Page 48 - CBT 2018
P. 48
46
Uncertainty quantification for collisions
In the reconstruction of automobile crashes, a continuing topic of interest is how to combine event data recorders (EDR) with other information. One approach has been to apply non-linear least-squares (NLS) minimization to two-vehicle crashes to estimate impact speeds from different combinations
of EDR, crush surface, and exit angle measurements. However, NLS does
not readily support quantification
of the uncertainty associated with
these estimates. Bayes Theorem from probability theory describes how a prior assessment of uncertainty regarding a quantity x should be updated to account for new information y. The computations needed to apply Bayes Theorem to practical cases can be carried using Markov chain Monte Carlo simulation. CEGE researchers have shown that Bayesian uncertainty quantification is feasible and readily supports fusion of different data sources. [90–92]
