Reliability Analysis and Verification for Pervasive Context Aware Systems

Members & Funding

The Reliability project is funded by CNRS under the PEPS INS2I/INSMI 2015 call for proposals. The thematic of the call is Fundamentals and Applications of Data Science (FaSciDo). This is a colaborative project between IPAL, LIRMM in Montpellier and NUS School of Computing in Singapore.


Several studies have been done on developing context aware systems during this last decade, but limited impact has been noticed due to the lake of reliability of such complex system. In this project we focus on a scientific approach for reliability measurement and prediction in pervasive context aware systems. These systems typically combine data intensiveness and near real-time constraints. The example of a smart homecare system is used to demonstrate the viability of proposed models and techniques. The work primarily provides development-level information such as guaranteed reliability across several design choices in order to help developers build the most reliable system taking into account predefined constraints. The partners, who have expertise in reasoning engines (IPAL and LIRMM) and formal methods for verification (NUS), may also explore online system reliability estimation to provide post-deployment reliability feedback to both developers and a semantically driven reasoning engine in order to allow a manual and/or automatic improvement of the rule system.

Research Focus

Quantitative analysis of probabilistic systems gains great importance, especially for complex systems with non-deterministic behaviours. Ambient Assisted Living (AAL) systems are typically user centred so their behaviours are non-deterministic due to unpredictable user activities. Thus, we propose to use Markov Decision Process (MDP) as the modelling formalism for its support of both non-deterministic as well as probabilistic choices. Three general but highly important reliability issues are investigated. First, “what is the overall system reliability if reliability of all its components is known, considering all possible user behaviours, and unreliable factors?”. This is the problem of reliability prediction. This question is to be answered necessarily before system deployment since end users would prefer to know how reliable the system is. Secondly, “what is the reliability required on certain components if there is a requirement on overall system reliability?”. This is the problem of reliability distribution. Addressing this issue is useful because we can have specific quantitative requirements on selecting software and hardware components, whose quality are often cost sensitive. Last but not the least, to find the most critical components affecting the system reliability via sensitivity analysis is essential to improve the overall system reliability effectively with limited resource [4, 5]. For example, if a system is shown to be not reliable enough based on current component’s reliability, it is desirable to prioritise the components such that reliability improvement of a higher priority component would result in more improvement on overall system reliability. 

Research Team