Modal identification of structures from input/output data using the expectation–maximization algorithm and uncertainty quantification by mean of the bootstrap

Abstract

Modal testing in civil engineering includes the possibility to apply measured forces in addition to the unmeasured ambient excitation. In these cases, it is necessary to consider mathematical models that account for both excitation sources, what explains the increasing interest in sophisticated system identification methods for modal analysis with input/output data. In this work, the maximum likelihood estimation of the state space model from input/output vibration data is investigated. This model can be estimated using different techniques: Among them, the maximum likelihood method has optimal statistical properties, so modal parameters computed using this approach will be optimum in a statistical point of view. The algorithm considered for maximizing the likelihood is the expectation–maximization algorithm. The quantification of modal parameters uncertainty is addressed using a Monte Carlo type approach called the bootstrap, which is based on resampling the residuals of the estimated model. Finally, the proposed techniques are applied to synthetic data and also to field data recorded on a stress-ribbon footbridge.