Geometric digital twin: A digital and intelligent model for aero-engine assembly accuracy prediction

Digital twin technology, as a representative paradigm of virtual modeling, has been increasingly applied to aero-engine assembly. However, existing approaches still fall short in predicting assembly accuracy, especially for internal parameters such as coaxiality that cannot be directly measured due to the engine’s enclosed structure. To address this challenge, this paper proposes a geometric digital twin (GDT) capable of high-fidelity modeling and accurate assembly prediction. The modeling framework consists of three levels: a part-level GDT based on geometric distribution error (GDE) reconstruction for incorporating machining-induced errors; a product-level GDT built through virtual assembly for integrating assembly-induced errors; and an intelligence-driven GDT developed using point cloud deep learning to reduce modeling complexity and cost. Together, these three levels enable high-precision and low-cost prediction of aero-engine assembly accuracy. The proposed approach was validated using simulated aero-engine components. The part-level GDT achieved modeling accuracies of 0.4 µm using high-precision point clouds and 8.3 µm using high-density point clouds. The product-level GDT yielded absolute differences between predicted and measured coaxiality not exceeding 0.056 mm. The intelligence-driven GDT achieved an R2 of 0.996, with a Mean Absolute Error (MAE) of 0.015 mm and a Root Mean Square Error (RMSE) of 0.025 mm. Moreover, the discrepancies between deep-learning-based coaxiality predictions and actual measurements remained within 0.070 mm. This paper provides a GDT for aero-engine assembly accuracy prediction, effectively enhancing both predictive performance and efficiency in aero-engine assembly.