Eliciting expert judgment on the probability of Loss of an AUV operating in four environments

Abstract

Assessing the risk in the operation of complex systems is based on the subjective judgment and knowledge of human experts as much as on hard statistics.The Autosub3 is an Autonomous Underwater Vehicle (AUV) designed and operated by the National Oceanography Centre, Southampton.The AUV is used to support studies carried out by scientists supported by the Natural Environment Research Council.Autosub3 will undertake campaigns in icecovered seas, where we consider the risk of loss to be significant.Our aim here is to quantify that risk.This study reports on an expert judgment elicitation project, where a group of ten world experts in AUV operation were asked to assign a probability of AUV loss given the emergence of a particular fault, a total of sixty three faults were considered.Reasons given by experts for their assessments are analyzed, and the differences of opinion noted.The annexes contain the full text of expert judgments.After discussing methods of aggregation and analysis, the aggregated risk estimates obtained from the expert judgments were used to estimate the AUV survival in four different environments (open water, coastal, sea ice and ice shelf).The Autosub3 survival function was used to estimate the risk of a campaign to the Antarctic, planned for January-March 2009.Below is the probability of loss for four mission sets using the risk model derived from the independent expert judgments.On the right is the risk acceptable to the responsible owner of Autosub3.In summary, the reasoning for accepting risk was stated as: Minimum set, no sea ice in front of glacier P(loss) = 10% Accept 10% Minimum set + 30km of sea ice in front of glacier P(loss) = 17% Accept 17% Desired set with no sea ice in front of glacier P(loss) = 20% Accept 20% Desired set + 30km of sea ice in front of glacier P(loss) = 23% Accept 23%The estimates presented above are based on the motor fault on D306 being retired, as demonstrated by the Norway trials, but with the new high impact fault on those trials included.These results are based on a risk mitigation strategy that requires the vehicle to be operated for an optimal monitoring distance in a area where recovery could be achieved prior to each and every mission.The optimal monitoring distance for scenario 1 is 28km, scenario 2 is 33km, scenario 3 is 43km and finally scenario 4 is 48km.The results above are also based on modeling the increased reliability of the vehicle after each successful mission.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.