Inverse problems applied to rotating systems, considering parameters uncertainties

• Grant number: 13/07542-3
• Support Opportunities: Regular Research Grants
• Start date: August 01, 2013
• End date: January 31, 2016
• Field of knowledge: Engineering - Mechanical Engineering - Mechanics of Solids
• Principal Investigator: Helio Fiori de Castro
• Grantee: Helio Fiori de Castro
• Host Institution: Faculdade de Engenharia Mecânica (FEM). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

Rotating systems are critical components _ power generation plants, as they are critical to its operation. Therefore, the study and modeling Rotating machinery is extremely important for the development of the country, facing expected energy demand for years to come. _ Rotating machinery models have been developed, where the rotor is modeled by finite element method and the bearings by finite differences or finite volume methods. These rotating systems have _ stochastic characteristic, which makes the identification of parameters by model updating (solution of inverse problems), when considered deterministic model responses. This project proposes the use of model updating techniques _ that take into account the stochastic characteristic of the system. The first method is the application of multi-objective optimization, in order to obtain a feasible region of possible responses (Pareto set). Thus, we can evaluate the effect of model uncertainties on the variability of the Pareto optimal set. The optimization methodology proposed is based on evolutionary algorithms, such as MOGA (Multi-objective Genetic Algorithm) and SPEA (Strength Pareto Evolutionary Algorithm). Another proposed approach is application of Bayesian inference. In this case the parameters to be identified can be related to the posterior distribution and its initial information to the prior distribution. The difference between the model response and experimental results is represented by the likelihood function. The solution to this method can be obtained through a numerical search method Monte Carlo Markov Chain (MCMC). For the application of model updating, experimental data from a test-rig will be obtained, where it will be considered modal parameters, frequency response function and unbalance response. (AU)