Integrated Computational Framework for Controlling Dimensional Accuracy of Thin-Walled Turbine Blades During Investment Casting

Daiyin Zhao,Leyao Zhou,Donghong Wang,Hong Zeng,Xiufang Gong,Da Shu,Baode Sun
DOI: https://doi.org/10.1007/s00170-023-12319-8
IF: 3.563
2023-01-01
The International Journal of Advanced Manufacturing Technology
Abstract:Nickel-based superalloy MM247 turbine blades produced through investment casting are widely used for aero engines and gas turbines. To control the dimensional accuracy of the blades, this study proposed a novel integrated computational framework named AICAST, implanted with the response surface methodology optimization model and multi-layer perceptron neural network. Firstly, the wax injection stage and the influence of the gating system and process parameters on the displacement field during the solidification process are simulated to predict the deformation distribution of the blade patterns. The maximum deformation of the wax pattern is then lowered by 60.39%, and the deformation of the blade casting reaches the minimum via AICAST of 0.1504 mm. The maximum prediction difference of the MLP neural network model is 0.0123 mm, and the mean absolute percentage error is 2.71%, indicating that the deformation of blade casting can be modeled and predicted. Finally, the experimental results show that the wax pattern and the blade casting meet the dimensional accuracy requirements. The largest fluctuation of the contour error is in the A3 section, which agrees with the simulation results. The proposed framework, AICAST, can give researchers new insight into integrated computational materials engineering. The framework takes the fluctuations and effects of multiple processes into account. It can also automatically operate all the script files, thereby improving the efficiency of traditional numerical simulation.
What problem does this paper attempt to address?