For developing model-based online applications such as condition monitoring and condition-based maintenance or real-time optimization, highly representative and yet simple physical models of centrifugal compressors are necessary. Previous investigations have shown that in this context meanline models represent a valid alternative to the commonly used empirical based modelling methodologies such as polynomial regression models or artificial neural networks.
This paper provides a methodology for tailoring meanline models to multistage centrifugal compressors by appropriate selection and adaptation of loss correlations. Guidelines for the selection of the boundary conditions are also provided.
The potential of the methodology is demonstrated in the Proof-of-concept section using two sets of data obtained from an air multistage centrifugal compressor operated in BASF SE, Ludwigshafen, Germany. The first set of data was used to calibrate the model whereas the second one was used for validation. The model results show that the predictions of stagnation temperature and pressure at the outlet of the stage deviate from the measurements respectively 0.15–3% and of 0.66–1.1% respectively. The results are discussed in the current paper.