Evaluation Of Measurement Uncertainty Based On Bayesian Information Fusion

With the development of science and technology, many fields and departmentsrequire a more adequate, complete and scientific measurement result. More and moreoccasion begin to believe that uncertainty of measurement must be included inmeasurement result. Therefore increasingly importance has been attached to theory ofuncertainty, which makes it a quite active field. And many new calculation and theory ofuncertainty are being put forward.According to principle of GUM measurement uncertainty, this paper studies onuncertainty with Bayes method, based on information fusion theory, to combine unity ofA and B evaluation methods, static and dynamic evaluation method. This method makesfull use of historical information and the current measurement data. To determine priordistribution by historical information and posterior distribution is deduced byintegrating prior distribution and current measurement data with Bayes model, and thisposterior distribution is used to estimate the uncertainty.In the view of normal distribution on sample information, characteristics of Bayesposterior are estimated on such separate situation, mean value is known but variance isunknown and both mean value and variance are unknown. As uncertainty componentshave different distributions, conjugate distribution method is adopted to determine priordistribution, uncertainty of measurement formula is derived from non-normaldistribution (uniform distribution, triangular distribution, T distribution) respectively.Since new information can be continuously integrated during measurement processwith Bayes information fusion method, therefore this method has got a very goodtiming characteristic. So Bayes dynamic uncertainty evaluation method is beingpreliminarily discussed on the basis of Bayes static uncertainty evaluation. To estimateuncertainty evaluation formula by adopting Bayes dynamic linear ordinary averagemodel, and analyze dynamic linear model component. Dynamic measurementuncertainty evaluation can be conducted by modeling for ergodic process.Based on estimated uncertainty formula of the Bayes information fusion, instanceanalysis has been carried on by adopting coordinate measuring machine to conductexperiment on uncertainty of measurement for aperture. Repeatability andreproducibility of the experimental results are concentrated to be estimated by Bayesuncertainty components, and finally aperture estimated measurement uncertainty isobtained.