Prediction of Scour Downstream Regulators Using ANNs

Regulators is considered one of the main irrigation structures that is used for many purposes. One of the main purposes of regulators is to measure and control the discharge of rivers; also it is used to control the water levels and to generate power. Scour is an inevitable problem that occurs downstream regulators. Different researchers tried to predict scour hole downstream regulators, but their results always gave scour dimensions less than that actually occur. Scour was studied either on solid bed by means of velocity distribution or on movable bed by investigating the topography of the scour hole. In this paper, the scour hole was studied on movable bed by using a new technique rather than the traditional techniques used by other researchers. Due to the complex and unexpected behaviour of water as well as sediment movement downstream regulators, a Back-Propagation Neural Network (BPN) model was developed to predict the dimensions of the scour hole formed downstream regulators, in order to overcome the problem of exclusive and non-linear relationships. A Three layered feed forward neural network using Levenberg-Marquardt algorithm was formulated. The inputs to the (BPN) model were obtained through an extensive experimental program carried out on a trapezoidal channel 0.0001 bed slope. The study covers free and submerged hydraulic jump conditions in both symmetrical and asymmetrical under-gated regulations. It was found that the scour hole dimensions in case of submerged hydraulic jump is always greater than the free one, also the scour hole dimensions in asymmetrical operation is greater than symmetrical one. From the comparison between the experimental results and the predicted ones by the (BPN) model, we found that the scour hole dimensions can be efficiently predicted using (BPN).