Utilization of artificial neural networks for detection and classification of damages in composite plate-like structures via ultrasonic guided waves

Using methods based on artificial intelligence to reduce the role of human interpretations in data analysis and obtaining the favorable results, in line with increasing the speed, reducing the errors and adjustment of the costs in the nondestructive evaluation and structural health monitoring is seriously concerned by researchers.
In this study, the design and implementation of a structural health monitoring system is performed by the intelligent signal processing of the ultrasonic waves in order to identify and classify the three common defects in the composite plate-like structures. By creating three types of damages including delamination, crack and hole in the multi-layer composite plate made of glass fiber reinforced polymer and dividing it into 4 different zones, 9 piezoelectric transducers with dual role of actuator and sensor are attached with their network arrangement and the propagated signals in the four mentioned zones on the 12 paths in three different directions including 240 signals were stored. In the next step, extraction of the features from the signals is conducted by the advanced signal processing techniques such as wavelet transform and the findings have been used to train a neural network of advanced multilayer perceptron by back-propagation error method.
The results show that the trained neural network algorithm is able to differentiate between the intact zone from the damaged ones. In addition, it has classified the types of current defects and damages in the structure with the acceptable efficiency (the average is about 80%), which can be generalized to the different conditions.