Uncertainty calibration and quantification of surrogate model for estimating the machining distortion of thin-walled parts

Thin-walled structural parts, such as aeroengine casings, impellers, blades, and disks, are widely used in the aerospace industry due to their outstanding performance. A more efficient and accurate prediction approach for machining distortion has also been widely considered. The traditional finite element analysis approach, which is commonly used in the industry, has some limitations in both efficiency and accuracy. To resolve the above problems, a hybrid surrogate model considering uncertainties is proposed to estimate the machining distortion of thin-walled parts. In terms of efficiency improvement, a hybrid three-layer gradient surrogate model to integrate the response surface regression model and the Gaussian process regression model is introduced to replace the multi-physical finite element analysis model. In terms of accuracy improvement, the calibration and quantification of uncertainties within the model are carried out. The uncertainties are divided into four parts, including cutting contact area uncertainty, tool wear uncertainty, mechanism modelling uncertainty, and unquantified uncertainty, which are determined through an iterative solving algorithm, an experimental calibration approach, and Bayesian inference simultaneously. An aeroengine combustor casing is selected as a case study to validate the effectiveness of the proposed estimation model. The predicted results indicate that the estimation time is less than \(24 \mathrm{ms}\), and the average estimation deviation is \(3.1400 \mu \mathrm{m}\). Compared with the traditional finite element analysis model, the proposed methodology can significantly reduce the estimation time with higher accuracy.