Performance Analysis on Modeling of Loop Heat Pipes Using Artificial Neural Networks

The absence of mechanical moving parts, simplicity and robustness of loop heat pipes (LHPs) has prompted many satellite manufacturers to employ LHPs as primary thermal control systems. The acknowledgement of temperature dynamics of LHP plays a vital role in conception, optimization and control of the same. The overall thermal conductance varies not only with the power input, sink temperature and ambient conditions, but also with the previous history of its operation. In this connection, the present work is focused on a different approach for modeling LHPs. Experimental data for a vertical loop heat pipe made of copper, with two different working fluids, viz., water and ethanol for a range of heat inputs and fill ratios is collected from the literature. An artificial neural network (ANN) is trained with the collected test data and validated. Fully connected feed forward multilayer configuration (MLFFN) with momentum back propagation algorithm is adopted for the ANN. The MLFFN architecture consists of two input nodes representing the parameters heat input and fill ratio, and a single output node representing the thermal resistance of the LHP. The MLFFN predictions were validated within the domain of total available experimental data. This study also emphasizes that the understanding of the physical phenomena of LHP to be modeled by ANN is a prerequisite for getting acceptable results. As there is a serious limitation of conventional techniques for understanding the LHP physical phenomena and thermal behaviour, ANN approach appears to be very promising, if sufficient experimental data is available covering the entire range of system operation.