Automatic Fault Diagnosis for Autosub 6000

Richard Dearden; University of Birmingham

27th July, 2010 at 3.00 pm - EM2.33

Internal HWU seminar

In this talk I report on progress made in the NERC-funded project"Automatic Fault Detection for Autosub 6000" (AFDA). The project aims toimprove the reliability of the Autosub autonomous underwater vehicles byimplementing a fault detection system on board. I will describe theconsistency-based diagnosis system we are using, Livingstone 2, anddescribe how it is used to model aspects of the vehicle. I will thendiscuss the considerable challenges that arise from the fact that like mostAUVs, Autosub has very short turnaround times between missions, and may besignificantly reconfigured from one mission to the next. To overcome this,we have developed a method to automatically generate a diagnosis model ofthe program the vehicle is executing that can be integrated with the modelsof the hardware. I will show that this combination of hardware and softwarediagnosis models allows us to detect faults that were not detectable byother means. I will also briefly discuss our approach to diagnosing vehiclesubsystems that are not amenable to consistency-based approaches.