A novel approach based on neural network to model the free convection within a differentially heated cavity partitioned by interior curved slats

Abstract

A vertical closed cavity is presented by a modified configuration wherein, the cavity is partitioned by horizontal curved non-conductive slats. Then, the free convection flow of air in the modified differentially heated partitioned cavity is experimentally investigated as a function of governing variables. The cavity is bounded by two hot and cold isothermal parallel walls along with four non-conductive walls. The governing variables consist of the Grashof number (1 × 104≤Gr ≤ 2.05 × 104), slat's tilt angle (0o≤θ ≤ 180o), slat's curvature (0≤Cb ≤ 4 mm) as well as distance between the isothermal parallel walls (15.8 mm ≤ W ≤ 18.8 mm). After that, an optimal combination of the teaching–learning-based optimization (TLBO) and wingsuit flying search (WFS) algorithms is proposed as a new combinatorial optimization algorithm. The proposed combinatorial algorithm is tested over some standard benchmark functions and then integrated with the artificial neural network (ANN) to construct a novel hybrid model. In the following, the developed hybrid model is utilized for forecasting the average Nusselt number (Nuavg) of the cavity, as a function the governing variables. It was found that, the proposed combinatorial algorithm enhances the forecast effectiveness of the main ANN. Moreover, the creation of curvature on the slats, results in the suppression up to 34.35% in the heat transfer within the cavity.