First Steps in Probabilistic Hybrid Twin Extrapolation Based on Nonparametric Probabilistic Method and Machine Learning Algorithms

Abstract

With emerging mechanical engineering applications on a daily basis, the incorporation of data into models and data-driven modeling are subjected to constant improvement nowadays. This work tackles the case of incorporating experimental variability coming from experimental measurements into a deterministic model, ending up with an updated stochastic hybrid-twin of the investigated object. Moreover, the work proposes that a limited measurement on a limited part of a system of interest, can be used to extrapolate the enhanced stochastic model onto the complete system. Thus, the non-parametric probabilistic method (NPM) is used to built an enhanced updated stochastic model on a subsystem of a complete system, where the measurements are available. Later on, an online learning algorithm will extrapolate quasi-instantly the results into the complete system. A comparison on a simplified 1D beam model built using NPM on a complete system, and the one obtained by the proposed extrapolation technique from a subsystem, showcases the success of the suggested method.