Pulsed eddy current NDT for quantitative evaluation of inclined cracks /
To date, Pulsed eddy current testing (PEC) has been one of the most attractive non-destructive testing (NDT) due to the richness of information provided by the broadband of frequencies of magnetic induction. It has been proposed to be used in various NDT applications. However, none of the research s...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur :
Kulliyyah of Engineering, International Islamic University Malaysia,
2018
|
| Subjects: | |
| Online Access: | http://studentrepo.iium.edu.my/handle/123456789/5154 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | To date, Pulsed eddy current testing (PEC) has been one of the most attractive non-destructive testing (NDT) due to the richness of information provided by the broadband of frequencies of magnetic induction. It has been proposed to be used in various NDT applications. However, none of the research sought to exploit the use of PEC in quantifying inclined cracks. Cracks with inclination angles tend to harm a bigger region in the tested structures due to the elongation of the cracks. This study aims at exploring the reliability of utilising PEC in characterising surface crack inclination angles and depths, through extraction of features from C-scan images. In the pursuit of doing so, finite element models (FEM) were first built to study the interaction of magnetic field with cracks of different inclination angles. Using the FEM, optimal excitation coil parameters were obtained, which then be used to develop a PEC system. The PEC system was developed incorporating a scanner system, a data acquisition system and an excitation circuit, capable of scanning in 2D while picking up normalised response signals at each spatial resolution. By examining 20 surface cracks and five subsurface cracks, C-scan images were obtained from each simulated crack. An image-based feature extraction technique was proposed, adopting multi-step algorithm involving local maxima localisation and Hough transform. Four novel features were extracted from each C-scan image. Two surface-subsurface crack classification features, termed as LLS and Gmax, were proven to be capable of distinguishing crack classes. Two other features, skewness and LSmax, alongside with the LLS, have been presented for estimating the crack inclination angle and depth. With the analysis of the linear correlation of the features with inclination angle and depth, a multiple linear regression (MLR) and a hierarchical linear model (HLM) were built, utilising 6020 observations from all the cracks, where each C-scan image contributed to 301 features. An artificial neural network (ANN) was also built to simultaneously predict the two crack parameters, as well as to consider the interdependency of the three features. Comparisons between the prediction performance of the MLR and the ANN did not differ much. On the other hand, the ANN was proven to handle interdependency of features to the responses better, thanks to the backpropagation algorithm within the ANN. The ANN outperformed the HLM by recording an RMSE of 4.7634º, while the HLM was 8.980º. Nevertheless, this research work was still limited by the hardware capabilities that it can only effectively quantify inclined surface cracks. Continual efforts on designing optimal probes capable of providing more information on the subsurface cracks, as well as proposing better image processing techniques, are expected to contribute to the limitations. |
|---|---|
| Physical Description: | xvii, 143 leaves : colour illustrations ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 126-132). |