Stefan Bosse, Frank Kirchner, Smart energy management and low-power embedded system design, SPIE Newsroom, 2011,
DOI:10.1117/2.1201106.003694. Paper PDFPublisher
Today there is an increasing demand for miniaturized smart sensors embedded in sensorial materials and smart actuators. Each sensor and actuator node provides some kind of sensor, electronics, data processing, and communication. With increasing miniaturization and sensor-actuator density, decentralized network and data processing architectures are preferred, but energy supply is still centralized. Using local energy-harvesting technologies a decentralized energy supply can be provided, too. Energy harvesting, for example using solar cells or thermo-electrical sources, actually delivers only low electrical power (due to technology or size constraints). We propose and demonstrate a design methodology for embedded systems satisfying low power requirements suitable for self-powered sensor and actuator nodes. This design methodology focuses on 1. smart energy management at runtime using advanced computer science algorithms (artifical intelligence) and 2. application-specific System-On-Chip (SoC) design using High-level synthesis at design time. Low-power systems are designed on algorithmic, rather than on technological level. Smart energy management is performed spatially at runtime by a selection from a set of different (implemented) algorithms classified by their demand of computation power, and temporally by varying data processing rates. It can be shown that power/ energy consumption of an application-specific SoC design depends strongly on computation complexity.
[c11.1]
T. Behrmann, C. Budelmann, M. Lemmel, S. Bosse, Tool chain for Harvesting, Simulation and Management of Energy for Sensorial Materials , Euromat 2011 Conference, Montpellier (Frankreich), 12.-15. September, 2011, 2011.
[c11.2]
F. Pantke, S. Bosse, D. Lehmhus, M. Lawo, M. Busse, Combining Simulation and Machine-Learning for Real-Time Load Identification in Sensorial Materials, Proceedings of the International Conference SIMBIO-M-2011, Simulations in BIO-Sciences and Multiphysics, 20-22.6.2011,
Marseille, France, 2011. Presentation PDF
Sensorial materials are biologically inspired technical systems – materials which are able to “feel” based on integrated miniaturized sensor networks, communication and data processing facilities. Seen from an engineering perspective, the nervous systems of animals complement the latter’s load-bearing structures, allowing active protection by, e.g., utilizing the notion of pain in cases of unexpected or accidental loads. Transfer of this basic principle to engineered technical structures ranging from bridges to robot arms, prosthetics, and implants affords studies in several fields. Simulation is one of them, both in the role of a supporting tool for layout and evaluation of such systems, and as a basis of internal sensor signal evaluation. Several links to classic biomaterials research exist: Smart implants can be envisaged which monitor healing, their condition, as well as changes in their integration with the natural structure, or prosthetics that autonomously infer, record, and monitor their loading history and give immediate feedback to their wearer and/or designer. The technical approach of the present study is to bring together multi-agent based simulation and finite element analysis. It is under the assumption that, in foreseeable future, the considerable computational power necessitated by the latter technique cannot be efficiently miniaturized to the scale required for embedment within the structure and, thus, is only available at design-time. As an intermediate step, to gain knowledge how sensorial structures can most effectively be built, an Artificial Intelligence based process for the construction of such structures was developed and realized in an experimental setup which uses machine learning for fast load location, force, and deformation identification. It is presented in this paper, along with evaluation results obtained in experiments using a finite element model and sensor data from a strain-gauge equipped plate which demonstrate the general practicability.
[c11.3]
T. Behrmann, M. Lemmel, S. Bosse, Energy management for self-powered sensor nodes, Energy Harvesting & Storage Europe 2011, München (Deutschland), 21.-22. Juni, 2011, 2011.
