عنوان مقاله [English]
Several studies have been performed to study the scouring depth at bridge
piers. Due to the complication of the problem and variety of the hydraulic and
geometric parameters affecting the scouring phenomena, a generalized
relationship has not been presented yet. Therefore, adaptive neuro-fuzzy
inference system (ANFIS) is an alternative to overcome these problems. This
approach is an effective tool to provide the hydraulic engineers, precise estimation of the scouring depth around the bridge piers. Although a large number of former studies have just focused on scouring at bridge piers under steady flow condition and uniform-graded bed materials even by applying ANFIS model, a lack of studies exists on scouring under unsteady flood flow condition as well as for non-uniform bed materials. Generally, river beds are composed mainly of non-uniform materials. Motion of the finer sediment particles initiates results in the protective effect of greater particles, namely armoring effect on the bed surface, thereby eliminating further erosion of the bed. Furthermore, in most of the rivers the flow regime is commonly unsteady. During a flood, the maximum scouring depth regarding to the peak of the flood hydrograph would be smaller than the equilibrium scouring depth which is commonly estimated using a constant flow discharge. When the flow unsteadiness is pronounced, the difference between the maximum scouring depth and the equilibrium scouring depth is quite substantial and thus should be addressed.In the present study, armoring effect on local scouring under unsteady flow condition was investigated based on a comprehensive dataset collected by different former investigators using ANFIS model. For this purpose, two different models were constructed. The first model was based on 372 dataset collected in a practical study on different bridges in USA. Measurements were conducted under steady flow condition. The second model was developed for estimating the maximum scouring depth in the beds of uniform and armored materials under unsteady flow condition. To present a more accurate model, some strategies including; reduction of dimension and detection of outlier were used to improve the performance of calculations. Genetic algorithm and particle swarm optimization methods were applied to develop a novel hybrid learning algorithm for ANFIS model. The new hybrid learning algorithm train the
antecedent part of the fuzzy rules. Then the least square method was applied
for training the conclusion part of the rules. It was shown that ANFIS model
gives more accurate results compared to the empirical equations. Results
highlighted the effectiveness of the data on the estimations of ANFIS model.
Furthermore, according to the results, this approach is potentially able to train the ANFIS model in both steady and unsteady flow conditions.