A Coupling Architecture for Remotely Validating Powertrain Assemblies

Abstract

Among the myriad of potential hybrid powertrain architectures, selecting the optimal for an application is a daunting task. Whenever available, computer models greatly assist in it. However, some aspects, such as pollutant emissions, are difficult to model, leaving no other option than to test. Validating plausible options before building the powertrain prototype has the potential of accelerating the vehicle development even more, doing so without shipping components around the world. This work concerns the design of a system to virtually couple—that is, avoiding physical contact—geographically distant test rigs in order to evaluate the components of a powertrain. In the past, methods have been attempted, either with or without assistance of mathematical models of the coupled components (observers). Existing methods are accurate only when the dynamics of the systems to couple are slow in relation to the communication delay. Also, existing methods seem to overlook the implications of operating a distributed system without a common time frame. In order to overcome the inherent latency arising from long-range communication, the proposed design combines two features: The exploitation of synchronized clocks for the simultaneous introduction of setpoint commands and the use of observers generated through machine learning algorithms. This novel design is subsequently tested in two scenarios: A simple one, involving the virtual coupling of two parts of an elementary device formed by three rotating inertias, and a more complex one, the coupling between an internal combustion engine and an electric motor/generator as representative of a series or parallel hybrid powertrain. Although the results are heavily influenced by the quality of the data-generated observers, the architecture improves the fidelity of the coupling by nearly an order of magnitude compared to the alternative of directly transmitting the signals. It also opens a niche application that leverages the accuracy of low-fidelity models.