Hide menu

Design and Optimization of Cyber-Physical Systems

Supervisor: Professor Zebo Peng, ESLAB, Department of Computer and Information Science

Context and Research Area:

There is a spectacular increase of cyber-physical systems (CPS) where the computational components interact tightly with the physical world. Such systems are used in a huge spectrum of application areas, including aerospace, manufacturing, chemical processes, healthcare, vehicles, telecommunication, and consumer appliances. Many of these systems are safety critical, such as automotive electronics and medical equipment, with stringent reliability and real-time constraints. In automotive systems, besides the safety-critical functions with high demands on reliability and strict timing constraints (e.g., brake-by-wire, active safety, and autonomous driving features), we have also functions that demand very high computation capacity with guaranteed quality-of-service (e.g., image/speech recognition, intelligent navigation, and sensor fusion). The increased complexity of all these functions can only be addressed by deploying heterogeneous multiprocessor architectures implemented with the most advanced silicon technology. 

We are therefore facing the challenge of how to design and build such complex heterogenous architecture to satisfy all the functional and non-functional requirements a CPS. This challenge has to be addressed at multiple levels of abstraction, including hardware, architecture, software, and control, since cross-layer optimization is the only way to efficiently design a CPS to satisfy all the design constraints. 

This research project will be part of ESLAB's long-term efforts in developing system-level design and optimization techniques for various CPS applications and architectures. System-level design decisions, such as functionality mapping, task scheduling, and power management are crucial in order to achieve timeliness, reliability, and power efficiency, as well as the expected level of service in complex CPS. Other system-level techniques can be applied at run-time and will be implemented as run-time management mechanisms. Such techniques can be used to deal with mode changes or criticality-level switches that occur at runtime either as responses to external events or at predetermined moments in time. For example, a mode change occurs when some control loops are deactivated, some are activated, or both. This leads to a change in the execution and communication demand and, consequently, a new schedule and some new controllers must be used to achieve the best possible control performance in this new mode. In this context, advanced techniques are needed to address the design and optimization of multi-mode CPS applications that are mapped on a heterogenous execution platform, which is dynamically reconfigured due to not only dynamic changes of workload, but also disturbances, such as hardware failure and malicious attacks. 

Objectives:

A CPS requires timely and reliable delivery of information from sensors to the computing fabrics and control signals from the computing fabrics to actuators of the physical world. This PhD project will deal with the issues of how to manage the limited resources, such as computation power and communication bandwidth, to meet the timing requirements of a CPS as well as graceful degradation of a CPS in the presence of various disturbances including hardware failure and malicious attacks. Advanced modeling and optimization techniques, including machine-learning based, will be used to achieve the required degree of adaptability, safety, robustness, and energy efficiency.

Required Qualification:

The applicant should have a degree in computer science, computer engineering, or closely related fields. Candidates with strong background in algorithms design/analysis, optimization techniques, machine learning, and programming will also be preferred.