Design of modern engineered systems is characterized by synergistic integration of mechanical, electrical, electronic, computer, optical and control disciplines'what has become known as mechatronics. We use the term "controlled systems" to include all systems where control functions are critical elements of their performance, and the term "co-design" (combined design) to indicate that design of the "plant" in the controls jargon and of the plant controller must be done in a combined, integral manner. We study theoretical foundations for quantifying the coupling between design and control functionality based on optimizing overall system performance. We also study how learning algorithms can be used for autonomous decision making in vehicle powertrain control. Applications include automotive vehicles, elevators, MEMS devices.

Keywords: optimal design, optimal control, global sensitivity, coupling strength, robust control, q-learning, markov decision processes

Incorporating Artificial Intelligence into Internal Combustion Engines
by Andreas A. Malikopoulos, Ph.D Candidate in Mechanical Engineering

Growing requests for better performance and fuel economy, and reduced emissions, have motivated continued research in advanced internal combustion engine technologies. These technologies, such as fuel injection systems, variable geometry turbocharging, variable valve actuation, and exhaust gas recirculation, have increased the number of accessible engine controllable variables, and our ability to optimize engine operation. In particular, the determination of the optimal values of these variables, referred to as engine calibration, have been shown to be especially critical for achieving high engine performance and fuel economy while meeting emission standards. Consequently, engine calibration is defined as a procedure that optimizes one or more engine performance criteria, e.g., fuel economy, emissions, or engine performance with respect to the engine controllable variables. Engine calibration generates a static correlation between the optimal values of the controllable variables and the corresponding steady-state operating points to coordinate optimal performance of the specified criteria. This correlation is incorporated into the electronic control unit of the engine to control engine operation, so that optimal values of the specified criteria are maintained. Current engine calibration schemes cannot guarantee continuously optimal engine operation for the entire operating domain, especially in transient cases encountered in driving styles of different drivers. These schemes rely on dynamometer static correlations for steady-state operating points accompanied by transient vehicle testing. Even for engines with simple technologies, achievement of optimal calibration may become impractical.

In our approach, engine is treated as a controlled stochastic system and engine operation as a stochastic process. The engine calibration is formulated as a sequential decision-making problem under uncertainty. A learning algorithm has been implemented allowing the engine to learn the optimal values of the controllable variables in real time. While the engine is running the vehicle, it progressively perceives the driver's driving style and eventually learns to operate in a manner that optimizes specified performance criteria, e.g., fuel economy, emissions, or engine performance with respect to the engine controllable variables. Consequently, optimal calibration is achieved for steady-state and transient engine operating points resulting from the driver's driving style. The engine's ability to learn its optimum calibration is not limited, however, to a particular driving style. The engine can learn to operate optimally for different drivers, if they indicate their identity before starting the vehicle. The engine can then adjust its operation to be optimal for a particular driver based on what it has learned in the past regarding his/her driving style. A major challenge to this approach is the increase of the problem's dimensionality when more than one controllable variable is considered. To address this problem, a decentralized learning method has been implemented to include two or more controllable variables. In decentralized learning, the algorithm no longer considers all combinations of values of the controllable variables; instead, it establishes a hierarchy of the variables, and learns the optimal values of each variable in parallel.

Downloadable materials: ODE poster

Co-Design of an Artifact and its Controller
by Diane Peters

My research involves co-design - the optimal design of systems in which the artifact design and controller design are coupled. In these systems, the traditional sequential approach of first designing the artifact, then designing an appropriate controller, may not produce a useful design. A simultaneous optimization can be formulated, but this presents challenges in terms of both computational requirements and organizational structure.

In my case study, a microelectrical mechanical system (MEMS) actuator is optimized for maximum displacement and minimum settling time, subject to appropriate constraints on the physical system and its dynamic characteristics. The system is optimized both sequentially and simultaneously, with the simultaneous optimization generating a Pareto solution.

Downloadable materials: ODE poster