ABSTRACT
ABSTRACT This chapter presents a novel approach to detecting and localizing structural defects based on a novelty detection method called outlier analysis and a multilayer perceptron (MLP) neural network. The main specimen used here was a rectangular carbon fiber-reinforced plastic composite plate. The effectiveness of the selected approach was assessed by analyzing the experimental data acquired from the PZT patches. The scope of this present work also comprises an investigation of the scattering effect of an ultrasonic-guided wave on the tested plate under both damaged and undamaged conditions. The wave propagation is sequentially transmitted and captured by eight PZT patches bonded on the plate, forming a sensor network on the tested composite’s rectangular structure. An in-house eight-channel multiplexer is incorporated into this small-scale and low-cost structural health monitoring system to effectively swap the PZTs role from sensor to actuator and vice versa. The Real Time Damage Demonstrator software is primarily developed to acquire and store the waveform responses. These sets of scattering waveform responses representing normal and damaged conditions are transformed into a set of novelty indices that ultimately determine the true conditions of the tested structure. The acquired novelty indices representing the available sensor paths are used as the inputs for the neural network incorporating the multilayer perceptron architecture to compute and predict the damage location as the x and y location on the tested composite plate.
