Jeff Doering

Model Predictive Control of Engine Speed During Vehicle Deceleration

We consider the speed control of a spark ignition engine during vehicle deceleration. When the torque converter bypass clutch is open, the engine speed needs to be kept close to the turbine speed to guarantee responsiveness of the vehicle for subsequent accelerations. However, to maintain vehicle drivability, undesired crossing between engine speed and turbine speed must not occur, despite the presence of significant torque disturbances. Hence, the engine speed during vehicle decelerations needs to be precisely controlled by feedback control, which has to coordinate airflow and spark timing and enforce several constraints including engine stall avoidance, combustion stability, and actuator limits. We develop a model predictive controller that manipulates airflow and spark to track the reference signal for engine speed while enforcing constraints, and synthesize it in the form of a feedback law. The controller is evaluated in simulations and in a vehicle, and it is shown to achieve a responsive and consistent deceleration and the potential for reducing fuel consumption.

MPC-based control of engine deceleration with open torque converter

We consider the speed control of a spark ignition engine with an open torque converter during vehicle deceleration. In these conditions, the torque transmitted from the engine to the wheels through the torque converter, and the corresponding torque load on the engine, is related to the torque converter speed ratio, the ratio between the turbine speed and the engine/impeller speed. The engine speed controller needs to effectively coordinate airflow and spark timing and to enforce constraints including engine stall avoidance, misfire avoidance, and desired actuator operating ranges. We formulate a model of the relevant powertrain dynamics, and propose a parametrization that removes a multiplicative nonlinearity, hence enabling the application of linear quadratic model predictive control (MPC). The MPC controller is synthesized as a piecewise affine state feedback and evaluated in simulations and in an experimental vehicle showing good performance in reference tracking and settling time, which translates in smooth and more consistent deceleration profiles and potential for fuel economy improvements.

Crowd sourced energy estimation in connected vehicles

Accurately forecasting the energy consumption profile of a vehicle is a key requirement of many growing research areas such as horizon based energy management and eco-routing. However, the energy consumption rate of a vehicle depends on many factors making it very difficult to estimate. Many of these factors such as traffic light timing, traffic congestion and weather, change from day to day and trip to trip. While real time traffic information and traffic light timing schedules can be used to help predict the effect of the first two factors, the impact of weather cannot be as easily predicted based on a weather report. Depending on the topology of the route including other vehicles on the road, the local wind speed relative to a vehicle can differ greatly from a predicted bulk wind speed. The effect of precipitation is also difficult to predict because it depends on the amount falling and the amount accumulated on the road. In this paper it is first shown that energy consumption prediction errors due to un-modeled effects, including most notably weather, exhibit a high amount of trip-to-trip variation and a smaller amount of variation within a trip. Next, it is demonstrated that moderate wind speeds have an observable effect on energy consumption and this effect varies based on the direction of travel and wind direction. This analysis also illustrates the challenges in predicting the effect of wind speed and precipitation on energy consumption based on a weather forecast. Finally, a case is made for future research involving the use of current and recent data from a large population of vehicles to provide a more accurate energy consumption profile by reducing the prediction errors due to un-modeled effects.