Prediction of flow field in a solar chimney using ANFIS technique

Abstract

Solar chimneys have been intensively studied as an effective method for natural ventilation of buildings. Though numerical methods, such as Computational Fluid Dynamics (CFD), have been widely utilized in such studies, they usually require extensive computational resources. Moreover, experimental study is quite complicated and costly. In recent years, machine learning has started to be used as a tool in the thermal-fluid field. In this study, in order to save time and cost, Adaptive Neuro-Fuzzy Inference System (ANFIS) technique, a class of adaptive networks that incorporate both neural networks and fuzzy logic principles, is combined with CFD. A simulation model was first validated by experiment from another study in the field. The result was documented as a dataset using CFD code ANSYS Fluent (Academic version 2020 R2). Then, they are used to train and validate the ANFIS model. In particular, the study is to predict the fluid flow field in a 2-dimensional typical solar chimney when heat flux changes in the range of 400 to 1000 W/m2. Inputs of the ANFIS model are position and heat flux, while outputs are temperature and velocity at that location. As a result, the 2 ANFIS models could achieve R2 values of 0.997, 0.97 (training set) and 0.994, 0.9715 (testing set); RMSE are 1.009, 0.00224 (training set) and 1.074, 0.0204 (testing set) for outputs of temperature and velocity, respectively. Those results are acceptable. By using the ANFIS model, large amounts of flow fields with different scenarios can be estimated simultaneously. Therefore, it is expected that engineers and architects can have a quick tool in the process of design.