"Model-Based Design of Uncertain Systems: Why and How"

Alessandro Pinto, United Technologies Research Center Inc.

June 3rd (Friday), 3:00pm
Webb Hall 1100

There are several reasons why design paradigms for cyber-physical systems should include the notion of uncertainty. The physical side of the system (i.e. the plant that the cyber side controls) is only known to some extent. For example, the weather condition that an aircraft will face during a mission is a random parameter. Even when the dynamic behavior of the environment is not subject to any randomness, the parameters of its model may not be known exactly either because difficult to measure, difficult to control during the manufacturing process, or simply because of an abstraction process required to limit the complexity of an accurate model. The cyber side of the system comprises the control software, the hardware and the communication network, and is subject to random failures and data dependent performance metrics. Optimal control strategies rely on the solution of optimization problems whose run-time depends on the input data. Moreover, the worst case execution time of software is data dependent because of low level implementation techniques such as cache memories, branch prediction, pipeline execution etc. Notoriously, communication delays are also uncertain, especially when collision-based and wireless protocols are used.

Many techniques have been developed to analyze systems in the presence of uncertainty such as fault tree analysis and Markov Chain based methods including probabilistic model checking. Together with analysis techniques, several languages have been introduced to capture uncertain systems, such as Stochastic Petri Nets and Stochastic Automata Networks. These tools and languages provide a solid groundwork to enable the analysis and design of uncertain systems. However, the adoption of these tools in industrial design flows is not straightforward. There is a semantic gap between the system specification captured using domain specific languages and the input to many analysis and design flows for uncertain systems. Furthermore, verification of correctness and assessment of performance are not the only type of questions to be answered. A realistic design environment should also provide some parametric analysis capability to perform design space exploration.

In this talk we present several model-based design tools and methods for dealing with uncertain systems. We first present an integrated end-to-end tool-chain that provides a flow form high level description to probabilistic analysis. We then highlight the challenges and present our effort to deal with the complexity induced by hybrid dynamics. By the end of the talk, we will add a question mark to the title, thereby turning the original intent of proposing solutions to concrete problems, into a quest for answers to the many hard riddles that are still open in this field.

About Alessandro Pinto:

Alessandro Pinto is a researcher in the Systems Department at the United Technologies Research Center (UTRC), Inc., Berkeley, California. His research interests are in the area of computer aided design for cyber-physical systems. He received a Ph.D. degree in Electrical Engineering and Computer Sciences from the University of California at Berkeley in 2008, and a M.S. degree in Electrical Engineering in 2003 from the same University. He holds a Laurea degree from the University of Rome “La Sapienza”. In 1999, he spent one year as a consultant at Ericsson Lab Italy in Rome, Italy, working on the design of systems-on-chips. He consulted for the same company from 2000 to 2001, developing system-level design flows for wireless access networks.

Hosted by: CCDC Seminar, Departmental host: Francesco Bullo