An artificial neural network model for non-iterative calculation of the friction coefficient in open channel flow

A non-iterative procedure tvas developed using an artificial neural network (ANN) for calculating the friction coefficient C, in the Chezy equation as applied to flow in open channels. The Regula-Falsi method was used as an i mplicit solution procedure to estimate the C values for a range of Reynolds numbers, Re, and relative roughness (e/R) values (where e is the absolute roughness of the channel, and R is the hydraulic radius). in developing the ANN model, two configurations were evaluated: (I) the input parameters Re and e/R were taken initially on a linear scale; and (2) both input paramete rs (Re and e/R) were transformed do a logarithmic scale. The second configu ration yielded an optimal ANN model with 10 neurons in each of the three hi dden layers. This configuration was capable of predicting the values of C i n the Chezy equation for any given Re in the range of 2 x 10(3) to I x 10(8 ) and e/R in the range of I x 10(-6) to 5 x 10(-2). These values were in cl ose agreement with those obtained using the numerical technique. The propos ed model is superior to existing approximations of the Powell equation sinc e it is applicable to the entire spectrum of the turbulent flow regime in o pen channels. It offers significant advantages when dealing with flow probl ems that involve repetitive calculations of the friction coefficients, such as those encountered in the hydraulic analysis of surface irrigation, over land flow! and unsteady flow in open channels.