Invited Talk by Fumbeya Marungo, Johns Hopkins University

• Title: A Residual Approach to Multi-agent Fault Detection and Isolation Using Asynchronous Belief Propagation for Fault-free Modelling of a Naval Vessel's Chilled Water System
• Speaker: Fumbeya Marungo, Johns Hopkins University, Baltimore, MD, USA
• Date: Friday, August 21, 2015
• Time: 10:30
• Location: 03.13.010

Abstract

The requirements of fault diagnosis and isolation (FDI) for fully or semi-autonomous systems that must operate for long durations in adverse conditions outstrips the capacity of centralized monolithic approaches. Multi-agent FDI (MAFDI) is a means of addressing the need by placing the workload across many components. A MAFDI approach that meets the needs of this scale requires extending the present state-of-the-art. The approach must be: (1) deployable on low powered computationally constrained devices; (2) use only local agent neighbor to neighbor communication; (3) employ heterogeneous agents to represent the system's different components; (4) and be capable of isolating multiple faults. Additionally, MAFDI design must account for the lack of global data availability. To address these issues, this work proposes a residual MAFDI methodology. The approach uses a fault-free model of a system’s configuration. A controller combines the configuration estimate with its local knowledge of previous commands and control objectives to create a fault assessment. The fault-free model is implemented using strictly local belief propagation messages over a factor graph. The model is tested on a simulation of a testbed for a naval vessel’s chilled water fluid system.