Crack Depth Specification of a Circular-Section Bar Using Peak Frequencies from Impact Test and Support Vector Machine Classification

Non-destructive data-driven approaches were noticed by researchers to crack identification of beam like structures. In this research, peak frequencies of the Fourier spectra of acceleration signals were used as crack depth classification features. Experimental data of impact tests are collected and first frequency peaks were extracted from circular bars with different crack depths. Extracted feature matrix was used to train an SVM model. The obtained performance of the classifier model shows that the frequency peaks can be used in the depth estimation of cracks, when the input force and consequently the FRFs are not available. While the sensor masses cause noticeable effects on the natural frequencies of the structure, peak frequencies of impact response of the sensor mounted systems still can indicate the depth of crack with acceptable accuracy. Also, the research showed that use of more number of peak frequencies can enhance the performance of classification. Acceptable performance of classification and cross-validation results were obtained using first 10 peak frequencies