Health diagnostics using multi-attribute classification fusion

Abstract

This paper presents a classification fusion approach for health diagnostics that can leverage the strengths of multiple member classifiers to form a robust classification model. The developed approach consists of three primary steps: (i) fusion formulation using a k-fold cross validation model; (ii) diagnostics with multiple multi-attribute classifiers as member algorithms; and (iii) classification fusion through a weighted majority voting with dominance approach. State-of-the-art classification techniques from three broad categories (i.e., supervised learning, unsupervised learning, and statistical inference) were employed as member algorithms. The diagnostics results from the fusion approach will be better than, or at least as good as, the best result provided by all individual member algorithms. The developed classification fusion approach is demonstrated with the 2008 PHM challenge problem and rolling bearing health diagnostics problem. Case study results indicated that, in both problems, the developed fusion diagnostics approach outperforms any stand-alone member algorithm with better diagnostic accuracy and robustness.