基于多任务学习的飞机起落架着陆载荷预测模型

Prediction Model of landing load on aircraft landing gear based on multi-task learning

摘要:

构建飞行参数与起落架着陆载荷之间的预测模型，对于起落架结构的应力分析、安全寿命评估和预测性维修具有重要意义。以多任务学习为框架，提出一种基于多门控混合专家网络（MMoE）的起落架着陆载荷预测模型；基于飞行参数数据集的皮尔逊相关系数分析，确定 MMoE 模型的输入特征，实现对起落架着陆载荷的精准预测，并与单任务模型和 Shared-Bottom 模型的性能进行对比。结果表明：基于 MMoE 模型的框架能显著提升载荷数据的预测能力，相较于其他两种模型，基于 MMoE 的起落架着陆载荷预测模型精度更高，在测试集上的均方误差（MSE）降低了 66% 以上。

Abstract:

Building a predictive model between flight parameters and landing load on landing gear is of great significance for the stress analysis and safety life assessment of landing gear structures, as well as predictive maintenance. This paper presents a landing gear landing load prediction model based on multi-task learning framework, using a Multi-gate Mixture of Experts (MMoE). The input features of the MMoE model were determined by Pearson correlation analysis on the flight parameter dataset, achieving accurate prediction of the landing gear landing load and comparing the performance with Single-Task model and Shared-Bottom model. The results show that the framework based on the MMoE model significantly improves the prediction ability of load data. Compared with the other two models, the MMoE model exhibits greater robustness, and the mean square error (MSE) on the test set is reduced by more than 66%.