ABSTRACT
Most real systems have multiple degrees of freedom. Additionality, real physical systems may be nonlinear. These two issues are addressed in the current chapter thereby taking the digital twin development process to a high level of fidelity. Unscented Kalman Filter uses concepts of unscented transform to analyze nonlinear models and approximates the mean and co-variance of the targeted distribution instead of approximating the nonlinear function. Gaussian process regression (GPR) along with neural network are perhaps the most popular machine learning techniques in today's time. A major concern in Digital Twin (DT) is its connectivity with the physical twin; in absence of which, a DT will be of no practical use. It can be observed that the proposed approach yields highly accurate estimate of the state vectors. However, the divergence is observed approximately after 3.5 years from the last observation, which for all practical purpose is sufficient for condition-based maintenance.
