# Systems Integration

In order to leverage the most energy efficient operation out of any single engineering component, care needs to be taken to consider how that component functions with other components that interact with it. No longer can the function of a component be optimized while it is isolated, a systems level integration analysis is needed to optimize system as a whole in order to achieve even the component by component peak performance.

Isotherm surfaces detected from limited measurements

The challenge that arises when optimizing the integration of such systems is that the size (number of variables) increases rapidly and the physical domains that need to be considered are often very disparate. To overcome this challenge, CEED has gathered top researchers in the fields of complex fluid flows, networked sensing and control among many others. CEED has also organized a team that has expertise in analysis of large complex dynamical systems using tools to isolate strong interconnections in systems, efficiently sample key design parameters, as well as experts to perform system reductions to capture the fundamental dynamics of these large complex processes.

Understanding complex fluid flows is essential for any energy efficient process as fluid is often one means to transport energy (and often linked to comfort in systems such as buildings). Professors Meiburg, Yuen, and Mezic have made significant advances in multiphysics modeling of multiphase flows even in complex geometries. For energy efficiency, mixing is essential, and tools have been developed to quantify these processes, along with heat transfer due to convection, and radiation. Based on this research, fluid dynamics of isolated units such as rooms in a building is well understood. Backed by knowledge of individual flow in single volumes or rooms, we strive to advance the understanding of the role of coupling in models that describe multiple rooms and heat and fluid and flows between them.

Understanding the natural dynamics in large interconnected systems is beneficial while in many cases these natural dynamics may need to be controlled for better efficiency, comfort, or other engineering goals. In order to accurately control such large systems, it is necessary to gather large amounts of sensor information. Building systems in particular demonstrate a proliferation of accessible data flows: local temperatures, local air flow rates, flow of people, lighting, status of electrical equipment, and many other variables can be measured. How to efficiently distribute sensors, and how to extract coherent information that captures the relevant behaviors of the system is the focus of research of Professors Bullo and Madhow.

Once adequate information is gathered at the systems-level, decision and control strategies are needed to obtain the desired performance. Due to the wide range of time scales and physical phenomena inherent in these integrated systems, novel decision and control strategies have to be devised. The expertise of professors Bullo and Moehlis in the control of networked engineering and biological systems, such as teams of robots, groups of neurons, and schools of fish, provides a valuable inspiration and insight into the control of large-scale interconnected systems such as those studied at CEED.

The design tools that are involved with the systems integration analysis of complex fluid flows and networked control discussed above results in large complicated physics-based models. Often these models are very challenging to analyze or numerically compute. To overcome this challenge, CEED is focusing on several techniques that target the analysis of large models: design of verification vectors, model reduction to understand characteristic behavior, as well as the identification of interdependency between subcomponents of the large integrated system.

Included in these large interconnected models are lists of parameter values (for component dimensions or characteristics) which are often only known within some range or distribution. To understand how the efficiency of the entire system behaves depending on where this component parameter lies, one must perform exhaustive simulations within the parameter range. Since hundreds of parameters are often uncertain in this way, this task becomes extremely time intensive.

Tools have been developed within CEED to efficiently analyze such systems. Using theoretic concepts in dynamical systems, Professor Mezic has developed a sampling routine (dSample) which efficiently calculates samples within in a parameter range. The procedure results in samples which are much like those available in Monte Carlo and Quasi Monte Carlo methods but are more computationally efficient and scale better with system size.

For generic qualitative understanding of integrated systems as well as advanced control strategies such as Model Predictive Control, reduced order representations of these complex dynamics are needed. Model reduction techniques produce smaller models that capture particular characteristic behaviors of large systems. The approach pursued in the research of Professor Moehlis uses proper orthogonal decomposition on either experimental or numerical data. These techniques have been verified for fluid flows in his group and can be extended or used for other multi-physics systems. Projecting governing equations onto these modes results in reduced-order models that capture enough physics to be meaningful but which are computationally inexpensive.

Within these large interacting systems, each component or sets of components play different roles. It is therefore essential to identify which of these components are strongly connected to key components or the important outputs of the system. Professor Mezic has focused on developing a tool that identifies components that are affected by changes in remote parts of the system, as well as negative effects of such changes on the overall system performance. Currently, such methods can quickly analyze systems involving thousands of interconnected variables, producing both qualitative and quantitative understanding of the inner-workings of these large complex systems.