Damage monitoring of pultruded GFRP composites using wavelet transform of vibration signals

Abstract

A statistical procedure is presented for monitoring the damage of pultruded specimens using vibration data. Pultruded composite samples are processed by using continuous glass fiber rovings and epoxy matrix. Two different types of damages, a notch in the top layer and a hole at the midsection has been introduced individually in these samples. Vibration signals from healthy and damaged samples are processed using wavelet transform. various time–frequency domain statistical features have been extracted. Also, the time-domain features have been computed from the raw amplitude data. Further, both these features have been exploited to construct the feature space and significant features have been identified by evaluating the contribution rates. Finally, significant feature set is channeled as input to various machine-learning classifiers (deep neural network and support vector machine) and different classification accuracies have been estimated. The capability of these methods to early detect the damage in pultruded composites has been discussed.