Damage detection by wavelet packet transform and multiclass SVM in structural health monitoring applications

In recent years, guided ultrasonic wave (GUW) technique has been widely used for detecting different damage types in aerospace, mechanical, and civil engineering structures. In this study, a feature extraction and pattern recognition algorithm based on wavelet packet transform (WPT) and multiclass support vector machines (SVM) is proposed to detect existence and severity of damage in a structural beam. A prismatic beam is simulated in healthy and damaged conditions and the corresponding signals are captured from finite element method (FEM) simulations. The damage is a slot having several depths, located in several positions in the beam. The computed signals from FEM simulations are decomposed by WPT. Then, statistical features of the decomposed signals are extracted. A multiclass SVM classifier is used to classify the conditions of simulation signals into four classes: healthy condition, low, medium and high severity damage conditions. By using 43 signals for training the classifier, it can classify 8 test signals into 4 condition classes perfectly. The performance of the SVM classifier is compared with two different artificial neural network (ANN) classifiers. The ANN classifier that has more hidden neurons can classify the damage conditions perfectly; however, another classifier can classify only 6 signals into correct conditions. Moreover, performance of the introduced algorithm is better than some other feature extraction algorithms including: WPT node energies and the statistical features algorithm

کلیدواژه ها:

Structural health monitoring (SHM) ، wavelet packet transform (WPT) ، multiclass support vector machines (SVM) ، damage detection