» version anglaise » RESEARCH » Main themes » Hybrid systems and sensing systems

Hybrid systems and sensing systems

In these areas, the main challenges come from the connexion and the interactions of numerous hybrid and sensing systems via communication networks. This leads to levels of complexity that must be addressed with new techniques and methods.

Hybrid Systems

Complex hybrid systems such as those considered here can be described as involving three abstraction layers :

  • a lower level corresponding to individual (analog) physical devices to be controlled,
  • an intermediate level corresponding to interactions between these devices (i.e., between electro-technical devices for the purpose of balancing produced and consumed power)
  • a higher level corresponding to networks of interactions (for regulating traffic flow, for ensuring overall power network security, etc.).

Each layer has its own "life" (size and time scale of models), but is
strongly connected to the others. For instance, small variations at a
particular point in the lower layer may generate chaos or blackout at
the higher level. The higher level needs to take into account computers
and devices (network) interconnected by communications channels.
Research on complex and networked hybrid systems is of major strategic
relevance for industry and society and calls for new methodological and
engineering approaches that ensure efficient, predictable, safe and
secure behaviour and increase robustness and performance. Domains of
application include manufacturing and process plants and large-scale
infrastructures such as those involved in distributed energy production
or distribution, airports or seaports, etc. New technologies such as
wireless sensing and actuation should be taken advantage of.

What follows is a non-limitative list of open questions in this context :

  • How to coordinate regenerative breaking, conventional breaking, battery voltage, and engine torque in a vehicle, possibly taking into account a stochastic model of the behaviour of the driver and information about incoming slopes as provided by navigation systems ?
  • How to integrate control and prediction in freeway traffic networks ? How to optimize traffic management with cooperative vehicle-infrastructure systems ?
  • How to increase the efficiency of power generation and to lower energy costs for households via distributed control of power networks ?
  • How to integrate new (stable and robust) technologies to the distributed control of large-scale electric power generation, transmission and distribution networks ?
  • How to optimise product yield and how to minimise batch time for batch reactor
  • operation under constraints on the utilization of infrastructures ?
  • How to equip populations of biological cells with distributed regulators mapped into biochemical networks for achieving desired collective behaviours such as the production of a protein in a synchronized fashion ?
  • How to support movements of paralyzed people by functional electrical stimulation of muscles and robotic assistance (related to Section 3.1) ?
  • How to design deep brain stimulation to treat Parkinson’s disease based on microscopic and macroscopic models of neuronal activities, in order to optimize the current clinical techniques ?

Key challenges include :

  • developing generic modelling and the design of model-based control methods, under the multiple constraints inherent to networked and/or embedded systems.
  • implementing languages and scalable algorithms for the control of evolvable, distributed and adaptable systems,
  • making performance robust to the effects of temporal spatial and parametric uncertainty, transmission delays, disturbances, limited communication bandwidth, actuation constraints and node availability.

Addressing these questions and challenges requires the merging of
competences to be found in computer, control, and communications
sciences and engineering together with biology, chemistry, physics and
social sciences. Fundamental and applied interactions must be
stimulated, coordinated, and organised.

Sensing Systems

The traditional approach in instrumentation follows a linear scheme
going from some physical quantity to be measured to signal via
transduction, and then from this signal to a numerical estimate of the
quantity of interest via signal processing. A multitude of new schemes
are now possible, thanks to the available computing power and the
massive use of modelling, simulation, prior knowledge, and data bases
and networks. The traditional linear scheme is therefore no longer the
only option available and new research avenues open. The key concept of
information, and information theory, should be at the core of sensing
systems, which are at the boundary between the physical world (where the
sciences are physics, chemistry, biology, etc. and the technology is
that of sensors and electronics) and the digital world (where the
sciences are signal processing, statistics, numerical analysis, machine
learning, etc. and the technology is mainly software).

The scientific and technological aspects on the physical and digital
sides are more and more interconnected, mixing fundamentals in physics,
chemistry, biology, sensor technologies, and mathematics and software
technologies. This leads to the multidisciplinary concept of sensing
system, where crossover and synergy between instrumentation, information
and computer sciences and technologies should be the source of new
functionalities and performances. This also sets new challenges out.

A generic definition of the scientific program would be :

  • Any approach based on information processing that transforms data produced by sensors (actual or simulated) into estimates of quantities of interest or into pertinent, higher-level information.
  • Any approach where sensor design is governed by information-processing issues.

A non-limitative list of topics of interest to Digiteo partners is :

  • Efficient algorithms to process voluminous and/or complex data and extract information of interest.
  • Inverse problems, coupling physical sensors, physical models and simulations.
  • Software sensors to estimate quantities that cannot be measured directly from available data and prior knowledge.
  • Experiment design by statistical approaches to minimise uncertainty.
  • Control theory for linear and non-linear parameter and state estimation (see the Decision and Control Systems section).
  • Coupling data bases and prior knowledge with physical measurements.
  • Data-mining and machine-learning techniques to extract high-level information from low-level and/or heterogeneous data (cf. Programming, Software Engineering and Information Systems).
  • Sensor networks.

Digiteo partners offer a global expertise in hybrid and sensing
systems, from physics, biology and chemistry to knowledge management and
information processing in complex and heterogeneous contexts. Digiteo
works at developing some synergy and cross-fertilisation between the
relevant groups.