Machine Learning for Nondestructive Evaluation

This paper reports on the current status of a collaborative project exploring applications of machine learning methods to Nondestructive Evaluation (NDE). It presents initial results of applying AI methods for inductive learning, feature extraction, and function-finding to support the ultrasonic diagnosis of defective metal parts. Experience with a simple classification approach using ID3 has led us to develop an adaptation of a machine vision technique to obtain a more abstract feature representation. Learning from examples expressed in the new representation is expected to be less sensitive to noise. In addition, the new representation better supports function-finding and more knowledge-intensive classification processes.