The research activity addresses the design of methodologies, techniques and solutions for adaptive and intelligent information processing systems able to interact proactively with the environment and react and adapt to evolving time-variant situations. In detail, the research focuses on:

  • the study and design of intelligent embedded and cyber-physical systems, i.e., embedded and cyber-physical systems inheriting intelligent mechanisms proper of human cognition;
  • the investigation and design of adaptive computational-intelligence techniques;
  • the deployment of credible networked intelligent embedded systems able to operate in harsh environments.

These research activities are strictly related and integrated and can be intended within a research cycle. The research on adaptive computational-intelligence techniques provides novel intelligent mechanisms making embedded and cyber-physical systems adaptive and able to interact with the environment. The research on intelligent embedded and cyber-physical systems provides novel strategies and solutions to design networked intelligent embedded and cyber-physical systems. Finally, the “from the lab to the real world” research activity on the deployment of networked intelligent embedded and cyber-physical systems in harsh environmental conditions identifies novel research challenges to be addressed both at the theory and embedded system level.


Intelligent Embedded and Cyber-physical Systems

The research aims at addressing methodological and application-related aspects of intelligent embedded and cyber-physical systems, i.e., embedded and cyber-physical systems endowed with computational intelligence and cognition abilities allowing them to deal with a pervasive uncertainty and learn from acquired data. Wireless Sensor Networks, complex sensor/actuator networks and hybrid systems are examples of the classes of embedded and cyber-physical systems considered in the research. More specifically, the research activity focuses on the design of methodologies and algorithms granting

  • the processing system to implement adaptation and robustness mechanisms. For example, fault diagnosis is a relevant aspect of this research activity enabling mechanisms to face the insurgence of hardware and software faults. In particular, cognitive fault diagnosis systems have been designed to carry out the identification and isolation tasks by exploiting temporal and spatial dependencies in acquired datastreams;
  • the units to expose intelligent distributed mechanisms for identifying variations in the environment and in the interaction of the technology with the environment. Hierarchical and distributed change detection tests have been designed to detect changes directly from incoming data and adapt the system to the new working conditions;
  • the ability to execute smart energy management mechanisms for intelligent embedded systems. Designed solutions aim at maintaining the Quality-of-Service and prolong the life-time of the systems by means of energy-aware local routing protocols and adaptive sensor sampling as well as supporting a remote reconfiguration of the network units within a distributed framework.

Particular attention has also been devoted to the design and the development of networked intelligent embedded and cyber-physical systems with sensor capability by addressing applications characterized by a large impact on the Society and able to operate in harsh environments . This “from the Lab to the Real world” research activity is described in the Section “Deployment of credible networked intelligent embedded and cyber-physical systems”.


Adaptive Computational-Intelligent Techniques

The research addresses theoretical, implementation and application-related aspects of machine learning/ computational intelligence-based systems, with a specific focus on adaptation mechanisms allowing the system/application to track evolving environments. In order to achieve this goal we need to weaken the stationarity/time invariant hypothesis and develop dynamic management mechanisms based on the available knowledge to allow the adaptive systems to react and track process changes. In turn, this research has requested

  • the definition of general purpose solutions able to assess the stationary of a data-generating process and estimate the temporal instant the process generating the data deviates from its nominal state;
  • the introduction of the novel concept of “Just-in-time” framework that allows the systems to reconfigure/update in a just-in-time fashion, i.e., exactly when needed, the knowledge base of the systems. The approach represents a form of active learning where a triggering mechanism activates the response to changes in the process under monitoring (differently from other passive approaches present in the literature that force the continuous update of the application);
  • the design of adaptive knowledge-based management systems able to track the process change by activating, whenever appropriate, previously acquired knowledge so as to take advantage of recurrent states.

The above research activity provides the theoretical foundation needed by intelligent embedded systems described in the previous section, since algorithms and techniques designed to support intelligent embedded systems are here investigated and perfected. Relevant examples of this combined research activity are the development of cognitive fault detection/isolation/identification techniques for networked embedded systems. These computational intelligence-based techniques allow the system for detecting, isolating and identifying faults affecting units of the network (to provide timely alarms and actuate accommodation procedures) as well as changes in the physical process under monitoring (which require a re-training of the computational intelligence-based system).

In addition, the research addressed the study of a methodology for the high-level synthesis of adaptive information processing systems endowed with computational intelligent mechanisms. The novelties of these adaptive information processing systems reside in the ability to work in time-varying environments (hence weakening the stationary hypothesis that is generally assumed in the related literature), the capability to adapt the feature set that is extracted from the data-generating process over time to improve/maintain the Quality of Service of the envisaged application and the ability to create and maintain evolving knowledge in adaptive processing systems. A relevant application scenario for this research activity is the analysis and quality assurance in industrial processes.


Deployment of credible networked intelligent embedded and cyber-physical systems: “From the Lab to the Real world”

The outcomes of the theoretical research activities on intelligent embedded and cyber-physical systems and adaptive computational-intelligence techniques provide novel tools and solutions for the “from the lab to the real world” applied research focusing on the deployment of credible networked embedded and cyber-physical systems.

A credible deployment in a harsh environment sets the basics for identifying the real needs that an intelligent embedded system must expose. In this direction, we designed and deployed a set of systems where some basic intelligent mechanisms show to be fundamental to grant a Quality of Service in harsh environmental conditions. Remarkably, this “from the Lab to the Real world” activity is fundamental since it sets the challenges both at the technological and methodological level that need to be addressed by the research community at the basic research level.

A robust, adaptive, solar powered WSN-based framework for marine environment monitoring has been deployed in Queensland, AUS (Nov. 2007) for monitoring the underwater light and temperature. All aspects of the environmental monitoring system such as sensing activity, local transmission (from sensor nodes to gateways), remote transmission (from the gateways to the control center), data storage and visualization have been addressed to guarantee robustness and adaptability to network changes.

A real-time monitoring system for rock collapse forecasting that exploits MEMS accelerometers and geophones (in addition to traditional sensors such as strain-gauges, tiltmeters, flowmeters) for non-invasive detection of micro-acoustic bursts associated with the formation and the evolution of cracks within the rocks has been designed and developed. This system has been deployed in several critical areas of the Italian-Swiss Alps (and systems are still active): S. Martino Mountain (April 2010-present); Torrioni di Rialba (July 2010-present); Val Canaria, Ticino, Switzerland (August 2011-present); Gallivaggio (July 2012-present). The proposed monitoring system relies on a network of intelligent embedded systems characterized by high-frequency sampling hybrid wireless-wired architectures tailored to detect and –hopefully- localize micro-acoustic emissions in the rock face, yet maintaining an high energy-efficiency by means of effective energy management policies and sophisticated adaptive energy harvesting mechanisms.

Finally, an intelligent landslide monitoring system based on a wireless network of intelligent embedded systems has been designed and developed. This system has been considered to investigate critical areas of Italian Alps: Torrioni di Rialba (July 2011-present); Premana (August 2012-). Aspects related to intelligent power management, remote units reconfigurability, remote code upload and effective data storage, aggregation and visualization have been considered and addressed in this research.

The design, implementation and deployment of sophisticated networked intelligent embedded systems required a high degree of cross-disciplinarity with continuous discussions with biologists, geologists and geophysicists. The interaction with phenomenon experts represented a challenging activity that, despite the communication difficulties, lead to effective monitoring systems.