Predicting Automotive Air Conditioning System performance through Deep learning

Abstract

Automotive Air Conditioning System (AACS) involves phase change of the refrigerant, to provide a comfortable environment in the vehicle cabin. The phase change is governed by many complex equations. Therefore, a technique that can validate the results and predict the system performance is required to avoid tedious calculations. Deep Neural Networks (DNN) are better at learning complex non-linear relationships between performance metrics. Experimental data is used to train the specified DNN model. Compressor speed, air temperature at the inlet of the evaporator, and refrigerant flow rate are used as input, while coefficient of performance, compressor work, and heat loss have been used as output parameters to train the model. Predicted results are compared by using statistical measures such as Root Mean Square Error, Mean Square Error as well as Correlation Coefficient. Based on the results obtained, the specified DNN model can be effectively used in predicting and validating the performance of the AACS.