The Data-Driven Surrogate Model-Based Dynamic Design of Aeroengine Fan Systems

Abstract

High cycle fatigue failures of fan blade systems due to vibrational loads are of great concern in the design of aero engines, where energy dissipation by the relative frictional motion in the dovetail joints provides the main damping to mitigate the vibrations. The performance of such a frictional damping can be enhanced by suitable coatings. However, the analysis and design of coated joint roots of gas turbine fan blades are computationally expensive due to strong contact friction nonlinearities and also complex physics involved in the dovetail. In this study, a data driven surrogate model, known as the Nonlinear in Parameter AutoRegressive with eXegenous input (NP-ARX) model, is introduced to circumvent the difficulties in the analysis and design of fan systems. The NP-ARX model is a linear input-output model, where the model coefficients are nonlinear functions of the design parameters of interest, such that the Frequency Response Function (FRF) can be directly obtained and used in the system analysis and design. A simplified fan bladed disc system is considered as the test case. The results show that by using the data driven surrogate model, an efficient and accurate design of aero-engine fan systems can be achieved. The approach is expected to be extended to solve the analysis and design problems of many other complex systems.

A Machine Learning-Based Surrogate Finite Element Model for Estimating Dynamic Response of Mechanical Systems

“…Hall et al [25] presented a new technique for developing both linear and nonlinear statistical forecasts of the winter North Atlantic Oscillation (NAO) based on complex systems modeling, that their polynomial NARMAX models demonstrated considerable skill in out-of-sample forecasts and their performance was superior to that of linear models. Zhu et al [26] introduced a data-driven surrogate model, known as the Nonlinear in Parameter AutoRegressive with exogenous input (NP-ARX) model, to circumvent the difficulties in the analysis and design of fan systems. Ren et al [27] proposed two active learning approaches to combine Kriging and ANN models for reliability analysis the efficiency and accuracy of the proposed approaches were demonstrated by representative examples and two finite element problems.…”