A Multidimensional Controls Framework for Diesel Engine Fuel-Air Management, 03-R8179
Inclusive Dates: 08/19/10 09/19/11
Background — Automotive emission standards continue to become stricter around the world, and industry often responds by incorporating innovative combustion and exhaust treatment technologies into new products. In this context of regulatory (e.g., emissions, diagnostics), consumer (e.g., performance), and corporate (e.g., CAFE) requirements, the automotive engine truly provides a multi-objective optimization and a multivariable control problem. The optimization problem pertains largely to operating the combustion and after-treatment system, and yields engine system characterization needed to address the control problem. However, the challenge of controlling the large number of actuators while managing interactions grows rather quickly. This increased complexity exposes the limits of the traditional method of using single-input single-output proportional-integral-derivative (PID) controllers. The main goal of this work was to develop an analytically sound and practically useful extension of the traditional PID controller to a cross-coupled multivariable form. SwRI researchers focused specifically on the diesel engine, but expected the basic results to be applicable to a wide variety of generic control challenges (including gasoline and alternative fuel engines).
Approach — The premise of the SwRI solution to the control problem is intuitively simple — manage production of pollutants and performance at source; i.e., in the combustion chamber. It is only natural that combustion can be better controlled if the conditions in the combustion chamber (also referred to as the cylinder state) are known and steered in the desired direction, and the fuel is matched to the prevailing cylinder state.
This premise naturally divided the work into three main tasks:
Cylinder-state-based fueling. Researchers established fueling as a function of the cylinder-state: oxygen mass, oxygen concentration, intake manifold temperature, and coolant temperature. Researchers showed that such a fueling scheme could become unstable at light loads and provided an alternative solution for that region of engine operation.
Active management of cylinder-state. The cylinder-state-based fueling manages pollutant production at source, but the cylinder-state must be managed properly to deliver the desired performance (torque response, emissions etc.). SwRI researchers spent most of the time on this part of the problem and derived a systematic method to devise multivariable PID controllers that account for and exploit the natural cross-coupling among the subsystems.
Adaptation of standard design of experiment (DoE). Items (1) and (2) require systematic and efficient ways to collect and process experimental data. SwRI adapted the standard methods of DoE within the framework of available experiment space. For example, the gridding (or levels) normally called for in the traditional DoE cannot be applied directly to the problem at hand because the so-called "independent" variables can be controlled only indirectly.
Accomplishments — Researchers began the effort with a physically based dynamical model of the diesel engine with a dual-loop exhaust gas recirculation (EGR) system. Some elementary, but powerful ideas from network and graph theory were applied to develop a compact representation of the engine system. Subsequently, this representation was used to develop a general method for design of controllers given a set of target states and inputs. In doing so, researchers "factored" the problem into dynamical and algebraic portions. Some ideas from the sliding mode control theory were adopted to address the dynamical portion and from multi-port network description to address the algebraic portion. Researchers developed methods to handle the issues of windup, over- and under-actuation and actuator saturation. On the fueling side, formulas were developed to compute fueling parameters suitable for the given cylinder state. The SwRI team conducted a stability analysis of the fueling scheme and showed ways to achieve stable combustion in all regions of engine operation (idle, light-load lean, lean, and rich). Over the course of the project, several dozen controllers were constructed and evaluated, and a combination of four controllers was selected to cover the entire operating range of engine including cold start.
Researchers evaluated the controller on two platforms: a 2-liter-class diesel passenger vehicle, and a 13-liter-class diesel truck. The evaluation criterion was to simultaneously meet the performance and regulated emission norms. The passenger vehicle was driven on the standard European (NEDC) and U.S. (FTP75) test cycles. At of the end of the project, the vehicle was close to meeting the NEDC standard without any NOx after-treatment. The vehicle also showed high potential to meet the FTP75 standard with NOx after-treatment. The truck engine was subjected to the standard U.S. (HDFTP) test cycle. The emission and performance metrics met or exceeded the standard. The method was applied successfully to two significantly different test platforms, establishing scalability of the approach.
The analytical depth of controller development and favorable results of testing allow researchers to be excited about the future of this method. SwRI expects the abstract thought process and its software implementation will form the foundation of its work in simultaneously meeting the aforementioned triad of requirements — performance for the consumer, fuel economy for the corporations and emission for the regulators. The methods are generic enough to be applicable to a large variety of control challenges.