Abstract

Predicting the response of nonlinear dynamical systems subject to random, broadband excitation is important across a range of engineering disciplines, such as structural dynamics. Building data-driven models requires experimental measurements of the system input and output, but it can be difficult to determine whether inaccuracies in the model stem from modelling errors or noise. Therefore, there is a need to determine the maximum component of the output that could theoretically be predicted using the input if an improved model was to be developed through the investment of resources. This talk presents a novel method to identify the component of the output that could potentially be modelled, and quantify the level of noise in the output, as a function of frequency. The method uses input-output measurements and an available, but approximate, model of the system. A trainable, frequency dependent parameter balances an output prediction generated by the model with noisy measurements of the output to predict the input to the system. This parameter is utilised to estimate the noise level and then calculate a nonlinear coherence metric as a measure of causality or predictability from the input. There are currently no solutions to this problem in the absence of an accurate benchmark model.