Understanding nonlinear systems with Gaussian processes: a new look at restoring forces  Seminar

The restoring force surface method for nonlinear system identification is a classical approach which is both elegant and powerful in structural dynamics. In this talk we will look at a modern take on this theme based on the use of Gaussian processes, a Bayesian nonlinear regression tool from the statistics and machine learning communities. It will be shown how these models can be used to efficiently recover equations of motion from nonlinear system response data. Examples will be shown in both input-output and output-only settings.

Speaker information

Dr Timothy Rogers is a Senior Lecturer in the Dynamics Research Group at The University of Sheffield. His research interests are focussed on effective use of data in engineering dynamics problems. This spans problems in linear and nonlinear system identification, structural health monitoring and model calibration/validation under uncertainty. Most of the methods used are probabilistic and in many cases a Bayesian viewpoint is adopted. The span of industrial problems for which these methods are relevant is very large, however, specific applications of interest are, load estimation for offshore structures, crack detection using ultrasonic waves, semi-supervised damage detection.