To improve the performance of electrically assisted turbochargers (EATs), the influences of the hub profile and the casing profile on EAT performance were numerically studied by controlling the upper and lower endwall profiles. An artificial neural network and a genetic algorithm were used to optimize the endwall profile, considering the total pressure ratio and the isentropic efficiency at the peak efficiency point. Different performances of the prototype EAT and the optimized EAT under variable clearance sizes were discussed. The endwall profile affects an EAT by making the main flow structure in the endwall area decelerate and then accelerate due to the expansion and contraction of the meridional surface, which weakens the secondary leakage flow of the prototype EAT and changes the momentum ratio of the clearance leakage flow and the separation flow in the suction surface corner area. Because the tip region flow has a more significant influence on EAT performance, the optimal casing scheme has a better effect than the hub scheme. The optimization design can increase the isentropic efficiency of the maximum efficiency point by 1.5%, the total pressure ratio by 0.67%, the mass flow rate by 1.2%, and the general margin by 6.4%.

To improve the performance of an electric assistant turbocharger in different backpressure environments, the compressor margin needed to be optimized individually. As an effective passive control method to improve the margin, the full-chord-covered pressure side tip winglet has been studied and applied for compressor and turbine rotors. The research on the effect of partial tip winglets, which can reduce added mass of winglet and the weight of the rotor, on the compressor margin has not been found in the available literature. Utilizing numerical simulation, this paper was focused on the research of a high-speed single-rotor compressor which the tip loading coefficient is 0.81. The shape of the pressure side tip winglet is optimized with the margin as the single objective. A database with 189 valid samples was generated using the Latin hypercube method. A self-developed ε-Support vector regression (SVR)-genetic algorithm (GA) optimization platform was applied to optimize the tip shape. The platform obtained a partial pressure side tip winglet shape. Compared with the CFD simulation results, the error of the ε-SVR-GA platform predicted the margin was 4.1%. The result obtained by the ε-SVR-GA platform was improved compared with the prototype margin by 25.36%.