Recognition Types of Cracked Material under Uniaxial Tension Based on Improved Mel Frequency Cepstral Coefficients (MFCC)

Acoustic emission (AE) waveforms contain rich feature information on material attributes, which is extremely important for identifying material types. In this work, a novel method for identifying material types is proposed in crack process based on improved Mel Frequency Cepstral Coefficients (MFCC) and K-Nearest Neighbor (KNN) algorithm. The dynamic characteristics of materials fracture were obtained based on the improved algorithm of MFCC. The parameters and their first derivative are calculated from AE signals by using MFCC algorithm. Features are divided into training and testing dataset. The recognition accuracy of material types exceeds 90% by using KNN classifier in crack process, which highlights that the proposed method has better practicability for the problems of material recognition including copper, epoxy resin and organic glass. The research results of this paper can locate fault sources quickly for multi-component and composite systems, which can enrich the connotation of structural health monitoring.