Time delay neural network for ultrasonic flaw detection

Ultrasonic flaw detection in large-grained materials is difficult because of multiple scattering from material microstructure (grain noise). The Time Delay Neural Network (TDNN) has been used quite successfUlly to process time series data. Therefore, it is possible to achieve SNR enhancement in ultrasonic testing by using this technique. In this paper, a new TDNN architecture with two input variable,viz. wave form and its phase difference, is developed to reduce the grain noise. Experimental results demonstrate the ability of the TDNN architecture to reveal flaw echoes from high level of background noise.