Incipient particle entraiment prediction with the use of machine learning methods

In natural water bodies, sheared turbulent flows are the forcing agent responsible for particle mobilization near the river bed surface. Several analytical approaches have been used to describe this phenomenon, with ambiguities in the analytical methods employed, resulting in methodological biases. The application of a machine learning technique, namely, Adaptive Neuro-Fuzzy Inference System (ANFIS), is proposed here to model sediment transport dynamics. It is hypothesized that turbulent flow of different magnitudes and sufficient duration or near bed instantaneous flow power is responsible for particle displacement. The entrainment of sediment is modeled using the dynamic incipient motion criteria of impulse and energetic turbulent flow events. Several ANFIS architectures have been developed to relate the hydrodynamic vectorial quantities to particle displacement. ANFIS combines artificial neural networks' adaptation and learning power with the advantage of fuzzy inference (IF-THEN) rules for knowledge representation. To demonstrate ANFIS applicability for near bed threshold conditions, streamwise velocity [1], and particle dislodgement [2], flume-based experimental data sets are obtained as input and output signals to train the ANFIS model of various architecture complexities. The energy-based criterion and impulse criterion are obtained as cubic and quadratic expressions of streamwise velocity, respectively, and they are also used as inputs to train the ANFIS model [3]. Following a trial and error approach, the models developed with these criteria are analyzed and compared in terms of their efficiency and predictability using several performance indices. The optimum performing model is found capable of replicating the complex dynamics of sediment transport.

References
[1] Liu, D., AlObaidi, K., Valyrakis, M.* (2022). The assessment of an Acoustic Doppler Velocimetry profiler from a user’s perspective, Acta Geophysica, 70, pp. 2297-2310. DOI: 10.1007/s11600-022-00896-3.
[2] AlObaidi, K., Valyrakis, M. (2021). Linking the explicit probability of entrainment of instrumented particles to flow hydrodynamics, Earth Surface Processes and Landforms, 46(12), pp. 2448-2465 DOI: 10.1002/esp.5188.
[3] Valyrakis, M., Diplas, P., Dancey, C.L. (2011). Prediction of coarse particle movement with adaptive neuro-fuzzy inference systems, Hydrological Processes, 25(22). pp.3513-3524, DOI:10.1002/hyp.8228.