Disbond detection through ultrasonic signal classification using an artificial neural network

Summary form only given. An ultrasound technique for detecting disbonds in aircraft lap joints and in the adhesive joints between aircraft skin and reinforcing doublers was discussed. A high-frequency ultrasonic pulse is transmitted into the aircraft skin by a contacting ultrasonic transducer. This pulse is reflected at the bond interface, and is picked up by the transducer. The output is a time-varying ultrasonic signal that characterizes the bond interface. The use of an artificial neural network for classifying the signals as corresponding to bonded and disbonded regions is discussed. Training and classification performances were obtained for several values of network parameters. A peak classification accuracy of 98.7% was obtained on the test signal set. The advantages of using a neural network are its low noise sensitivity, low classification time, high classification accuracy, and convenient threshold capability of output for disbond detection