An improved PSO-SVM model for online recognition defects in eddy current testing
| Autor: | Pingjie Huang, Peihua Chen, Dibo Hou, Baoling Liu, Guangxin Zhang, Huayi Tang, Wubo Zhang, Banteng Liu |
|---|---|
| Rok vydání: | 2013 |
| Předmět: |
Engineering
business.industry Mechanical Engineering General Physics and Astronomy Differential structure Particle swarm optimization Control engineering law.invention Support vector machine Nonlinear system Mechanics of Materials law Eddy-current testing Simulated annealing Eddy current General Materials Science business Algorithm Classifier (UML) |
| Zdroj: | Nondestructive Testing and Evaluation. 28:367-385 |
| ISSN: | 1477-2671 1058-9759 |
| DOI: | 10.1080/10589759.2013.823608 |
| Popis: | Accurate and rapid recognition of defects is essential for structural integrity and health monitoring of in-service device using eddy current (EC) non-destructive testing. This paper introduces a novel model-free method that includes three main modules: a signal pre-processing module, a classifier module and an optimisation module. In the signal pre-processing module, a kind of two-stage differential structure is proposed to suppress the lift-off fluctuation that could contaminate the EC signal. In the classifier module, multi-class support vector machine (SVM) based on one-against-one strategy is utilised for its good accuracy. In the optimisation module, the optimal parameters of classifier are obtained by an improved particle swarm optimisation (IPSO) algorithm. The proposed IPSO technique can improve convergence performance of the primary PSO through the following strategies: nonlinear processing of inertia weight, introductions of the black hole and simulated annealing model with extremum disturbance... |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
| Nepřihlášeným uživatelům se plný text nezobrazuje | K zobrazení výsledku je třeba se přihlásit. |