ANN to correct coarse-mesh CFD errors in cell culture bioreactors

This data was generated from CFD models of a spinner flask developed in COMSOL with two different impeller geometries and varying rotational speeds. The 1st derivative, 2nd derivative, and cell Reynolds data are used as inputs to train the ANN model. The fine and coarse velocity data are used to calculate the mesh error and used as training output. Then, the trained ANN is used on a set of testing data to predict mesh errors and correct the test coarse data. Results are saved as a correlation plot between corrected and coarse data against fine data and also as a table containing correlation coefficients, MSE, RMSE, and normalized RMSE. If the corrected results have lower RMSE than the coarse data, the ANN model has worked properly in correcting the coarse data.