Rivers provide water and energy for humans and the nature and it could be stated that providing water is the most crucial economic role of rivers. Water is deviated from its main course through the intakes in order to fulfill various purposes such as for agriculture, urban water supply, electricity generation and etc. Constructing the water intakes is one of the ancient cheapest methods to use the river water for different purposes. In the past, the gravity was used to collect the water from the rivers but it has now turned into an evolved hydraulic structure with designing criteria. Considering the fact that the river flow includes fine- and coarse-grained sediments and that the river regime changes during the floods, the intake’s inlet mouth must carry out the two essential duties: absorbing and controlling the flow deviated from the river and directing it into the intake channel and preventing the sediments and floating objects from entering the intake. Therefore, one of the crucial points to be considered in designing an intake located within a river is to select the conditions under which the water deviated by the intake will have maximum flow discharge and minimum sediment discharge. Using the new method, known as the soft computing, has gained significant popularity within the past decade due to the complexity of most engineering problems. One of the advantages of this method is its efficiency and desirable level of accuracy in solving complex and challenging engineering problems therefore, it brings accuracy and speed to the works of researchers. Numerous studies have been conducted to evaluate the flow in different types of open channels and hydrologic and hydraulic phenomena using the soft computing. The obtained results indicate the high accuracy of this method in solving complex water and hydraulic structures despite the few registered researches done on dividing open channel flow modeling. The present research examines the effect of the channel roughness on flow hydraulics and predicts the velocity in the intakes using a combination of artificial neural network and numerical model.
The artificial neural network is an idea based on the human mind, which is widely used in solving complex problems in different sciences. An artificial neural network generally comprises a number of connected nodes (known as neurons) including an input layer, a number of hidden layers and an output layer, each layer is made up of a number of neurons. Considering that the number of the hidden layers neurons is excessively great, the network will take an unacceptably long time to train for each value. An ANN model is presented in this study in order to predict the flow mean velocity. It is essential to model and verify a numerical model in order to train the model under flow conditions where there are no available experimental data. Ramamurthy et al.’s (2007) experimental model has been used in this study in order to verify the numerical model. The experiments have been conducted within rectangular channels. The main channel is straight while the branch channel is attached to the main channel with an angle of 90 degree. Therefore, Ramamurthy et al.’s (2007) experimental model has been simulated using the ANSYS-CFX software and verified through using the existing experimental data. After assuring the simulation accuracy in the CFX, different models have been conducted using the CFX and are used in training and verifying the ANN in order to predict the flow mean velocity.
The verification results, using a MAPE by 5 percent average error, indicates the high accuracy of the generated numerical model. After verifying the CFX model’s results and assuring that they were highly accurate, measures were taken to design five artificial neural network models with five different roughnesses by using the numerical and experimental data. The longitudinal velocities were predicted using the ANN model in cross sections, which possessed no experimental data, and then they were compared with the CFX model’s results. The result of this comparison indicates that the ANN model is highly accurate to predict the flow velocity in different areas of the intake channel moreover as the channel wall roughness increases from 0.0 to 0.0005 m, the flow longitudinal velocity drops significantly in most areas of the branch channel however as the roughness increases from 0.0005 to 0.001 m and more, the longitudinal velocities slightly drop. As the channel wall becomes rougher, the size of the flow separation zone decreases and the curve of the streamlines and the flow turbulence also decrease at the entrance of the branch channel