Share

Predicting vapor injected compressor performance using artificial neural networks.

Number: 1421

Author(s) : KHAN A., BRADSHAW C. R., SCHMITT J., LANGEBACH R.

Summary

Positive displacement compressors have recently begun to include vapor injection more frequently to adapt to energy efficiency and decarbonization goals. High-accuracy models are crucial to predict the compressor performance for rapid integration into HVAC&R systems. Most existing empirical models use more than 10 experimental data points for accurate performance prediction, which can prove burdensome. This study aims to address the need for more universal and versatile compressor mapping methodologies that do not require such intensive and expensive experimental testing. An artificial neural network (ANN) based vapor-injected compressor performance mapping approach is proposed. The proposed ANN model architecture comprises of one input layer, one output layer, and one hidden layer. Input layer includes input parameters such as compressor speed, and suction, injection, and discharge pressures while output layer includes output parameters such as evaporator mass flow rate, injection mass flow rate, compressor power, and discharge temperature. In addition, this study qualifies the feasibility and reliability of the proposed ANN model using Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). Data is collected on vapor injected scroll and rotary compressors with R410A and R454B to train and test the model. The model can predict the evaporator mass flow rate, injection mass flow rate, and compressor input power within 5% MAPE, and discharge temperature with 5K MAE.

Available documents

Format PDF

Pages: 10

Available

Free

Download

Details

  • Original title: Predicting vapor injected compressor performance using artificial neural networks.
  • Record ID : 30033616
  • Languages: English
  • Subject: Technology
  • Source: 2024 Purdue Conferences. 27th International Compressor Engineering Conference at Purdue.
  • Publication date: 2024/07/18

Links


See other articles from the proceedings (80)
See the conference proceedings

Indexing