Publication: Delamination under insulation assessment using CNN base
Loading...
Date
2023-07
Authors
Lee, Ren Kai
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Non-destructive testing (NDT) is an essential and increasingly significant technique in various industries as it enables the detection of defects and flaws without
compromising the integrity of tested materials or components. The goal of this research is to create a technology for measuring sample defect size utilising a
microwave-NDT method and convolutional neural networks (CNN) by using various signal processing technique and modify the pooling layer of CNN using outlier removal techniques. The input data is received from a vector network analyzer (VNA) the sample utilised for defect testing which call Macor sample a ceramic coating sample. The representative ceramic coating sample is scanned using a Q-band open ended rectangular waveguide, acquiring data at 101 frequency points between 33 and 50 GHz. Subsequent to data processing, a CNN model is trained using the processed data for defect identification and classification. By comparing the projected outcomes with the actual results, the defect patterns are estimated, and accuracy is assessed. To obtain more precise data, this process is repeated. The proposed method aims to provide a valuable tool for industrial applications by addressing the issue of delamination under insulation assessment in a non-destructive and effective manner. The project's results show that the accuracy rate of 97.84% using Tukey Rules a signal
processing method to change the data from frequency domain to frequency time domain by implemented the Z-score to the pooling layer of the CNN model. This is slightly better compared with other signal processing methods and other outlier removal techniques used in this project.