digilib@itb.ac.id +62 812 2508 8800

Biomimicry has been one the most interesting scientific topics that might lead to groundbreaking technologies, and particle image velocimetry (PIV) is one of the instruments to observe it. Hence, developing a state-of-the-art motion estimator program to perform PIV should be a promising option to support it. Up until now, the cross-correlation with window deformation iterative method (WIDIM) is the go-to method to perform PIV image processing because of its fast computation and decent accuracy. Another option for it is using the optical flow method or the variation of it. The optical flow can estimate a detailed velocity field even for complex flow cases, but it took too long to process. As an emerging technology, using a deep learning approach for motion estimation should be an interesting option. It is later proven that a PIV image processing program with the LiteFlowNet deep learning model can provide a robust solution for either simple or more complex flow pattern. It also provides a much faster computing time, that would be helpful for handling experiments with numerous images. A PIV experimental setup to measure and investigate the flow around a flapping wing will be discussed also to provide a real test to the motion estimator program. Finally, it is later shown that the developed motion estimator program is able to diagnose the flow behaviour of a root canal irrigation with eddy tip activation without the help of a PIV experimental setup. Thus, providing a novel approach in the root canal irrigation quantitative analysis through visual images.