Toward a digital twin of a mid-speed marine engine: From detailed 1D engine model to real-time implementation on a target platform

Abstract

System complexity is challenging for development of marine mid-speed engines when striving to meet increasingly stringent emission targets. Control-oriented modeling offers a solution, cutting calibration time and enabling robust control strategies. Simultaneously, real-time, physics-based engine models (digital twins) are emerging as they offer better predictive capability and scalability than typical mean-value, data-driven approaches. This study explored development of a control-oriented digital twin of a Wärtsilä 4L20 marine engine. Starting from a detailed one-dimensional model (GT-Suite), it explored reduction strategies toward a fast-running engine model (FRM), balancing the calculation speed and accuracy trade-off. Finally, the FRM was tested for real-time implementation on a target machine. Comprehensive experimental data from the 4L20 platform in the VEBIC Engine Laboratory provided the baseline for model calibration. Model calibration and validation covered four representative operating points and involved correlation of crank-angle, resolved in-cylinder pressures, thermal state at several locations of the engine air-path and relevant performance indicators. The results shed new light on the feasibility of digital twins in the marine engine domain. The obtained FRM was three times faster than real-time, while the accuracy loss was comfortably within the 5% tolerance levels for the governing outputs, including crank angle resolved in-cylinder pressure. The grid-resolved simulation was obtained with four times fewer flow components and internal discretization length of 100% and 150% of the cylinder bore for intake and exhaust components respectively. The balance between predictivity, accuracy and real-time surplus, was ultimately more favorable than in state of the art automotive applications and enables exploring further coupling with semi-predictive emission sub-models. The real-time capable FRM is considered applicable in hardware-in-the-loop simulation, and this application is scheduled in a follow-up project.