Intelligent signal processing system to investigate damage severity in kenaf fibre composite
The emergence of natural fibre as potential alternative for glass fibre replacement has seen various development and investigation for various applications. However, the main issue with the natural fibre reinforced with composites is related to its susceptibility to impact damage. The capability to...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2016
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/9050/ http://eprints.uthm.edu.my/9050/1/Zaleha_Mohamad.compressed.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The emergence of natural fibre as potential alternative for glass fibre replacement has seen various development and investigation for various applications. However, the main issue with the natural fibre reinforced with composites is related to its susceptibility to impact damage. The capability to detect damage at early stages reduces any risk of catastrophic failure. This research investigated impact damage severity in Kenaf fibre composites (KFC) structures using signal processing procedures. The mechanical properties for Kenaf fibre composite were investigated based on 5%, 10%, 15% and 20% fibre volume fraction. Then the KFC was instrumented with piezoceramic transducers and excited with impact hammer. The response signals captured from each sensor were recorded by a data acquisition system. The impacted specimens were examined with visual inspection (dye penetrant) to examine the extent of damaged areas. Then, an effective damage severity regression and classification procedure is established using a multilayer perceptron neural network approach. The system was trained to predict the damage area based on the actual experimental data. The comparison between experimental and simulation for wave velocity propagation in KFC demonstrated a good agreement in their pattern. For the network, the result demonstrated that the trained networks were capable to predict the damage size accurately. The best performance was achieved for an MLP network trained with minimum signal features, which recorded the error less than 0.40%. While that, for the classification, the data features were mapped into five output class labels, presented as a target confusion matrix. The results revealed that the damage sizes were successfully mapped according to its respective classes, with the minimum peak feature gives the highest classification rate at 97.9%. While that, the principal component analysis (PCA) is used to support the ANN classification analysis. The result revealed that PCA can cluster and separate the group within their classes. |
|---|