An Automated Proper Orthogonal Decomposition-Based Post-processing of In-Cylinder Raw Flow Datasets

Laser-based diagnostic techniques like particle image velocimetry (PIV) and molecular tagging velocimetry (MTV) are used to measure flow fields at a high spatial resolution. Post-processing of the obtained flow fields is essential for outlier correction as the datasets may be skewed by local flow vectors with a disproportionate disparity in magnitude or directions from neighborhood vectors. The rationale behind this work is to propose an algorithm using proper orthogonal decomposition (POD), namely, POD-OROC (POD-based outlier removal and outlier correction), which can correct outliers in an ensemble of flow fields. The proposed algorithm is first validated on synthetic flows with a known percentage of outlier rate and then applied to engine in-cylinder flow fields. The algorithm ran for a few iterations for both flow datasets and rejected frames with high outlier rates (above 15%) and then post-processed the remaining ones to detect and correct local spurious vectors. It was found that outlier vectors with larger deviation from neighboring vectors are detected in earlier iterations. An error analysis was performed to quantify the total error in an ensemble and, in using it, two types of errors—over-detection and under-detection—were identified. With this insight, several parameters of the model for synthetic flows were optimized for best performance, and then the model was modified for application to in-cylinder flows. The impact of POD-OROC was studied through changes in the POD energy spectra where the energy share of the first mode increased to 99.9% for synthetic flows and to 82.5% and 68.9% for the two in-cylinder flow sets. Finally, POD-OROC is now matured enough to be applied to in-cylinder flow datasets and can detect and correct both single and cluster outliers.