An Artificial Neural Network and Taguchi Method Integrated Approach to Predicting the Local Scour Depth around the Bridge Pier during Flood Event

Experiment design is believed to be an important part of investigating an engineering phenomenon for characterizing and optimizing the process. In this study, the Taguchi method (TM) reduced the number of experiments and was used to analyze the results of an artificial neural network (ANN) and find the optimal combination of the relevant parameters in the ANN. Accordingly, the phenomenon of the local scour depth around the bridge during flood events was considered as a case study. The study results indicated that TM could reduce the number of experiments compared to the previous original study and the full factorial method by 28% and 67%, respectively. According to TM, the flow intensity at the hydrograph peak was the most effective parameter providing the optimal state (minimum scour depth). Additionally, an ANN with three hidden layers and the main parameters, including several neurons in the first and second hidden layers, training function, and transfer function, was introduced. Adjusting the input parameters of the ANN, TM led to the emergence of networks with a reasonable correlation coefficient of R= 0.952. Finally, the results demonstrated that the transfer function had the most significant effect on the results of the ANN.