[c11.4]
Stefan Bosse, Thomas Behrmann, Smart Energy Management and Low-Power Design of Sensor and Actuator Nodes on Algorithmic Level for Self-Powered Sensorial Materials and Robotics, Proceedings of the SPIE Microtechnologies 2011 Conference, 18.4.-20.4.2011, Prague, Session EMT 101 Smart Sensors, Actuators and MEMS, 2011,
DOI:10.1117/12.888124. Paper PDFPresentation PDF
We propose and demonstrate a design methodology for embedded systems satisfying low power requirements suitable for self-powered sensor and actuator nodes. This design methodology focuses on 1. smart energy management at run-time and 2. application-specific System-On-Chip (SoC) design at design time, contributing to low-power systems on both algorithmic and technology level. Smart energy management is performed spatially at runtime by a behaviour-based or state-action-driven selection from a set of different (implemented) algorithms classified by their demand of computation power, and temporally by varying data processing rates. It can be shown that power/energy consumption of an application-specific SoC design depends strongly on computation complexity. Signal and control processing is modelled on abstract level using signal flow diagrams. These signal flow graphs are mapped to Petri Nets to enable direct high-level synthesis of digital SoC circuits using a multi-process architecture with the Communicating- Sequential-Process model on execution level. Power analysis using simulation techniques on gate-level provides input for the algorithmic selection during run-time of the system, leading to a closed-loop design flow. Additionally, the signal flow approach enables power management by varying the signal flow and data processing rates depending on actual energy consumption, estimated energy deposit, and required Quality-of-Service.
[c11.5]
Stefan Bosse, Hardware-Software-Co-Design of Parallel and Distributed Systems Using a unique Behavioural Programming and Multi-Process Model with High-Level Synthesis, Proceedings of the SPIE Microtechnologies 2011 Conference, 18.4.-20.4.2011, Prague, Session EMT 102 VLSI Circuits and Systems, 2011,
DOI:10.1117/12.888122. Paper PDFPresentation PDF
A new design methodology for parallel and distributed embedded systems is presented using the behavioural hardware compiler ConPro providing an imperative programming model based on concurrently communicating sequential processes (CSP) with an extensive set of inter-process-communication primitives and guarded atomic actions. The programming language and the compiler-based synthesis process enables the design of constrained power- and resource-aware embedded systems with pure Register-Transfer-Logic (RTL) efficiently mapped to FPGA and ASIC technologies. Concurrency is modelled explicitly on control- and datapath level. Additionally, concurrency on data-path level can be automatically explored and optimized by different schedulers. The CSP programming model can be synthesized to hardware (SoC) and software (C,ML) models and targets. A common source for both hardware and software implementation with identical functional behaviour is used. Processes and objects of the entire design can be distributed on different hardware and software platforms, for example, several FPGA components and software executed on several microprocessors, providing a parallel and distributed system. Inter-system-, inter-process-, and object communication is automatically implemented with serial links, not visible on programming level. The presented design methodology has the benefit of high modularity, freedom of choice of target technologies, and system architecture. Algorithms can be well matched to and distributed on different suitable execution platforms and implementation technologies, using a unique programming model, providing a balance of concurrency and resource complexity.
[c11.6]
Stefan Bosse, Thomas Behrmann, Frank Kirchner, Smart Energy Management and Low-Power Design of Embedded Systems on Algorithmic Level for Self-Powered Sensorial Materials and Robotics, Proceedings of the Smart Systems Integration Conference 2011, Session 4, Dresden, 22 – 23 Mar. 2011, VDE VERLAG GMBH, 2011,
ISBN: 978-3-8007-3324-8. Paper PDFPoster PDF
A new design methodology for low-power embedded systems is presented which is based on advanced algorithms from computer science. System-On- Chip architecture, power analysis, and advanced system modelling methodologies are used.
[c11.7]
F. Pantke, S. Bosse, D. Lehmhus, M. Lawo, An Artificial Intelligence Approach Towards Sensorial Materials, Proceedings of The Third International Conference on Future Computational Technologies and Applications (Future Computing 2011),
Sept. 25-30, 2011. Paper PDF
Sensorization aims at equipping technical structures with an analog of a nervous system by providing a network of sensors and communication facilities that link them. The objective is that, instead of having been designed to loads and tested to conditions, a structure can experience and report design constraint violations by means of realtime self-monitoring. Specialized electronic components and computational algorithms are needed to derive meaning from the combined signals. For this task, artificial intelligence approaches constantly gain importance; the more so as the trend of ever increasing sensor network size and density suggests that sensor and structure may soon become one, forming a sensorial material. Current simulation techniques capture many aspects of sensor networks and structures. For decision making and communication, the Intelligent Agent paradigm is an accepted approach, as is finite element analysis for structural behavior. To gain knowledge how sensorial structures can most effectively be built, an artificial intelligence based process for the construction of such structures was developed that uses machine learning methods for fast load inference. It is presented in this paper, along with evaluation results obtained in experiments using a finite element model of a strain gauge equipped plate which demonstrate the general practicability.
Interdisciplinary Science & Education Laboratory