Operational Damage Identification Scheme Utilizing De-Noised Frequency Response Functions and Artificial Neural Network

A damage identification scheme combining impact-synchronous modal analysis (ISMA) and artificial neural network is developed in this study. The ISMA de-noising method makes it feasible to detect and classify the damage states with high accuracy when the machine is under operation. The feed-forward backprop network was utilized in this study. The input feature vector of the network consisted of the FRF changes in a selected vibrational mode frequency interval at several measurement points. The scheme was tested on a rectangular Perspex plate. It is proved that the trained network can successfully identify damage locations with the testing data collected by ISMA, which allows the damage detection to be carried out without shutting down the tested machine. For the plate structure in this study, an overall accuracy reached 100% when all five measurement points were used. With the input features optimized by mode shape assessment, 100% accuracy was also achieved with only two measurement points.