The PHM community has long been dedicated to the advancement of various techniques for carrying out prediction of remaining useful life. Compared to top-down modelling provided by the conventional first-principle model-based approaches, data-driven approaches and ensemble data-driven models offer a new paradigm of bottom-up solution for predictive asset health management. The challenge still remains that accurate prognosis has eluded most to date. Data mining and Artificial Intelligence may offer some insight but these techniques required an understanding of correlations of numerous operational parameters.
Canonical Variate Analysis (CVA), typically applied to dynamic processes, is a dimensionality reduction technique that can be applied to machine condition monitor of rotating machinery. This method maximizes the correlation between combinations of any process measurements, including vibration data, thereby generating accurate health and prognostic indicators from the serially correlated data.
Examples will be given to demonstrate the actual application of CVA on operational machinery using process and vibration data for fault detection of compressors, fault identification and remaining useful life prediction of pumps and turbofan engines.
De Montfort University, UK
Professor David Mba is currently the Pro Vice-Chancellor and Dean of Computing, Engineering and Media at De Montfort University, UK. He is a leading authority in machine condition monitoring, diagnosis and prognosis. He has contributed to the development and publication of international standards in the subject area of Acoustic Emission and Vibration diagnosis. Professor Mba has written over 300 technical papers and his current research is focused on machine fault diagnosis, model based prognostics and machine performance prediction