Please use this identifier to cite or link to this item:
|Title:||Advanced feature extraction and selection for fault diagnosis and prognosis||Authors:||Zhou, Junhong.||Keywords:||DRNTU::Engineering::Mechanical engineering::Control engineering
|Issue Date:||2012||Source:||Zhou, J. (2012). Advanced feature extraction and selection for fault diagnosis and prognosis. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||Condition based maintenance (CBM) has become increasingly important over the past decade as a means to maximize the asset usage and plant operating efficiency. CBM reduces maintenance cost by reducing the unnecessary scheduled preventive maintenance where equipment outages are predicted and maintenance is carried out only when necessary. The CBM methodology and techniques have been extensively researched in a variety of application areas such as rotating machinery, aerospace systems, chemical manufacturing, electronic and electrical components by universities, laboratories and the industry. Among these, advanced feature extraction and selection are essential research issues for realizing CBM. In this research, a novel Dominant Feature Identification (DFI) methodology is first proposed. Singular Value Decomposition (SVD) is used to decompose the inner product matrix of collected data from the monitoring sensors. The principal components are optimized in a least squares sense in a certain reduced space, and the dominant features are extracted using the K-means clustering algorithm. DFI provides an autonomous rule-base for data and sensor reduction. To efficiently use DFI in the CBM framework, a methodology is developed to integrate the proposed DFI with traditional feature extraction methods of time domain and frequency domain analysis, working together with Recursive Least Squares (RLS) and Extended Least Squares (ELS) modelling methods. The developed methodology and framework is evaluated based on the accuracy of prediction of the actual milling tool wear. Further improvement has been developed to integrate DFI with advanced feature extraction techniques of wavelet based correlation modelling. The efficiency of the advanced feature extraction and selection methodology is evaluated based on the accuracy of health condition assessment of the brushless DC motor. The research scope was extended from single machine CBM to complex manufacturing environment which may consist of many different types of machines. A system framework of iDiagnosis & Prognosis was developed to provide distributed intelligence of diagnosis and prognosis, to provide system condition and job execution capability assessment, and provide real time machine condition feedback to the supervisory control. The system framework, architecture, functions, and information flow are developed for implementing the diagnosis and prognosis in a complex manufacturing environment.||URI:||http://hdl.handle.net/10356/50600||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||MAE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.