Date of Award
Master of Science
All current engineering prognostic practices require prior off-line tests. These are needed to: (1) Determine the exact conservative principle or utility function being satisfied, and (2) Determine associated material, geometric and process parameters. In addition, prediction of onset of instability or failure requires a failure criterion. The data driven prognosis (DDP) approach, developed here, obviates the need for such off-line testing and facilitates true predictive capability using only on-line data being sensed. To achieve this end, the DDP algorithm makes an assumption regarding polynomial order of the potential or utility function in the neighborhood of each observation points. Thus, an assumption regarding the local piecewise behavior replaces any global assumption. The needed system parameters in dimensionless forms are then estimated based on prior data or experience from the same experiment. A multi-physics model based on the concept of excess curvature is then developed to predict short-term and long term stability profiles of any system. The model is first validated against simple "Balloon Burst" experiment and later used for analyzing "Gulf Stream" and "Economics" systems. The proposed DDP algorithm may be used for general conservative systems provided the variables involved in the conservation principle are observable. The developed multi-physics model also provides an objective basis for data driven prediction of system stability and associated decision making in various mechanical, economic and societal systems.
Kar, Oliva, "Data Driven Prognosis: A Multi-Scale And Multi-Physics Approach" (2012). Graduate Theses and Dissertations. 12879.