A complete online-SVM pipeline for case-based reasoning system: a study on pipe defect detection system
- PDF / 2,063,100 Bytes
- 17 Pages / 595.276 x 790.866 pts Page_size
- 94 Downloads / 196 Views
METHODOLOGIES AND APPLICATION
A complete online-SVM pipeline for case-based reasoning system: a study on pipe defect detection system D. Van-Khoa Le1
· Zhiyuan Chen1 · Yee Wan Wong2 · Dino Isa3
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Recent developments in case-based reasoning system (CBR) have led to an interest in favoring machine learning (ML) approaches as a replacement for traditional weighted distance methods. However, valuable information obtained through a training process was relinquished as transferring to other phases. This paper proposed a complete pipeline integration of CBR using kernel method designated with support vector machine (SVM) as the main engine. Since the system requires learning SVM model to be invoked in every phase, the online learning mechanism is nominated to effectively update the model when a new case adjoins. The proposed full SVM-CBR integration has been successfully built into a pipe defect detection. The achieved result indicates a substantial improvement by transferring learning information accurately. Keywords Defect detection · Case-based reasoning · Online learning · SVM · Expert system
1 Introduction CBR has long built a solid cornerstone in the expert system domain. The application is favorable and not limited to faults and troubleshooting problems. As approaching problems in a human perspective (i.e., judgment is concluded according to the outcome learnt from experience) is seen essential, Communicated by V. Loia. Dino Isa: The author was deceased.
B
D. Van-Khoa Le [email protected] Zhiyuan Chen [email protected] Yee Wan Wong [email protected] Dino Isa [email protected]
1
School of Computer Science, Faculty of Science and Engineering, The University of Nottingham, Malaysia Campus, Nottingham, UK
2
Department of Electrical and Electronic Engineering, Faculty of Science and Engineering, The University of Nottingham, Malaysia Campus, Nottingham, UK
3
CFFRC Research Centre, Crop For The Future Research Centre, Jalan Broga, 43500 Semenyih, Selangor, Malaysia
industrial organizations are keen on consolidating system following CBR concept. Consequently, the process of CBR is designed to indicate the phases of analyzing past experience, abbreviated into 4-R cycles: retrieve, reuse, revise, and retain. However, designing an explicit CBR endures inevitable restraints. Many existing CBR systems require indicators using expert domains (Hashemi et al. 2014; Gu et al. 2017; Khosravani et al. 2019; Marling et al. 2014) (Begum et al. 2014). Generally, most of CBR systems are designed to retrieve similar cases automatically. The common methods including clustering techniques such as k-nearest neighbors (kNN) (Li and Sun 2009), kernel methods (Fyfe and Corchado 2002), and distance measurement (Gu et al. 2017) rule-based approaches are favorable (Sharaf-El-Deen et al. 2014) for some particular systems (Marling et al. 2014; Herrero et al. 2015; Mohammed et al. 2018). Unlike clustering methods, some rule-based
Data Loading...