Outlier Analysis and Artificial Neural Network for the Noncontact Nondestructive Evaluation of Immersed Plates

We present an experimental study where guided ultrasonic waves were used for the noncontact nondestructive evaluation of an aluminum plate immersed in water. Broadband leaky Lamb waves were generated using a pulsed laser and were detected with an array of immersion transducers arranged in a semicircle. The signals were processed to extract some features from the time, frequency, and joint time-frequency domains. These features were then fed to an unsupervised learning algorithm based on the outlier analysis to detect the presence of damage, and to a supervised learning algorithm based on artificial neural networks to classify the types of defect. We found that the hybrid laser-immersion transducers system and both learning algorithms enable the detection of the defects and their classification with good success rate.