Machine learning approach for multi-parameter performance estimations of EC fan coil units using heating tentative database

Abstract

Fan coils (FCs) are widely utilized, yet little is known about their performance under different operating conditions. A segment of a comprehensive experimental dataset with 1727 data points is used to develop and train four artificial neural network (ANN) architectures to computationally estimate the heat output and available power of a ceiling-mounted FC. The tests have been done via a specifically devised AMCA 210 test apparatus, under interior air and heat exchanger (HEX) fluid conditions recommended by EUROVENT. Utilizing six given inputs, the 1st ANN estimated the airflow rate and fan power. With five dissimilar input parameters, the exit temperature of air as well as the heating capacity was forecasted. Considering five separate inputs, the 3rd ANN assessed the pressure drops at the water side pertaining to the HEX. Depending on eight diverse inputs, the air exit temperature and power of the fan alongside total heating capacity were estimated. In the network models of 10 neurons in the hidden layer, the Levenberg-Marquardt training method has been utilized. Considering the 1st ANN, the deviation that pertained to the air flow rate was found to be -0.59%, as the deviations relevant to the air outlet temperature and heating capacity in the 2nd ANN were detected to be 0.001% and 0.03%, respectively. Additionally, the 3rd ANN resulted in a deviant value of -0.07%, referring to the fluid pressure loss. The 4th ANN has also brought about deviations of -0.005%, -0.13%, and + 0.09%, referring to exit air temperature, heating capacity, and fan power, respectively.