ModelConductor: An On-Line Data Management Architecture for Digital Twins

In this article we introduce a software architecture, along with an experimental implementation called ModelConductor, for input-output data management in on-line digital twin (DT) simulation model applications. The goal here is to facilitate asynchronous and standardized data interexchange between a (potentially computationally expensive) computer simulation model and a physical device (with potential latency in measurement data). The presented object-oriented architecture defines a DT design pattern by introducing a set of abstract methods for data retrieval, manipulation and consumption for communication of physical and digital assets. As a concrete example, we demonstrate the use of ModelConductor in machine learning based on-line prediction of diesel engine nitrogen oxide (NO
x
) emissions, using simulated live data from a real engine.