Optimal Modular Control with Applications to Automotive Powertrain Control
The success of modern industrial control strategies requires the consideration of many competing performance criteria through a controller that is easily calibrated and computationally tractable. This is particularly true for automotive powertrain control, where it is important to manage the competing interests of fuel economy, drivability, and emissions. This talk will examine the use of model predictive control (MPC) in order to manage competing performance interests, subject to constraints. In order to reduce computational load and obtain a tractable optimization, we employ a modular control approach, wherein the optimization only considers a subset of the overall system model and legacy controllers are incorporated to address other aspects of the system. To ensure seamless integration of the different modules, we fuse concepts from model reference control with our use of MPC, specifying a dynamic reference model for each module that reflects desired input-output performance of the module and is commonly known throughout the system. The first part of the talk will examine theoretical properties relating to the integration of the modules, focusing especially on the derivation of analytical modular vs. centralized performance comparison results for a specific linear framework, highlighting the role that the reference model plays in our derivations. The second part of the talk will examine the use of modular MPC for two powertrain applications. First, we will examine a thermal management system that is used for tightly regulating coolant and oil temperatures during the operation of an engine test cell. Second, we will consider an engine torque controller for a spark ignition (SI) engine, where modular MPC is used to manage fuel economy, drivability, and emissions in a computationally tractable manner. The talk will conclude with an examination of avenues for future work, many of which focus on closing the gap between our theoretical and applications results.
Chris Vermillion received his Ph.D. in Electrical Engineering from the University of Michigan in 2009, and received his undergraduate degrees in Aerospace and Mechanical Engineering from the University of Michigan in 2004. In his dissertation work, Chris studied optimal modular control strategies for overactuated systems, with an emphasis on applying the research results to advanced automotive powertrain control. His research interests include optimal control (including model predictive control), decentralized control, and robust control. Chris currently works as a Postdoctoral Researcher at the Toyota Technical Center in Ann Arbor, Michigan. He is currently applying optimal control strategies to the problem of simultaneously managing competing performance interests in advanced automotive powertrain control.