Evaluation of artificial neural network in data reduction for a natural convection conjugate heat transfer problem in an inverse approach: experiments combined with CFD solutions

• Published in: Sadhana (Bangalore) Vol. 45; no. 1
• Main Authors: Kumar, M K Harsha, Vishweshwara, P S, Gnanasekaran, N
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.12.2020; Springer Nature B.V
• Subjects: Aerodynamics; Aluminum; Artificial neural networks; Computational fluid dynamics; Computing costs; Data reduction; Domains; Engineering; Experiments; Free convection; Heat; Heat flux; Heat transfer; Inverse estimation; Inverse problems; Low carbon steels; Mathematical models; Metal plates; Neural networks; Numerical models; Temperature distribution; Thermocouples; Working fluids; estimation; ANN; inverse; forward model; natural convection; fin; Heat transfer
• ISSN: 0256-2499; 0973-7677
• DOI: 10.1007/s12046-020-1303-x

In this work, natural convection fin experiments are performed with mild steel as the fin and an aluminium plate as base. The dimension of the mild steel fin is 250 mm × 150 mm × 6 mm and the aluminium base plate is 250 mm × 150 mm × 8 mm. A heater is provided on one side of the aluminium base plate and the mild steel fin emerges on the other side of the plate. The heater provides required heat flux to the fin base; several steady-state natural convection experiments are performed for different heat fluxes and corresponding temperature distributions are recorded using thermocouples at different locations of the fin. In addition, a numerical model is developed that contains the dimensions of the fin set-up along with extended domain to capture the information of the fluid. Air is treated as a working fluid that enters the extended domain and absorbs heat from the heated fin. The temperature and the velocity of the fluid in the extended domain are obtained by solving the Navier–Stokes equation. The numerical model is now treated as a forward model that provides the temperature distribution of the fin for a given heat flux. An inverse problem is proposed to determine the heat flux that leads to the temperature distributions during experiments. The temperature distributions of the experiments and forward model are compared to identify the unknown heat flux. In order to reduce computational cost of the inverse problem the forward model is then replaced with artificial neural network (ANN) as data reduction, which is developed using several computational fluid dynamics solutions, and the inverse estimation is accomplished. The results indicate that a quick solution can be obtained using ANN with a limited number of experiments.