An optimized Adaptive-Neuro Fuzzy Inference System (ANFIS) for reliable prediction of entrance length in pipes

In this paper, an attempt has been made to predict and evaluate the entrance length in pipe for low Reynolds number flow using evolutionary-optimized Adaptive-Neuro Fuzzy Inference System (ANFIS). For optimization purpose, evolutionary algorithm namely particle swarm optimization (PSO) is adopted. We trained and tested the model using 100 experimental records from computational fluid dynamics (CFD) technique that established the basic dataset under various working conditions such as air and water. This experimental data set is included input parameters namely Reynolds number, pipe diameter, and inlet velocity and entrance length as output parameter. The dataset is divided to two part for training and testing with 87 and 13 data number respectively. The structure of developed PSO-optimized ANFIS model in comparison with another models is very simple such that the model is composed of 3 membership functions for each input and only 27 rules in rule base. Evaluation of predicted entrance length values obtained by the optimized model was performed by using indicators such as coefficient of determination (R2) = 0.9966 and Root Mean Square Error (RMSE) = 0.011 that prove satisfactory efficiency of this model. The model can also be used for prediction of online entrance length without any constraint in selection of data points or training phase.