Missing data recovery using data fusion of incomplete complementary data sets: A particle image velocimetry application

A data-fusion approach is reported to reconstruct missing data and is applied to particle image velocimetry (PIV) measurements. This approach departs from the existing ones in that the datasets involved in its operation are incomplete. Two sets of incomplete but complementary data with fault regions, obtained using different measurement setups, are combined to yield a complete dataset by reconstructing the missing data. In this report, the capability of the current approach is first demonstrated by using three fabricated scalar patterns with different frequency spectra. Second, this method is applied to PIV measurements pertaining to the natural wake of a circular cylinder with a Reynolds number ReD = 1.8 × 103. The performance of this approach is also examined under different configurations, size, location, and direction, of the fault regions. For the real-world data with turbulence and fluctuations, this approach encounters an overfitting problem. To employ this approach in real-world applications without ground-truth data, a method is also proposed to avoid the overfitting problem and estimate reconstruction accuracy. Then, the method is applied on a spectrally richer flow, i.e.