Real-time predictive model for reactivity controlled compression ignition marine engines

Publication channel


Year of the publication



Storm, X., Vasudev, A., Shamekhi, A-M., Modabberian, A., Zenger, K., Hyvonen, J., Mikulski, M. Real-time predictive model for reactivity controlled compression ignition marine engines, Control Engineering Practice, Volume 141, 2023.



Related to:


Model-based design is proven to be essential for the development of control systems. This paper presents a real-time predictive control-orientated model (COM) for low-temperature combustion (LTC), dual-fuel, reactivity-controlled compression ignition (RCCI) engines. A comprehensive model-based design methodology must be capable of constructing an RCCI control-orientated model with high accuracy, high noise immunity, good response, predictivity in governing mechanisms, and low computation time. This work attains all of these for the first time for a cutting-edge RCCI marine engine. The real-time model (RTM) captures the key sensitivities of RCCI by controlling the total fuel energy and the blend ratio (BR) of two fuels, while also considering uncertainties arising from variations of inlet temperature and the gas exchange process. It provides not only the cycle-wise combustion indicators but also the crank-angle-based cylinder pressure trend. The RTM is derived by direct linearisation of a physics-based model and is successfully validated against experimental results from a large-bore, RCCI engine and the previously acknowledged UVATZ (University of Vaasa Advanced Thermo-kinetic Multi-zone) model. Validation covers both steady-state and transient modes. With high accuracy in several case studies representing typical load transients and air-path disturbance rejection tests, the model predicts maximum cylinder pressure (Pmax), crank-angle of 5 % burnt (CA5), crank-angle of 50 % burnt (CA50) and indicated mean effective pressure (IMEP) with root means square (RMS) errors of 8.6 %, 0.3 %, 0.6 %, and 0.6 % respectively. The average simulation time without any code optimisation is around 5 ms/cycle, offering sufficient real-time surplus to incorporate a semi-predictive emission submodel within the current approach.