Novel Hybrid Method Based on Advanced Signal Processing and Soft Computing Techniques for Condition Assessment of Timber Utility Poles
- Publication Type:
- Journal Article
- Journal of Aerospace Engineering, 2019, 32 (4)
- Issue Date:
|Novel Hybrid Method Based on Advanced Signal Processing and Soft Computing Techniques for Condition Assessment of Timber Utility Poles.pdf||Published Version||1.81 MB|
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
© 2019 American Society of Civil Engineers. Recently, a variety of nondestructive evaluation (NDE) approaches have been developed for health assessment and residual capacity estimation of timber structures. Among these methods, guided wave (GW)-based techniques are highly regarded as effective tools for potential use in real situations. Nevertheless, because it is hard to comprehensively grasp the behavior of wave propagation in a wood structure, existing NDE-based techniques mainly depend on an oversimplified hypothesis, which can result in inaccurate or even misleading results in practice. Understanding the complex behavior of GW propagation in wood structures and extracting appropriate information from captured GW signals is a key for successful assessments of in situ conditions of timber structures. This paper analyzes the existing feature extraction and damage detection algorithms, and proposes a novel approach based on an integration of wavelet packet transform (WPT) and ensemble empirical mode decomposition (EEMD) for extracting damage-sensitive patterns, and then a soft computing method like support vector machine (SVM) for pole condition identification. In the proposed method, GW signals measured from a multisensing system with pole health condition as the baseline are divided into a series of subfrequency bands based on WPT. Then EEMD is adopted to extract the intrinsic mode functions (IMFs) that possess the features extracted at corresponding subfrequency bands. Hence, the IMF component was segregated from the original signals of tested poles, and the IMF Shannon entropy was employed to build up the feature vector to effectively demonstrate the health condition. To decrease the size of the feature vector and avoid multiple collinearity among obtained patterns, principal component analysis was employed and entropy information in the feature vector was replaced with main principal components, which will be employed as input variables of the developed SVM model for identifying pole health condition. In order to reduce the assessment error of the SVM model, genetic algorithm was introduced to select optimal parameters in SVM. Finally, the performance of the proposed method was assessed using laboratory timber specimens on which the experimental tests were conducted.
Please use this identifier to cite or link to this item: