Author(s): Manousos Valyrakis; Panos S; Clinton Dancey
Linked Author(s): Manousos Valyrakis
Keywords: No Keywords
Abstract: This paper reviews the utilization of different softcomputing techniques for the prediction of entrainment of a coarse particle by rolling. A set of variable complexity Artificial Neural Networks (ANN) and Adaptive NeuroFuzzy Inference Systems (ANFIS) are developed and employed for the short time prediction of entrainment events of individual particles. Each of the models are trained and validated using flow velocity and particle displacement data from a number of mobile grain flume experiments. The models use different representations of lagged time series of the flow velocity component upstream of a mobile exposed particle-obtained with a Laser Doppler Velocimeter (LDV) -as input, while the particle's displacements are tracked synchronously via specifically designed optical arrangements. The developed models perform well in predicting the entrainment dynamics of the exposed particle, validating the hypothesis that the recent history of streamwise local velocity component suffices to predict particle response such as displacement and deposition. The best performing models are compared in terms of their efficiency and forecast accuracy using certain performance metrics. It is also demonstrated that ANFIS can aid provide a better understanding for the phenomenon of particle entrainment as a dynamical process, in agreement with recently introduced theoretical models outlining the importance of magnitude and duration of energetic flow events.