Towards Automatic and Systematic Compressor Model Selection Based on Measurement Data

Abstract: Centrifugal compressors are mechanical devices that move and compress gas and are used in a wide range of applications. To get as much efficiency out of the compressors as possible, high demandsare put on having good compressor models. One method of compressor modeling is the curve fitting method, where empirical and mathematical models are used to parameterize relationships between different compressor characteristics. The models come in a vast variety of formats, complexity and difficulty of applying them. With the variety of compressors and models to select from, it is important to select an appropriate model. A tool for automatic parameterization, evaluation and selection of compressor model would be very useful. A model library is proposed in the thesis that utilizes the Total Least Squares method for parameterizing and evaluating the models. By initializing the model library with measurement data, called a compressor map, the compressor is broken down into smaller components in a hierarchy, based on physical expressions and characteristics. The individual compressor models are defined and parameterized on top of the base model, overriding and replacing the base model feature. Total Least Squares offers a countermeasure to avoid large model prediction errors associated with the nonlinear behavior of the compressor. To utilize the model library further, a simulation interface is proposed, so the selected compressor model can easily be applied in an engine and vehicle simulation environment. The model library is evaluated using three different compressor maps and shows promising results, where the TLS method seem to better capture bad visual curve-fitted models, compared to other simpler methods. Parameterization of the models are done automatically without any manual user input needed. The simulation interface is tested by replacing the compressor block of an engine simulation model, where the results show that the model component hierarchy can be integrated in a simulation environment.