compressor performance mapping, calorimeter testing, artificial neuron network
In the last decades, several research and development efforts led to new compressor technologies that have been successfully introduced into market such as hermetic compressors with variable-speed motors, compressors with economization lines (both vapor and liquid injection lines), hermetic linear compressors, novel capacity modulations, and oil-flooding among others. During the process of implementing new compressor technologies, performance mapping is essential to predict the system behavior across different operating conditions. However, current standard AHRI-540 for compressor performance rating utilizes a 10-coefficient polynomial model that has severe limitations to include compressor enhancement technologies and variable operating range. In addition, it is common practice in industry to calibrate such polynomial correlations with at least 15 to 20 compressor calorimeter test points for a single compressor to ensure a good fit, which results in extended laboratory testing time and relatively high associated costs. Therefore, an automated compressor performance mapping approach based on artificial neural network (ANN) modeling is proposed to address and overcome the limitations of the current standard including applicability to any positive displacement compressors and minimization of number of test points required to accurately predict the compressor envelope. In this paper, the performance of a positive displacement compressor is mapped by this novel methodology, which relies on an algorithm that effectively determines the minimum set of data points required and optimizes the training/testing of ANN architecture. The accuracy and reliability of the proposed methodology are compared to the conventional 10-coefficient polynomial mapping. Lastly, the propagated uncertainties through the model and its extrapolation capabilities are also analyzed.