An interpretable quality prediction method with adaptive hidden representation graph neural networks for compressor blade manufacturing

As the core power component of an aircraft engine, compressor blade quality critically affects the performance and stability of the aircraft propulsion system. As a quintessential process in precision manufacturing, compressor blade fabrication involves producing complex curved-surface components with micron-level tolerances, where error propagation exhibits strong cross-parameter correlations throughout the manufacturing chain. Therefore, there is a critical need for a quality prediction method that characterizes the interaction between features and processing quality. However, most technologies neither satisfy the tolerance requirements for compressor blade manufacturing nor provide sufficient interpretability to support processing parameter optimization. In this article, we propose an interpretable quality prediction method for compressor blade manufacturing using adaptive hidden representation graph neural networks (AHRGNN). An adaptive hidden representation graph neural networks can explore the interpretable mapping between processing parameters and the final processing quality, and the final quality prediction accuracy meets the manufacturing requirements of compressor blades. To evaluate the efficacy of the proposed method, two compressor blade production lines at Wuxi Turbine Blade Co., Ltd., China were selected for experimental validation. The results achieved the smallest root mean square error at 0.032mm and 0.020mm, respectively.