Study of Time-sequenced In-cylinder Engine Flow Fields Prediction

Learned high-speed particle image velocimetry (PIV) techniques to describe the 3D features of direct injection engine in-cylinder flow. Surveyed cutting edge methods of detecting and quantifying the transient vortex characteristics to provide a reliable way of improving temporal resolution in PIV flow data.

Applied K-means and Fuzzy-C-Means clustering algorithm for detection of time-resolved transient vortexpatterns to mitigate the loss of transient flow information based on conventional ensemble flow field analysis.

Predicted the underlying dynamics of the interaction between in-cylinder flows using Long Short-termMemory (LSTM) based bidirectional recurrent neural network (bi-RNN) model. Experiment conducted show that the bi-RNN model can accurately predict the bulk flow and vortex motions from early intake stroke to compression stroke.

Report can be viewd here.