Physics-based and data-driven modeling of the parameter-varying vibration dynamics of a simplified gantry manipulator

The work of this conference paper considers a simplified cartesian manipulator of a gantry robot set up as a pinned-pinned Euler-Bernoulli beam with the purpose of analysing the structural dynamics when subjected to a moving load/cart and a surface roughness. A physics-based beam-cart model is created utilising principles of physical domain substructuring methods to create coupled system matrices for modal analysis purposes. Here, the frozen modal properties as well as frequency response functions are calculated and used in a time simulation of the induced vertical vibrations when the cart traverses the beam. The simulated time signal of the physics-based model is extracted and used for further data-driven system identification. Here, the Linear Parameter-Varying Auto Regressive model is used with a Maximum Likelihood estimator as well as a Bayesian non-linear regression estimator to calculate the frozen power spectral density and natural frequencies. Furthermore, the two estimators are compared where they demonstrate similar predictive performance whereas the Bayesian estimator yields lower standard deviation of the calculated parameters.