Identification of the time-varying modal parameters of a spacecraft with flexible appendages using a recursive predictor-based subspace identification algorithm

Abstract

This study focuses on the recursive identification of the time-varying modal parameters of on-orbit spacecraft caused by structural configuration changes. For this purpose, an algorithm called recursive predictor-based subspace identification is applied as an alternative method to improve the computational efficiency and noise robustness, and to implement an online identification of system parameters. In the existing time-domain identification methods, the eigensystem realization algorithm and subspace identification methods are usually applied to obtain the on-orbit spacecraft modal parameters. However, these approaches are designed based on a time-invariant system and singular value decomposition, which require a significant amount of computational time. Thus, these methods are difficult to employ for online identification. According to the adaptive filter theory, the recursive predictor-based subspace identification algorithm can not only avoid the singular value decomposition computation but also provide unbiased estimates in a general noisy framework using the recursive least squares approach. Furthermore, in comparison with the classical projection approximation subspace tracking series recursive algorithm, the recursive predictor-based subspace identification method is more suitable for systems with strong noise disturbances. By establishing the dynamics model of a large rigid-flexible coupling spacecraft, three cases of on-orbit modal parameter variation with time are investigated, and the corresponding system frequencies are identified using the recursive predictor-based subspace identification, projection approximation subspace tracking, and singular value decomposition methods. The results demonstrate that the recursive predictor-based subspace identification algorithm can be used to effectively perform an online parameter identification, and the corresponding computational efficiency and noise robustness are better than those of the singular value decomposition and projection approximation subspace tracking series approaches, respectively. Finally, the applicability of this method is also verified through a numerical simulation.