Estimation of the delamination size of carbon fiber panels from the impact sound
Conference
Uriarte, P, Shah, KY, Tansel, I et al. (2019). Estimation of the delamination size of carbon fiber panels from the impact sound
. 1 89-96. 10.12783/shm2019/32102
Uriarte, P, Shah, KY, Tansel, I et al. (2019). Estimation of the delamination size of carbon fiber panels from the impact sound
. 1 89-96. 10.12783/shm2019/32102
Large composite panels of airplanes are regularly inspected, and delaminated sections are repaired. Experienced technicians can detect the delaminated sections of composite panels from the sound created from a small impact. In this study, performance of computerized analysis for impact sound was evaluated for composite panels at different delamination levels. The impact sound of carbon fiber panels without any defect and panels with simulated delaminated sections were recorded. The recorded audio data were processed and interpreted. Fast Fourier Transformation was used for converting the audio data into the frequency (spectrum) or time-frequency domain (spectrogram). Neural networks were used for the identification of carbon fiber panels. Three different approaches were used for encoding the data before they were used for the training and testing of the neural networks. The simulated delamination sizes of the panels were 5mm X 5mm, 10 mm X 10 mm, 20 mm X 20 mm and 30 mm X 30 mm. The Levenberg-Marquardt neural network algorithm provided the most consistent estimation compared to the Gradient Descent with Momentum, Fletcher Reeves and Resilient Backpropagation algorithms. During test cases, the Levenberg-Marquardt neural network algorithm identified a carbon fiber panel's delamination level with lower than 10% error. The study indicated that the audio data from the sound of impact could be used for the identification of delamination severity in composite structures.
