Low temperature combustion engines such as homogeneous charge compression ignition (HCCI) offer fuel flexibility with high fuel efficiency and low emissions. If the type and composition of the fuel, such as bio-fuel, is not known a-priori, a ?smart? engine has to be capable of sensing heat release and adjusting combustion system parameters online for minimized emissions and fuel consumption. This necessitates a more advanced adaptive control schemes for the control of these types of complex non-affine nonlinear systems
The overall goal of this study is to provide the next generation adaptive critic neural net controllers for complex non-affine, nonlinear systems supported by a rigorous and repeatable design and mathematical framework. The controller performance will be validated for the HCCI engine for a range of bio-mass based fuel stocks using conventional and novel input sensors for measuring cyclic heat release.
Intellectual Merit: The project will advance the state of the art in Adaptive Dynamic Programming for control by providing rigorous mathematical analysis for convergence and stability, and performance guarantees in the presence of approximation errors, actuator constraints and delays. Moreover, by applying the theoretical results to an emerging control application of fuel-flexible engines, this type of controllers will be implemented and tested in hardware in the Co-PI's internal combustion engine laboratory.
Broader Impact: Improved control of next generation fuel-flexible engines and multi mode engines, such as plug-in hybrids, is expected to improve fuel efficiency and reduce harmful emission, thus directly impacting the environment and reducing dependence on foreign oil. Research results will be integrated as part of undergraduate course and laboratories. Dissemination plans include distribution of software through websites, patents, journal and conference publications. The PIs have a track record of hiring underrepresented minorities through MST?s Minority Engineering Program, extending research opportunities to undergraduates via REU supplements and interactions with EPSCOR states. International collaborations will be pursued. Technology transfer to industrial members is planned through the NSF I/UCRC Site on Intelligent Maintenance Systems where the PI is the Site Director.
Unlike typical "diesel" engines, coupling the implementation of these advanced combustion strategies with the desired use of alternate fuel sources such as bio-fuels presents combustion phasing issues due to strong dependence of these advanced combustion modes on the physical and chemical properties of the fuel. To implement these advanced combustion modes in a fuel-flexible sense requires a "smart" engine capable of sensing heat release patterns, where the fuel is not known a-priori, and adjust combustion system parameters for minimized emissions and peak fuel efficiency. Similarly, the engine will operate in conventional mode when high power operation is needed. The proposed work seeks to develop a near optimal nonlinear adaptive controller using approximate dynamic programming (ADP), based on a phenomenological LTC engine model, through which heat release pattern sensing will provide a path to control of a LTC engine. The conceived controller will allow the use of fuel from non-traditional sources (e.g. bio-mass) to be used without advanced knowledge of the fuel’s physical or chemical composition and engine dynamics through sensing and control of heat release pattern phasing. Specifically, the variation in combustion behavior for the bio-fuels has led to development of two models representing two classes of fuels: pure hydrocarbons and oxygenated fuels. Utilizing both the model and the experiments, greater understanding has been gained toward the engine’s operational characteristics at the limits of combustion. The HCCI engine capable of multiple fuel type is represented as an nonaffine nonlinear discrete-time system in the input-output form. The controller development for the nonaffine nonlinear discrete-time system has been revised and demonstrated irrespetive of the fuel utilized due to the neural network based optimal adapive controller. Project outcomes include several journal and conference articles, support of two doctoral students and one undergraduate student.
