QUANTIFICATION OF EXPERIMENTAL AND COMPUTATIONAL UNCERTAINTIES FOR MEDICAL DEVICES USING PROBABILISTIC METHODS

摘要

Uncertainties are inherent to physical testing of medical devices. They need to be accounted for in computational models in order to achieve proper degree of validation. These uncertainties may arise due to various factors like the applied boundary conditions, the behavior measurement and the geometry and material properties of a physical prototype. On the other hand, computational models have different types of uncertainties that are mostly related to modeling assumptions and refinement of the physical phenomena modeled as well as uncertainties with respect to the solver and numerical discretization. Previously (Van der Velden et al. 2012), it has been demonstrated how uncertainties such as simulated behavior, manufacturing tolerances, material and loading condition variations could be incorporated into computational models. A key conclusion from the study was that even though there are a near infinite number of uncertainties impacting measured or computed behavior, for non-chaotic systems the behavior uncertainty can still be quantified with a limited number of factors using a single metric called the Probabilistic Certificate of Correctness (PCC). This metric computes the probability that the actual physical prototype will meet its benchmark acceptance tests, based on virtual prototype behavior simulations with known confidence and verified model assumptions. This metric has recently been adopted by DARPA in its Adaptive Vehicle Make Program (DARPA 2012).