Data driven assessment of a small scale evaporative condenser based on a combined artificial neural network with design of experiment approach

The performance of evaporative condensers depends on operating parameters such as the state of ambient air and circulating water (environmental) and the condition of the refrigerant fluid (operational). The equipment behavior can be analyzed in a laboratory environment with the aid of Design of Experiment DoE tools, which effectively assists in the identification of trends and couplings, but the procedure depends on data collected in a controlled manner. The aim of this paper is to analyze the behavior of a small-scale evaporative condenser tested in the laboratory environment with the aid of DoE, based on an uncontrolled experimental dataset. A data driven approach is applied to the problem by creating an neural network algorithm capable of reproducing the equipment behavior as if it were obtained from controlled factors, with coefficients of determination of 0.973 and 0.988 for the heat reject q ˙ c o n d and the overall heat transfer coefficient Ucond. General functions for these outputs are obtained out from a factorial 2k DoE approach, allowing to identify the environmental air wet bulb temperature Twb, in as the most relevant parameter for the q ¯ c o n d prediction and m ˙ s w and (Twb, in, m ˙ a i r ) as the most relevant ones concerning Ucond prediction. The errors from these prediction functions are calculated to be 3.48% and 3.69% respectively, with coefficient of determination of 0.793 and 0.752. The proposed data driven metamodels showed to be useful tools to represent and simulate complex systems in a much easier way, concerning both their mathematical implementation and computational running time.