Overview of my PhD work and experience (cont)

In a distributed system, components are interconnected and interact to produce an emergent complex behaviour. As the number of components increases, the complexity of the system also increases. We perform model-based diagnosis by matching scenarios in the model that are consistent with the observations. The main challenge is that the representation of these scenarios is usually exponential in the number of components. My supervisor and I proposed an approach to handle this problem by avoiding computing diagnosis scenarios on the whole system. Instead, we perform local computation that only involves subsets of the components and ensure consistency among those local computations. We first transform the topological graph of the system into a junction tree. Each cluster of that tree is a subsystem. We apply an algorithm that ensures consistency within the subsystems, then propagates the information among the subsystems in two passes to ensure consistency of the whole system. This work has been presented at the European Conference on Artificial Intelligence in 2008. It started during my Masters year and beginning of my PhD. It is the result of numerous discussions between my supervisor Dr. Alban Grastien and myself.

I started mingling in student reading groups and met students from other disciplines at the beginning of my PhD program. Mr. Lachlan Blackhall and I learnt about each other’s work and saw the parallel between what we are doing. He is doing in PhD in the control of complex networks. We thought of symbiosing our approaches and over numerous brainstorming sessions and refining our concepts we came up with a hybrid system framework to integrate a discrete event system approach with a continuous system modelling approach. The idea is that as the mode changes (e.g. generator goes from spinning reserve to generating power) the set of dynamics modelling continuous operation also changes. Hence the different modes of the system can be considered discrete and be abstracted from the continuous dynamics. We have presented our hybrid framework at DX08, Safe Process 09 and DX09 (see publications section for references).