Cracks and physical damages are a threat to the strength of structures. Non-destructive test (NDT) measures are used to detect the damages at the earlier phase to avoid any major damage to the structures. Vibration signal processing is one of the NDT methods to determine the damage based on the experimental modal analysis. In this study, an experimental setup is devised to freely suspend a steel plate of size 30 cm by 60 cm. Based on the experimental modal analysis, the steel structure is struck using an impact hammer and the dispersed mechanical energy is bagged as vibration response using an accelerometer. The damages of size 512–1852 µm were manually simulated at arbitrary locations on the surface of the steel structure. The data acquisition procedure is repeated before and after the simulation of damage. The vibration signals are then processed, and the wavelet packet transform is applied to extract the dominant features from both the normal and fault data. The feature set is normalized between 0 and 1 are then mapped toward the condition of the plate to formulate the final dataset. Using a K-fold cross-validation technique, the dataset is trained and tested using support vector machine (SVM) and K-nearest neighbor (KNN) classifiers. The results are compared and discussed.
