses ultrasound measurements for ultrasound feature extraction. This phase aims to detect the location of ultrasound emissions by analyzing ultrasound characteristics including the intensity and density. The second phase detects and classifies a damaged object with optical images recorded using a deep convolutional neural network. This phase not only discards the noise from the ultrasound measurements but also classifies a damaged system among many components in power transmission facilities. The experiments validate the effectiveness of the proposed method using ultrasound measurements and recorded images and finally suggest scenarios for potential applications.