Prediction of coarse particle movement with adaptive neuro-fuzzy inference systems

The use of a neuro-fuzzy approach is proposed to model the dynamics of entrainment of a coarse particle by rolling. It is hypothesized that near-bed turbulent flow structures of different magnitude and duration or frequency and energy content are responsible for the particle displacement. A number of Adaptive Neuro-Fuzzy Inference System (ANFIS) architectures are proposed and developed to link the hydrodynamic forcing exerted on a solid particle to its response, and model the underlying nonlinear dynamics of the system. ANFIS combines the advantages of fuzzy inference (If-Then) rules with the power of learning and adaptation of the neural networks. The model components and forecasting procedure are discussed in detail. To demonstrate the model's applicability for near-threshold flow conditions an example is provided, where flow velocity and particle displacement data from flume experiments are used as input and output for the training and testing of the ANFIS models. In particular, a Laser Doppler velocimeter (LDV) is employed to obtain long records of local streamwise velocity components upstream of a mobile exposed particle. These measurements are acquired synchronously with the time history of the particle's position detected by a setup including a He-Ne laser and a photodetector. The representation of the input signal in the time and frequency domain is implemented and the best performing models are found capable of reproducing the complex dynamics of particle response. Following a trial and error approach the different models are compared in terms of their efficiency and forecast accuracy using a number of performance indices