Assessing the Quality of Experimental Data with Gaussian Processes: Example with an Injection Scroll Compressor
injection compressor, scroll, machine learning, Gaussian Processes, experimental
This paper describes an experimental study carried out on a refrigeration scroll compressor with and without vapour injection. The test rig designed for that purposed allows evaluating the performance over a wide range of operating conditions, by varying the supply pressure, the injection pressure, the exhaust pressure, the supply superheating and the injection superheating. 97 Steady-state points are measured, with a maximum isentropic efficiency of 64.1% and a maximum consumed electrical power of 13.1 kW. A critical analysis of the experimental results is then carried out to evaluate the quality of the data using a machine learning method. This method based on Gaussian Processes regression, is used to build a statistical operating map of the compressor as a function of the different inputs. This statistical operating map can then be compared to the experimental data points to evaluate their accuracy.