There has been an enormous global research effort to alleviate the current and projected environmental consequences incurred by internal combustion (IC) engines, the dominant propulsion systems in ground vehicles. Two technologies have the potential to improve the efficiency and emissions of IC engines in the near future: variable valve actuation (VVA) and homogeneous charge compression ignition (HCCI).
IC engines equipped with VVA systems are proven to show better performance by adjusting the valve lift and timing appropriately. An electro-hydraulic valve system (EHVS) is a type of VVA system that possesses full flexibility, i.e., the ability to change the valve lift and timing independently and continuously, making it an ideal rapid prototyping tool in a research environment. Unfortunately, an EHVS typically shows a significant response time delay that limits the achievable closed-loop bandwidth and, as a result, shows poor tracking performance. In this thesis, a control framework that includes system identification, feedback control design, and repetitive control design is presented. The combined control law shows excellent performance with a root-mean-square tracking error below 40 µm over a maximum valve lift of 4 mm. A stability analysis is also provided to show that the mean tracking error converges to zero asymptotically with the combined control law.
HCCI, the other technology presented in this thesis, is a combustion strategy initiated by compressing a homogeneous air-fuel mixture to auto-ignition, therefore, ignition occurs at multiple points inside the cylinder without noticeable flame propagation. The result is rapid combustion with low peak in-cylinder temperature, which gives HCCI improved efficiency and reduces NOx formation. To initiate HCCI with a typical compression ratio, the sensible energy of the mixture needs to be high iv compared to a spark ignited (SI) strategy. One approach to achieve this, called recompression HCCI, is by closing the exhaust valve early to trap a portion of the exhaust gas in the cylinder.
Unlike a SI or Diesel strategy, HCCI lacks an explicit combustion trigger, as autoignition is governed by chemical kinetics. Therefore, the thermo-chemical conditions of the air-fuel mixture need to be carefully controlled for HCCI to occur at the desired timing. Compounding this challenge in recompression HCCI is the re-utilization of the exhaust gas which creates cycle-to-cycle coupling. Furthermore, the coupling characteristics can change drastically around different operating points, making combustion timing control difficult across a wide range of conditions. In this thesis, a graphical analysis examines the in-cylinder temperature dynamics of recompression HCCI and reveals three qualitative types of temperature dynamics. With this insight, a switching linear model is formulated by combining three linear models: one for each of the three types of temperature dynamics. A switching controller that is composed of three local linear feedback controllers can then be designed based on the switching model. This switching model/control formulation is tested on an experimental HCCI testbed and shows good performance in controlling the combustion timing across a wide range. A semi-definite program is formulated to find a Lyapunov function for the switching model/control framework and shows that it is stable.
As HCCI is dictated by the in-cylinder thermo-chemical conditions, there are further concerns about the robustness of HCCI, i.e., the boundedness of the thermochemical conditions with uncertainty existing in the ambient conditions and in the engine’s own characteristics due to aging. To assess HCCI’s robustness, this thesis presents a linear parameter varying (LPV) model that captures the dynamics of recompression HCCI and possesses an elegant model structure that is more amenable to analysis. Based on this model, a recursive algorithm using convex optimization is formulated to generate analytical statements about the boundedness of the in-cylinder thermo-chemical conditions. The bounds generated by the algorithm are also shown to relate well to the data from the experimental testbed.