On-machine Measurement Error Modeling And Predicting Research Based On Bayesian Network

The main content of this dissertation is the model of on-machine measurement error basedon Bayesian Network. On the basis of deeply analyzing the current status of research andapplication of on-machine measurement error and the application effect of Bayesian Networktheory in the filed of uncertainty, Bayesian Network modeling is proposed.Bayesian Network is one of the most important theories to dealing with uncertainty problem.It has complete mathematic support and is based on the casual and probabilistic theory. Becauseof its concise and visualization knowledge representation, powerful learning and reasoningability, it is applied in many domains. So, the on-machine measurement error modeling is studiedbased on the Bayesian Network theory. The following works are done:1. Bayesian Network theory is systematically studied, including constitution of BayesianNetwork, parameter learning of Bayesian Network and reasoning of Bayesian Network, andprovides a clear thinking to build the model of on-machine measurement error based onBayesian Network.2. The main error sources of on-machine measurement system are obtained after the processof on-machine measurement is studied experimentally. On this basis, theoretical researchmainly on the influence of probe system, grating measuring system, deformation of body andthe vibration of rail support are carried out, and has fully understanding about the variablesneeded for modeling.3. The model of on-machine measurement error based on Bayesian Network is establishedand priori probability distributions of each node in the model are assigned with the principle ofuniform distribution. Besides that, all functional modules of Bayesian Network model aredesigned through MATLAB.4. The high-accuracy measurement error data gathering experiment system is developed andmeasurement error sample database is established through experiment. Then，the parameterlearning and the analysis of model reasoning accuracy are complete based on the sample, andthe result compared with the BP neural modeling verified the superiority of Bayesian Networkmodeling.