![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
ABSTRAK Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
BAB 1 Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
BAB 2 Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
BAB 3 Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
BAB 4 Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
BAB 5 Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
COVER Calvin Christian Chandra
PUBLIC Irwan Sofiyan ![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
DAFTAR PUSTAKA Calvin Christian Chandra
PUBLIC Irwan Sofiyan
One method of optical flow experimentation is particle image velocimetry (PIV). But one shortcoming is the discreet nature of PIV outputs. In this work, deep learning which includes feedforward neural network (FNN) and physics-informed neural network (PINN) are utilized to provide an alternative that is expected to cover this shortcoming. Three cases will be used to explore deep learning for time resolved PIV which are fluidic pinball, turbulent channel, and flow past a 2-D wing. Based on the methods done in this work, FNN provides a more accurate prediction in terms of velocity. Meanwhile, PINN enables the prediction of pressure data, but has lower velocity prediction accuracy compared to FNN. Other than that, after experimenting with different training parameters, a couple of findings are obtained. The first is that using the Cauchy stress tensor (ST) over velocity-pressure formulation for the Navier-Stokes equations will increase the prediction accuracy of both velocity and pressure when using PINN. The second is that through this study’s parametric study, in general, more training data and a deeper network will increase the accuracy of the trained network.