Uma abordagem alternativa para avaliação da convergência do método de Monte Carlo adaptativo do GUM S1

Abstract

This dissertation presents a new approach to the measurement uncertainty evaluation in order to solve the problems found in the Adaptive Monte Carlo method (AMC), mainly about the loss of stability. This new approach, the Sequential Monte Carlo method (SMC), is based on the same principles guided by GUM and GUM S1 documents and aims to expand the scope of the proposed methods in these documents. When compared with the AMC, the SMC has two modifications; the first one is related to the storage of data resulting from simulation; and the second consists in applying a new convergence criterion. The SMC has been applied in order to estimate the uncertainty of measurements made with a caliper, external micrometer, profile projector and roughness meter. The SMC effectiveness was validated by the evaluation of the repeatability of the simulation results and through the comparison of the uncertainty values and those resulting from the application of the GUM, MC and AMC methods. Furthermore, the SMC showed greater repeatability when was compared to AMC as regards the number of necessary iterations to achieve the convergence of the simulation. This fact attributes to this method higher reliability in comparison with AMC. The new way of storing data resulting from the simulation method used in SMC decreased significantly the amount of manipulated data, and consequently extended the scope of the AMC method.