Performance prediction and Bayesian optimization of screw compressors using Gaussian Process Regression

Injecting oil inside the compressor chambers of the oil-flooded, twin-screw compressors has several advantages. Oil cools the compressing fluid upon mixing with it and hence the compression process is brought nearer to the ideal isothermal compression process. The oil also serves as a lubricant between the meshing rotors and other clearance gaps in the compressor. The thin film of oil formed in the clearance gaps prevents internal leakages too; enhancing the volumetric efficiency of the compressor. Among these desirable effects of injecting oil in screw compressors, there is an undesirable effect too. The interaction (friction) of oil films formed in various clearance gaps with the rotors leads to a drag power loss. Recent studies such as Abdan et al. have proposed more detailed and accurate methods to estimate the oil drag losses in screw compressors. These methods enable the modelling of the effect of even minor changes in rotor profile on the drag loss power. Predictions of this model were hence used to tweak the screw rotor profiles with an objective to reduce the oil drag losses. Such profiles were then retrofitted in the existing machines and tested. Comparing the differences in power consumption of these machines, the component of oil drag loss was deduced. The experimental results show close agreement with the oil drag loss prediction model. The reduction of oil drag losses through profile modifications led to an improvement in the specific power of oil-flooded screw compressor.