Experimental Investigation and Neural network based parametric prediction in a multistage reciprocating humidifier

• Published in: Applied energy Vol. 293; p. 116958
• Main Authors: Salins, Sampath Suranjan, Kota Reddy, S.V., Shiva Kumar
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2021
• Subjects: Artificial Neural network; Cooling effect; Energy consumption; Mean relative error; Multistage; Reciprocating cooling pads; Energy consumption; Multistage; Cooling effect; Mean relative error; Artificial Neural network; Reciprocating cooling pads
• ISSN: 0306-2619; 1872-9118
• DOI: 10.1016/j.apenergy.2021.116958

•A novel reciprocating multistage evaporative cooling method has been proposed to cool huge rooms.•Better performance in terms of COP and SCC observed for the proposed design as compared with single stage.•ANN techniques has been adopted to predict the responses.•Trainbr outperformed other algorithms with lower values of MRE for all responses. Cooling of the buildings is very much mandatory in summer and to meet this, considerable energy will be spent for cooling purpose across the world. Present work focuses on the multistage evaporative cooling pads where four different packing are used to analyze the different humidification output parameters. Cam shaft which is powered by the motor gives reciprocating motion to the cooling pads which is made to dip inside the stagnant water. Input operating parameters such as air velocity, cam shaft speed and the number of cooling pads are varied and the output parameters like pressure drop, cooling effect, coefficient of performance, relative humidity drop and energy consumption rate are determined. Results indicated that, there is an increase in COP, pressure drop and the energy consumption rate with the rise in the air velocity. Artificial neural network has been used for predicting the performance parameters of the experimental results. 3-15-4 structured MLP based network is considered and is trained by using trainscg, trainlm and using trainbr networks. Results indicated a good prediction capability of ANN techniques with MRE of test data lying below 12%. Trainbr outperformed the other two networks as the correlation coefficient was much higher and MRE was lower for both training as well as test data.