Point-Collocation NIPC 기법 기반 천이 난류 모델 계수의 베이지안 추론 연구 (BAYESIAN INFERENCE OF TRANSITIONAL MODEL COEFFICIENTS BASED ON POINT-COLLOCATION NON-INTRUSIVE POLYNOMIAL CHAOS METHOD)

서지정보

· 발행기관 : 한국전산유체공학회
· 수록지 정보 : 한국전산유체공학회지 / 25권 / 2호 / 111 ~ 118페이지
· 저자명 : 배재현, 장경식, 호아이탄 응우옌, 김병천

초록

In the present work, Bayesian inference based on the point-collocation Non-intrusive Polynomial Chaos was conducted in transitional flow around the flat plate using Menter’s γ-Reθ transition model. Three model coefficients, Ca2, Ce1 and Ce2 were considered for random variables based on the sensitivity analysis of the present turbulence model and quantity of interest was set to the drag coefficient, Cd. With the assumption of uniform distribution of three model coefficients within ±10%, the surrogate model was obtained based on the general Polynomial Chaos expansion. The simulation results were used for observation to calculate the likelihood function and MCMC sampling algorithm was adopted for Bayesian inference. The correlation between Ce1 and Cd was predicted with highest value than other two model coefficients. Based on the original Bayesian inference result, the effect of number of observation data was studied. Also the 2nd order of gPC for surrogate model was adequate through comparison with results of 3rd order gPC. Finally, two prior distributions of input random variables for Bayesian inference were considered and the differences of posterior distributions of input and output were investigated.

영어초록

In the present work, Bayesian inference based on the point-collocation Non-intrusive Polynomial Chaos was conducted in transitional flow around the flat plate using Menter’s γ-Reθ transition model. Three model coefficients, Ca2, Ce1 and Ce2 were considered for random variables based on the sensitivity analysis of the present turbulence model and quantity of interest was set to the drag coefficient, Cd. With the assumption of uniform distribution of three model coefficients within ±10%, the surrogate model was obtained based on the general Polynomial Chaos expansion. The simulation results were used for observation to calculate the likelihood function and MCMC sampling algorithm was adopted for Bayesian inference. The correlation between Ce1 and Cd was predicted with highest value than other two model coefficients. Based on the original Bayesian inference result, the effect of number of observation data was studied. Also the 2nd order of gPC for surrogate model was adequate through comparison with results of 3rd order gPC. Finally, two prior distributions of input random variables for Bayesian inference were considered and the differences of posterior distributions of input and output were investigated.

참고자료

· 없음