An extended analytical Bayesian framework for comparison of disparate test articles
Holger M. Jaenisch, James W. Handley · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2013
We present a framework for Bayesian analysis of test articles. For test articles comprised of two datasets, we derive analytical distribution models from the histograms of datasets. We introduce a novel alternative to traditional hypothesis testing that compares the analytical Cumulative Density Functions (CDF). This comparison between two datasets yields the probability that the two datasets are equivalent. If the test articles are two simulations, we derive trigonometric polynomial Data Models of the output and the inputs. Principal component analysis (PCA) reduces the number of modes used to reconstruct the simulation input or output to only the significant contributor. Kolmogorov-Gabor (KG) polynomial Data Models of the reduced mode simulation output are derived as a function of the reduced set of simulation input modes. These KG Data Models are analyzed to determine critical, sensitive, and key parameters. When the simulations have similar structure or when the range, standard deviation, and expectation are within similar ranges, the simulations are labeled as similar. If the test articles consist of a dataset and a simulation, the output from the simulation is first recorded as a dataset and compared to the second dataset test article using the CDF method.