Guided Waves Damage Detection in Beams utilising Bayesian Artificial Neural Network Model Class selection

This paper presents a damage detection methodology utilising an Artificial Neural Network (ANN) pattern recognition approach to identify the location and severity of damages in beams. The Bayesian model class selection method is employed to determine an optimum ANN model, hence avoiding subjective judgement during the ANN design process. Simulated first order longitudinal guided waves signals for beams with different damage locations and severities are pre-processed using discrete wavelet transformation to reduce ANN processing time and are then used as input patterns to train the ANN. The trained network is employed to identify unknown damages in beams with simulated damage cases.