Sensor and CFD data fusion for airflow field estimation

This paper presents how an airflow field can be quickly estimated by fusing isolated and sparse sensor observations with a given computational fluid dynamics (CFD) solution. The given CFD data represent the airflow field with a certain boundary condition while the sensor data provide sparse and sampled information of the actual field with a new/changed boundary condition. Essentially, we need to determine a dominant proper orthogonal decomposition (POD) basis from a limited number of snapshots obtained from CFD simulations of a spatial domain. Then, for the given CFD solution, we estimate its difference from the actual field based on the sensor observations and the dominant POD basis. With the estimation result, the given CFD solution can then be efficiently corrected based on new sensor information. That is, with sparse sensor observations, we are able to transfer the given CFD solution into an 'actual' one that corresponds to the new boundary condition. Additionally, we present a new greedy algorithm to determine the sensor placement, which plays a critical role in airflow field estimation. Finally, a simple example is provided to illustrate the effectiveness of the proposed sensor-CFD data fusion procedure and the new sensor placement algorithm.