Mean value modeling and control of a diesel engine using neural networks

Abstract

Increasingly stringent emissions legislation and demands for improved fuel economy have mandated the need for advanced control algorithms and complicated the diesel engine calibration procedure. To this end, a neural networkbased mean value model of a modern turbocharged direct injection diesel engine has been developed and validated. For a pre-specified engine speed schedule and control vector trajectory, the engine model was shown to produce accurate predictions of the turbocharger manifold charging dynamics and combustion efficiency through engine brake torque predictions. The mean value model was coupled with several sub-models to predict exhaust gaseous and particulate emissions and satisfactory predictions were reported over highly transient engine test schedules. The mean value model was used to develop and validate through simulation a neural network-based engine torque controller for both non-governed as well as governed engine operation. Two types of proportional governors were considered where one governor employed a more aggressive fueling strategy than the other. The engine performance and exhaust gas emissions for both strategies were quantified through simulation, showing steeper rises in torque and larger excursions in transient emissions for the more aggressive fueling strategy. The controller was adapted online using the standard back-propagation algorithm. For a pre-specified engine speed schedule and desired engine torque trajectory, excellent torque tracking was predicted using the neural network (NN) controller over transient operation compared to a classical proportional plus integral (PI) controller, which was tuned heuristically. The mean value model was also used to develop and validate through simulation a neural network-based all-speed governor. For a pre-specified engine load schedule and accelerator position trajectory, accurate tracking was predicted for the desired engine speed for both classical and NN-based controllers under high load transients. For the test engine used, it was shown through simulation that tighter control of engine torque over the Federal Test Procedure (FTP) cycle resulted in higher brake specific emissions of carbon dioxide, oxides of nitrogen, and particulate matter. Also, the EPA validation criteria for the prescribed engine torque over the FTP cycle allow for significant variations in brake specific emissions, especially particulate matter, total hydrocarbons, and carbon monoxide emissions, while still meeting the legal requirements for a valid engine certification test. iii ACKNOWLEDGEMENTS The author wishes to express his gratitude to his thesis advisor, Dr. Chris Atkinson, for the invaluable opportunity he has given him to work on this very interesting subject. Thanks for many unlimited office hours of sincere discussions on engine controls, a subject greatly complicated by confidentiality agreements and propriety information. The author wishes to thank his committee members, specifically Dr. Napolitano for accepting him as a 'guest' in his advanced con...

Performance and Emission Modelling and Simulation of Marine Diesel Engines using Publicly Available Engine Data

“…Taking into account the possibilities that a mean-value model can bring to the research, which needs a model of the engine as part of a wider range model, recently more attention has been paid to this type of model; see, for example, Lee [58], Hendricks [59], Sengupta [60], Yacoub [61], Theotokatos [62], Guzzella [63], Yum [64], Scappin [65], Kharroubi [66], Baldi [67], Altosole et al [68].…”

Modelling of Diesel Generator Sets That Assist Off-Grid Renewable Energy Micro-grids

“…However, PID schemes might not be able to handle large variations in the engine dead time. Here, several advanced control techniques for the speed control of the diesel engine systems have been reported in the literature, based on the methods of H∞ [19], adaptive control [20] and neural network [21].…”

MPC-Bases energy mangement system for hybrid renewable energies

“…However, PID schemes might not be able to handle large variations in the engine dead time. Here, several advanced control techniques for the speed control of the diesel engine systems have been reported in the literature, based on the methods of H∞ (Kuang et al, [144]), adaptive control (Roy et al, [145]) and neural network (Yacoub et al, [146]).…”