Implementation of vibration-based structural health monitoring technique for identification of simulated corrosion damage in steel pipeline using neural network
Corrosion defect has inevitably causes serious incidents in pipeline structures. Reduction in corrosion related incidents are highly desirable due to safety and cost efficiency. Current approaches have implemented destructive testing which highly cost and time consumptions. Moreover, the techniques...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/234/1/24p%20MUHEIEDDIN%20MEFTAH%20FARAG%20ELGHANUDI.pdf http://eprints.uthm.edu.my/234/2/MUHEIEDDIN%20MEFTAH%20FARAG%20ELGHANUDI%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/234/3/MUHEIEDDIN%20MEFTAH%20FARAG%20ELGHANUDI%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Corrosion defect has inevitably causes serious incidents in pipeline structures. Reduction in corrosion related incidents are highly desirable due to safety and cost efficiency. Current approaches have implemented destructive testing which highly cost and time consumptions. Moreover, the techniques were lacking in correlating corrosion behaviour and its damage severity. This research proposed several signal corrosion features extracted from time domain analysis which provide substantial information related to corrosion behaviour for damage classification analysis. Several corrosion damage scenarios were simulated with different depths indicating its severity conditions. Seven corrosion features in time domain were introduced and extracted from the strain signal obtained from multiple sensors attached to the pipeline structure. The aim was to obtain the monotonically linear behaviour in features which could provide good correlation between corrosion features and corrosion damage. The experimental features were validated with the computational simulation works done for undamaged case only representing the baseline conditions. These features were subsequently used as input parameters for artificial neural network to classify corrosion damage into six type of damage depth representing different damage severity. The results demonstrated only four corrosion features were found to have linear monotonically behaviour with impact damage which were maximum, minimum, peak to peak and standard deviation features. The simulation works obtained an average of 2 - 8% in relative error with the experimental results. The classification analysis also has demonstrated a feasible method for classifying damage into classes with the accuracy ranged from 84 – 98%. These findings were substantial in providing information for pipeline corrosion monitoring activities. |
|---|