2023 年 71 巻 5 号 p. 209-217
Uncertainty quantification (UQ) of a numerical model of JAXA 6.5m × 5.5m Low-speed Wind Tunnel (LWT1) is conducted to realize model-based anomaly detection. The uncertainty evaluated by UQ is used to estimate the normal space of LWT1, and anomaly detection can be achieved by comparing the measurements and the estimated normal space. In the operating condition with the wind velocity in the test section of 9.7m/s, the 95% confidence interval of the predicted wind velocity is ranging from 5.50 to 9.80m/s, which is too large to realize anomaly detection with sufficient accuracy. Uncertainty management (UM) is conducted for reducing the uncertainty of the numerical model. Sobol's method is employed to find explanatory variables sensitive to the objective variable, and then, an additional experiment is conducted to reduce the uncertainty originating in those variables. The 95% confidence interval of the predicted wind velocity can be reduced to the range of 7.61 to 9.80m/s. Successful reduction of uncertainty is achieved through UM.