Neural network applications to ultrasonic nondestructive evaluation

Ultrasonics, between 500 kHz and 5 MHz, is used to inspect welded piping in electric power generating plants. Detection, characterization, and sizing of defects, possibly detrimental to the pipe’s structural integrity (e.g., cracks, lacks of weld fusion) are the tasks of ultrasonic inspection. These tasks are compromised by the similarities and subtle differences between ultrasonic responses from the nonconsequential geometric reflectors of the weld and pipe structure and other critical reflectors. This paper shows how neural networks can be used to discriminate integrity-threatening flaws from geometric artifacts. A sample application, using ultrasonic waveforms collected from welded pipes, is presented as part of the paper.