Robust Design of Cyber-Physical Systems
Control based on Data-Intensive Sensing


The design of next generation high-tech professional systems for medical imaging, lithography, smart electricity grids, intelligent transportation, electron microscopy and high-end printing requires a tight coordination between computation, communication and control elements (the cyber part) on the one hand, and physical processes such as heating, cooling, motion, vibrations, etc. (the physical part) on the other hand. Despite the need for integrated design of these so-called cyber-physical systems (CPS), the corresponding scientific disciplines have predominantly developed independently. This separation of disciplines can no longer be sustained and urgently needs to be bridged. This is particularly urgent at present, as our high-tech industry is currently facing challenges of (i) exponential growth in complexity, due to increasing numbers of sensor and actuators in systems, increasing performance requirements, the trend to multiprocessing, increased network connectivity, and increased dynamic interaction between systems, components and environment; (ii) increasing uncertainty, due to increasing miniaturization of electronic components, changing conditions in the physical environment, unpredictable workload variations and failures in the cyber part, and incomplete information about the properties of components and the environment they have to operate in.


The project of Control based on Data-Intensive Sensing is part of the program on Robust Design of Cyber-Physical Systems. It aims at providing a multi-disciplinary design approach for complex and data-intensive control problems that is cost-effective, utilizes resources effectively, and results in high performance and robustness. An exemplary case of data-intensive control is found in electron microscopy. At the current state of the art, the observation of sub-Ã…ngstrom details requires complex mechanisms to navigate and stabilize the sample under investigation. To decrease the currently high stabilization time and to improve the overall performance, new control concepts and high-speed data acquisition and processing techniques need to be developed and integrated. In particular, the microscopic images themselves are indispensable as a source of information for the feedback control part to position the sample and adjust the electron beam to keep the sample in focus. The facts that high performance control is only possible based on high-quality sensing information, and that high-quality sensing information is only available when high-performance control is realized, create a vicious coupling that can only be circumvented by intensive computation on preliminary sensor data. This setting results in a complex hybrid control problem in which discrete and continuous decisions have to be made both offline and during run-time: different actuators can be used (adjusting beam or (re)positioning the sample), different processing algorithms can be applied (facing a trade-off between fast/inaccurate and slow/accurate computations), and the frequency, order and reliability of sending and computing data can be scheduled in different ways.

Useful links