Data driven modelling approach for design assessment of spacecraft equipment

• Propose a new regularization method enhancing data-driven model stability. • Demanding loads are integrated into the proposed regularization approach. • The data-driven modelling has the applicability for spacecraft assessment. Ground vibration tests are conducted on spacecraft equipment prior to launch to simulate the demanding vibration loads, characterized by high amplitudes and energies, that occur during rocket launching. These vibration loads may be destructive to spacecraft equipment. Therefore, spacecraft equipment must be assessed under these demanding loads before undergoing destructive ground vibration tests. Existing assessment methods, whether based on finite element analysis or other physical techniques, struggle to quantitatively analyze the nonlinear dynamic behavior of spacecraft equipment, such as the frequency-drift phenomenon, due to their complex structures. The objective of this study was to develop a data driven model to implement the design assessment of spacecraft equipment based on training data obtained from non-destructive tests. However, an identified data driven model is prone to overfitting due to noise effects in system identification, which may destabilize the model when subjected to demanding loads with larger amplitudes. To address this issue, a novel regularization approach based data driven modelling approach for the design assessment of spacecraft equipment was proposed. In this approach, the system identification algorithm is regularized by model predicted output errors of the training inputs and the demanding loads for design assessment. Then, an iterative computational approach was used to automatically discard unstable data driven models, following which the efficiency of the proposed data driven modelling approach was verified by comparison with other traditional methods for an engineering case of large spacecraft equipment. The results demonstrated that the proposed method could address the instability problem in extrapolation predictions and demonstrated superiority over traditional methods.