CONTROL CO-DESIGN WITH VARYING AVAILABLE INFORMATION APPLIED TO VEHICLE SUSPENSIONS

Recent optimization strategies for Control Co-Design (CCD) often utilize open-loop optimal control (OLOC) to explore the physical performance limits of actively controlled engineering systems. For most real systems, however, closed-loop control (CLC) is required for implementation. OLOC methods incorporate the use of present, past, and future information in making control decisions at each point in time. For systems with any uncertainty, CLC is needed for stability and robustness. The physical (plant) design generated by an OLOC CCD method will normally not interact optimally with CLC, producing results that are not system optimal. The ideal outcome of CCD optimization is the combined physical and control system design that produces maximum system utility, while accounting for the realities of implementable control systems, such as causality and other limits on information available to inform real-time control decisions. In this article an intuitive strategy is presented for investigating empirically the impact of information availability on CCD optimization results. Model Predictive Control (MPC) provides a flexible means to vary what information is used in making realtime control decisions. This is used as a proxy for the vast space of potential controllers, from simple to sophisticated. This method for studying information-based characteristics of CCD problems is demonstrated using a canonical CCD problem based on an active automotive suspension problem. Different plant architectures with various plant design variables are considered. Results show that varying the amount of information in the control design yields different plant designs and different objective values, and has the potential to yield insights into promising CLC architectures (beyond MPC), fruitful directions to head for plant design, and a deeper understanding of the interface between physical and control system design. It is also observed in the studies here that by using more advanced system architectures, the MPC prediction horizon can be reduced and still produce superior system performance compared to cases with simpler architectures and a longer prediction horizon, or, even when using an open-loop controller. Comparison studies are performed that provide in-depth knowledge of the effect of control sampling time and plant design on states and control. Furthermore, a hybrid Kalman filter is designed that couples with the MPC controller to provide state estimation in the presence of measurement noise and process noise. This article introduces the concept of information-based studies in CCD, but utilizes an applied approach based on MPC to generate insights. A more theoretical approach could be taken in the future that yields more generalizable understanding of how information limitations influence CCD optimization outcomes.