Perpustakaan Digital ITB

Improvements in centrifugal compressor performance have a major influence on total system efficiency and operating costs. One of the most significant components of centrifugal compressors is the impeller . There are two important parameters to improve compressor performance yield ,Pressure Ratio and Efficiency. In this thesis to improve the performance of centrifugal compressor using multi-objective optimisation method based on kriging-surrogate model assisted by Expexted HyperVolume Improvement. To evaluate the parameters of both parameters, computational fluid dynamics is used. After that, the model interpretation process is carried out by explaining the model against the predicted results obtained using the SHAP method. Optimization was performed on the impeller geometry by changing the control points as Design variables. There are 10 Design variables used in this thesis and Pressure ratio and Efficiency as two objective functions. Optimization was successfully carried out and produced a set of optimum solutions known as Pareto front. The optimization resulted in 4 main designs, namely Baseline, Balanced, PRMAX, EFFMAX. The highest pressure ratio occurs in the PRMAX design with an increase of 9.61% compared to the baseline. while the highest Efficiency is in the EFFMAX design with an increase of 6.21% compared to the baseline. The most optimum design is Balanced. The interpretability of the Machine Learning model using SHAP provides good insight for the case of multi-objective optimization for impeller centrifugal compressor. The results provide insight that the most influential input variable are Design variable TS1 and Hh for objective Efficiency and Pressure Ratio, respectively.