Neural network, reciprocating compressor, temperature field.
The understanding of heat transfer interactions in refrigeration compressors is of fundamental importance to characterize their overall performance. Certain temperatures, such as those of the motor, oil, shell, and at suction and discharge chambers, have strong influence on the compressor electrical consumption and reliability. Experimental and numerical approaches have been successfully employed to characterize the thermal profile of compressors under different operating conditions. This paper presents a multi-layered feed-forward neural network developed to predict the main temperatures of a hermetic reciprocating compressor. Such a model can be used for different compressor layouts without major modifications, being a fast method for estimating temperatures without the solution of the compression cycle. Predictions of the neural network were compared with experimental data and numerical results from comprehensive thermodynamic simulations, and good agreement was observed in a wide range of evaporating and condensing temperatures. The neural network was found to predict the temperature distribution with sufficient accuracy for compressor analysis and development.