A modular transfer learning-based method for predicting high-load support reactions in full-scale static tests

基于模块化迁移学习的全机静力试验高载支反力预估方法

¹ School of Aeronautics, Northwestern Polytechnical University, Xi'an 710072, China
² National Key Laboratory of Aircraft Configuration Design, Northwestern Polytechnical University, Xi'an 710072, China
³ National Key Laboratory of Strength and Structural Integrity, Aircraft Strength Research Institute, Xi'an 710068, China
⁴ Key Laboratory of Aviation Science and Technology on Full Scale Aircraft Structure Static and Fatigue Strength, Aircraft Strength Research Institute, Xi'an 710068, China

Received: 1 July 2025

Abstract

To address the problem of neglecting load uncertainties and structural deformation in conventional support reaction calculations for full-scale aircraft static tests, this paper proposes a novel approach based on modular transfer learning for accurately predicting high-load support reactions. The proposed method introduces a modular transfer learning framework in which separate models are developed to account for loading uncertainties. A primary prediction model is first established using historical test data to extract general static testing characteristics. Subsequently, a residual correction model trained with low-load data from the current test condition is integrated, enabling adaptation and correction of prediction errors specific to the present testing scenario. Additionally, to explicitly consider structural deformation effects, a binomial regression technique is employed to estimate loading-point coordinates at high-load levels, facilitating accurate determination of loading directions. By combining these coordinates with spatial force equilibrium equations, precise prediction of support reactions is achieved. The effectiveness and accuracy of the proposed method are validated through a full-scale static test case involving a representative verification aircraft. Comparative analyses demonstrate that, compared with traditional rigid-body assumptions, the developed method significantly enhances the prediction accuracy for high-load stages, thereby providing robust theoretical and technical support for support reaction estimation in large-scale and complex testing environments.

摘要

针对全机静力试验支反力计算忽略加载载荷不确定性和飞机结构变形的问题, 提出一种基于模块化迁移学习的全机静力试验高载支反力预估方法。该方法引入模块化迁移学习框架, 分模块构建加载载荷不确定性预测模型: 利用历史试验数据构建主预测模型, 提取静力试验的通用规律; 结合现场试验低载数据训练残差修正模型, 实现对当前试验工况的适应与误差校正。同时, 为考虑结构变形的影响, 采用二项式回归方法预估高加载级数下的加载点坐标, 进而计算加载力线方向, 并结合空间力系平衡方程实现精确的支反力预估。此外, 运用某型验证机的全机静力试验案例验证所提方法的预估准确性。与传统刚体假设下的支反力计算结果相比, 文中方法显著提升了高载阶段支反力预估的精度, 为大规模复杂试验系统中的支反力计算提供了理论和技术支撑。