Nine damage cases, at different locations, orientations and length, were introduced into the pipe structure. The pipe was impacted 300 times using an impact hammer, after each damage case, the vibration data were collected using 3 PZT wafers which were installed on the outer surface of the pipe. At first, damage sensitive features were extracted using the frequency response function approach followed by recursive feature elimination for dimensionality reduction. Then, a multi-class support vector machine learning algorithm was employed to train the data and generate a statistical model. Once the model is established, decision values and distances from the hyper-plane were generated for the new collected data using the trained model. This process was repeated on the data collected from each sensor. Overall, using a single sensor for training and testing led to a very high accuracy reaching 98% in the assessment of the 9 damage cases used in this study.
ACCESS THE FULL ARTICLE
S. Mustapha, A. Braytee, L. Ye, "Detection of surface cracking in steel pipes based on vibration data using a multi-class support vector machine classifier," Proc. SPIE 10168, Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2017, 101682K (12 April 2017);