A Study on Defect Identification of Ultrasonic Testing by Neural Network

Abstract

In recent year, the extremely thick plate becomes used in the deck plate of the large ships. To prevent the brittle fracture it is important to detect and measure the inner defects. Ultrasonic testing is used to find defects in structures. However, the precise estimation of defect shape and size is difficult through ultrasound testing of welds; currently, such high precision estimation depends on the experience and skill of the inspecting technician. This study aims to develop a method to quantitatively identify the shape of defects easily, quickly, and with high accuracy from the wave reflected from a defect during an ultrasonic test; in addition, such a method should not require the knowledge and experience of an expert. For this purpose, a neural network that uses numerical simulation results as training data was developed, and a defect identification system based on ultrasonic wave information was investigated. To confirm the usability and estimation accuracy of this system, measurement data from actual ultrasonic tests were used as the input in the defect identification system.