Use of reliability measures to improve the performance of fuzzy ARTMAP networks

Neural network based signal classification systems are being applied increasingly in nondestructive evaluation to solve the inverse problem. In general, two issues not usually addressed are (i) estimation of reliability measures of the network decision and (ii) ability of the network to learn and improve its performance with time. This paper presents a signal classification system using the fuzzy ARTMAP network. Fuzzy logic based reliability measures are developed for the fuzzy ARTMAP network and used as a feedback for retraining the network to improve its performance. The performance of the algorithm is demonstrated using ultrasonic data obtained from the inspection of welds in nuclear flow plant piping.