摘要
为精准构建起落架载荷谱标定模型,提出了全域整体传感测试方法,结合机器学习算法搭建智能地面标定模型。首先,分析全域受力特性,选择起落架多个关键截面,实现多向载荷协同感知,建立覆盖整体的“应变-载荷”映射网络。其次,在建模环节,引入主成分分析(principal components analysis,PCA)提取独立特征变量,完成高维应变参数的降维与解耦,同时采用贝叶斯优化算法对轻量级梯度提升机(light gradient boosting machine,LightGBM)进行超参数寻优,最终建立高精度非线性载荷标定预测模型。试验结果表明:该方法平均相对误差为1.72%,均方根误差为49.53,性能优于多元线性回归(multiple linear regression,MLR)与PCA-BP神经网络等传统方法。训练时间较采用随机搜索的极限梯度提升(XGBoost)模型和未降维LightGBM分别缩短78.5%和39.5%。
A holistic full-field sensing test method was proposed to accurately construct a calibration model for the landing gear load spectrum,integrating machine learning techniques to build an intelligent ground calibration model.First,the full-field force characteristics were analyzed,and multiple critical sections of the landing gear were selected to achieve collaborative sensing of multi-directional loads,thereby establishing a covering full-field“strain-load”mapping network.Second,in the modeling phase,principal components analysis(PCA)was introduced to extract independent feature variables,accomplishing the dimensionality reduction and decoupling of high-dimensional strain parameters.Simultaneously,a Bayesian optimization algorithm was utilized for hyperparameter optimization of the light gradient boosting machine(LightGBM),ultimately establishing a high-precision nonlinear load calibration and prediction model.Experimental results indicate that the proposed method achieves an average relative error of 1.72%and a root mean square error(RMSE)of 49.53,outperforming traditional methods such as multiple linear regression(MLR)and PCA-BP neural networks.The training time is reduced by 78.5%compared to an XGBoost model using random search and by 39.5%compared to a LightGBM model without dimensionality reduction.
作者
刘力宏
刘克格
刘彦鹏
韩小进
LIU Lihong;LIU Kege;LIU Yanpeng;HAN Xiaojin(China National Machinery Industry Corporation Beijing Aircraft Strength Research Institute Co.,Ltd.)
出处
《仪表技术与传感器》
北大核心
2026年第2期115-120,共6页
Instrument Technique and Sensor
基金
国家自然科学基金项目(50135010)。
