MIMO LMS-ARMAX IDENTIFICATION OF VIBRATING STRUCTURES––PART II: A CRITICAL ASSESSMENT

In this part of the paper, a critical assessment of the MIMO (multiple-input multiple-output) LMS-ARMAX method is presented, along with comparisons with a pure ARX version and the Eigensystem Realisation Algorithm (ERA) based upon two-input three-output vibration data obtained from a scale aircraft skeleton structure. This structure is characterised by light (zeta<1%) damping and seven modes within the considered frequency range, two of which are closely spaced and another is a `localʹ tail mode. The study focuses on the: (i) ability of the methods to handle higher-dimensional problems, (ii) ability to estimate closely spaced and `localʹ modes. (iii) ability to accurately estimate light modal damping, (iv) required model overdetermination, (v) distinction of structural from `extraneousʹ modes, (vi) effects of various (white/colour) noise environments, and (vii) suitability of various discrete-time representations for effective identification. he LMS-ARMAX method is shown to be effective, achieving high accuracy in all considered cases. Its various features, such s the stochastic ARMAX representation, the guaranteed stability version, the modest computational complexity and the digital ispersion analysis, are of critical importance. Compared to the pure ARX version, the LMS-ARMAX method is shown to ead to more parsimonious representations and overall better model fits. The ERA is shown to lag behind in terms of overall erformance, leading to somewhat lower accuracy, in particular for the two closely spaced modes, and encountering difficulties n distinguishing structural modes via the modal amplitude coherence (MAC).