i-Sense Results

Which are the main research challenges that i-Sense project is facing?

Recent technological advances in computing hardware, sensors/actuators, communications and real-time software have facilitated the development of networked intelligent systems with the capability to generate and deliver huge amount of data in real time. The “Internet of Things” and “Big Data” are key upcoming research challenges, which also go in this direction. As the volume of the acquired data rapidly increases, one of the most important research challenges is the design of methodologies for extracting the relevant information, make real-time decisions and bring back to the remote distributed system.

However, when data are faulty, missing or inconsistent, decisions are, in the best case only partly appropriate, more often causing the system to fail due to the undertaken wrong actions. The i-Sense project acts at this level and provides a systematic approach for cognitive fault diagnosis and fault tolerant control in uncertain distributed environments. These methodologies have a significant impact on the availability, reliability and dependability of engineering systems, and provide a new paradigm for fault tolerance in incompletely specified environments.


Two years after the beginning of the project, which are the main results obtained?

We obtained significant results, well published in international top venues, facing the problem at different levels, depending on the a priori information available regarding the nature of the system and the type of uncertainty affecting it. More in detail, we designed novel Fault Detection and Isolation (FDI) algorithms operating both at the sensor level and at the distributed system level. Methods identify potential changes in the datastream by inspecting the evolution of some signal features within a non-parametric approach and make a decision when enough confidence is grasped. Distributed information at the network layer is then exploited by the cognitive level, which aggregates and combines results provided by the sensor level. Here, exploitation of spatial and time dependencies provide important components of the developed methodology. Moreover, the i-Sense team has also been working in developing Fault Tolerant Control (FTC) strategies using virtual sensors and actuators under the Linear Parameter Varying (LPV) framework. This framework allows us to develop adaptive and robust fault tolerant control strategies able to detect and cope with the fault in a quite natural way.

Within i-Sense we developed also an innovative cognitive fault diagnosis framework that investigates fault diagnosis both in the model and parameter spaces of linear time invariant approximating models instead of operating in the signal space. Changes are automatically detected, isolated and classified when faulty instances are known while the faults are differentiated from non-stationarity phenomena or time variance of the environment. In order to emulate faults the team designed and developed the i-Sense fault emulator platform, an electronic board able to inject silently a large class of faults into the mounted sensors. As such, the outcome datastreams can be either faulty or error-free depending on the chosen set-up.


What are the next steps to be performed during the next year?

The main objective of the project is to build a fault diagnosis system prototype based on the i-Sense Platform for the intelligent building application, the water distribution network of the city of Barcelona and the environmental monitoring application. The various approaches and the developed methodologies will be further advanced, completed and integrated in the applications by exploiting the available information. Applications introduce a significant variety of diversified scenarios allowing us to test the different solutions. For instance, we have applications, such as the water distribution network of Barcelona, where the process is mostly known and described by physical equations. At the other extreme we envisage data coming from environmental monitoring, where nothing is known about the process generating the data and the unique source of information is the acquired datastreams.


i-Sense Project Coordinator

Prof. Marios Polycarpou