Regression model of rotor shape errors based on the ISSA-BP neural network

Ultraprecision aerostatic spindles are the key functional component of ultraprecision machining tools and precision measuring instruments. The uniformity and consistency of air film thickness are influenced by the shape errors of the ultraprecision aerostatic spindle. However, given that mass-produced rotors have different local error characteristics, existing shape error models fail to express the variation of the air film thickness accurately when the rotor rotates in the flow field. Thus, the influence of the distribution characteristics of rotor shape errors on the aerostatic bearing flow field and rotor performance cannot be accurately evaluated. In this study, a multistrategy fusion and improved sparrow search algorithm (ISSA) was proposed to optimize the backpropagation (BP) neural network, and a regression model of rotor shape error was established on the basis of the ISSA-BP algorithm. Numerous rotor surface profiles were collected by a cylindricity instrument, and the Gaussian-RK filter was applied to remove the noise and burrs from the rotor profiles. The distribution regularity of local error characteristics was statistically analyzed, and the ISSA-BP neural network was used to fit the rotor surface profiles. The shape error regression model of the rotor was established by regression analysis. Error and harmonic analyses were performed on the model and experimental measurement data. Experimental results showed that the relative error of the rotor shape error regression model in harmonic analysis was less than 1 %, and the maximum relative error of the rotor profile was lower than 0.1 %. Thus, the accuracy of the constructed rotor shape error regression model is verified.