Use of artificial neural network in rotorcraft cooling system

In this study, an Artificial Neural Network (ANN) is used to determine the surface temperatures of the avionics equipment located in an avionics bay of a rotorcraft. The bay is cooled via a system of a fan that supplies ambient air to the interior of the bay and an exhaust. A Feedforward Multi-Layer ANN is used with the input parameters of the fan and exhaust locations and the air mass flow rate of the fan. For training of the network, the results obtained by a large number of Computational Fluid Dynamics (CFD) analyses are used. An analysis on the accuracy of the ANN algorithm through the use of different ANN architectures revealed that an ANN with fifteen neurons in the hidden layer provides the best accuracy among the considered options. The size of the training data is increased progressively and its effect on the prediction accuracy of the ANN algorithm is also observed. The regression capability of the ANN is later compared with a response surface built by a commonly used full quadratic linear model. The comparison shows that the ANN predicts the avionics surface temperatures with much better accuracy