Research on application of machine learning to reverse identification of aerospace structure loads

DOI:	10.11809/bqzbgcxb2023.04.014
Keywords:	aerospace structure; load identification; machine learning; high fidelity modeling and simulation; high information entropy
Abstract:	Considering that the existing load identification methods are difficult to be applied to complex structures and various coupled loads on spacecraft, this paper proposes a load identification method based on machine learning. The huge training data required for machine learning are provided through high fidelity modeling and simulation technology, and the layout of sensor measurement points is optimized based on high information entropy to make the system simple and efficient. Various machine learning models are considered for different types of structural loads and the optimal model is selected for load identification. The method is verified by identifying cross section loads with coupling effects on typical spacecraft cabin structures, and the relative errors of the identification do not exceed 3%, which is obviously better than the traditional calibration method of load response transfer coefficient matrix. The result indicates that the machine learning method can better identify multiple coupling loads on the complex structures, the identification accuracy of which can meet the need of engineering application.
Issue:	Vol. 44 No. 4 (2023)
Published:	2023-04-28
PDF HTML