Hybrid Modeling for Condition Monitoring in Digital Twin Systems

Digital twin (DT) modeling is an emerging framework for modeling complex system which can improve modeling fidelity and accuracy. This improved modeling accuracy can be used to improve performance and reliability of these systems through performance optimization and condition monitoring (CM). One aspect of CM that can be potentially greatly improved is soft sensing, also known as virtual sensing or smart sensing, which is estimating the values of parameters or states without directly measuring them. Because a DT seeks to virtually replicate a system, it is necessary to model a great many parameters and states, and it is not economically feasible to directly measure each of them. To overcome this, soft sensing is used to estimate these using models. Because there is risk, uncertainty, and inaccuracy associated with relying on just one model, it is ideal to utilize multiple models. Hybrid modeling uses several models to improve the accuracy, precision, and reliability of these estimates. There are several approaches seen in the literature which can broadly be categorized as series, parallel, and combined models. Each has their own advantages and use cases, and they were each shown to improve CM capabilities of their respective systems. This work examines Hybrid modeling for CM in the context of DTs, and displays the effectiveness through examining existing literature applying the concept.