A hybrid driven approach to integrate surrogate model and Bayesian framework for the prediction of machining errors of thin-walled parts

Abstract Thin-walled structural parts are widely used in aerospace industry, the prediction of machining errors of these parts has attracted more and more attention over the past decades. Due to the complicated time-varying nonlinear characteristics and the lack of experimental data, it is difficult to use the mechanism model or data-driven model to predict the machining errors directly. To solve those problems, a hybrid driven approach to integrate surrogate model and Bayesian framework is proposed to predict the machining errors of thin-walled parts. The Gaussian process regression algorithm is embedded in the mechanism model to predict the cutting forces and flexibility, so as to establish the surrogate model for the prediction of machining errors. Besides, the surrogate model is calibrated by considering the uncertainties of cutting forces and tool wear. Five unknown calibration coefficients are analyzed through the Bayesian framework and determined by the Markov Chain Monte Carlo algorithm. A hybrid driven approach is introduced to train the prediction model. Both simulation data through mechanism analysis and experimental data through technological test are extracted. Compared with the mechanism model or data-driven model, the testing results have indicated that the proposed prediction model has higher efficiency and accuracy. The prediction time of a single point is reduced to 5.46ms and the root mean square error of the proposed prediction model is 4.5137μm.