Mathematical Frameworks for Diagnostics, Prognostics and Condition Based Maintenance Problems
Abstract
This report documents the theoretical and computational investigation of statistical pattern recognition techniques and Bayesian Networks (BN). The application of statistical pattern recognition methodology and Bayesian Networks to automatic fault diagnostics, fault prognostics, and condition based maintenance (CBM) is explored. The theory of Margin-Setting, a new pattern recognition method developed by researchers at Alabama A&M University, is documented and its applicability to problems of interest to Army is investigated. Extensive parametric studies of Margin-Setting have been carried out through training and testing the algorithms with real and simulated data. Effects of various parameters including size of the training set, the evolution of prototypes, threshold level, margin value, etc. on the accuracy of the algorithm will be studied for various types of classification problems. The interplay between false positives and false negatives as they relate to system parameters and the application environment will be studied. Algorithms for mapping the data sets of large-scale Bayesian Network graph structures in a parallel and distributed computing environment were researched. In support of the Condition Based Maintenance (CBM) philosophy, a theoretical framework and algorithmic methodology for obtaining useful diagnostic and prognostic data from electro-mechanical systems was developed. The methods are based on vibration and modal analyses of the physical components. To illustrate the concept of the derived process, two "real world" models, a PCI circuit card, and an example rotor hub were considered. Models were created using finite element analysis (FEA) techniques, and analyzed to determine fundamental mode shapes and vibration frequencies. The results yielded vibration modes characteristic of both undamaged and damaged systems. An ?(N?logN) pattern recognition technique was utilized for signal discrimination.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 15, 2008
- Accession Number
- ADA500002
Entities
People
- Andrew E Scott
- Kaveh Heidary
Organizations
- Alabama A & M College