A probabilistic approach for aero-engine component performance margins design

Abstract

Uncertainties in the design and manufacturing processes may lead the overall performance of individual engines deviate from their original design, posing a risk of failing to meet intended performance requirements. Traditional empirical methods often fail to provide appropriate component performance margins and tend to assign overly conservative margins, which exacerbates the difficulty of achieving overall performance matching. To address the problem of component performance margin design under specific aero-engine performance success rate requirements, a novel approach is proposed based on neural network and global optimization algorithm, within the framework of the Monte Carlo simulation (MCS). The approach utilizes benchmark commercial software to generate a large amount of structured data to train a multi-layer perceptron (MLP) model, thereby obtaining a surrogate model for the impact of component performance variation on overall performance under engine design condition, and then combines the MLP model with a differential evolution algorithm (DE) to carry out the component performance margin design. The results show that the coefficient of determination (R²) and the p-value for the constructed MLP model are both 1.00, with a maximum relative error of only 0.052% compared to the benchmark model outputs. The proposed margin design approach increases the engine performance success rate from the original 44.31% to the target of 99.00%, with a computation time of approximately only 10 seconds. The extended case study considering both design and off-design conditions further demonstrates the robustness of the proposed method.