Bayesian Neural Network-Based Uncertainty Quantification of Impact of Geometric Deviations on Ultra-High-Lift Low-Pressure Turbine Blades

The aerodynamic performance of low-pressure turbine blades is significantly influenced by geometric deviations, which is further intensified by the ultra-high-lift (UHL) design approach aimed at weight reduction. To quantify this impact, an enhanced surrogate model, the Bayesian neural network (BNN), is employed to predict the effects of these deviations. In addition to estimating prediction uncertainty, the BNN demonstrates strong resistance to overfitting, exhibiting superior robustness compared to conventional neural networks, particularly when trained on limited datasets. Notably, the BNNbased quantification method can still provide meaningful results when the training data is directly reduced to 1/10 of its original size, representing the potential true solution as a probability distribution and thus offering high interpretability. This approach effectively reduces data requirements and is well-suited to high-reliability uncertainty quantification. Quantitative results reveal that geometric deviations introduce considerable random scatter in aerodynamic performance, increasing stall susceptibility due to large-scale open separation. Consequently, this significant uncertainty impact underscores the necessity for robust design strategies, especially for UHL configurations.