A Markov Chain State Transition Approach to Establishing Critical Phases for AUV Reliability

Abstract

Abstract-The deployment of complex autonomous underwater platforms for marine science comprises a series of sequential steps. Each step is critical to the success of the mission. In this paper we present a state transition approach, in the form of a Markov chain, which models the sequence of steps from pre-launch to operation to recovery. The aim is to identify the states and state transitions that present higher risk to the vehicle and hence to the mission, based on evidence and judgment.Developing a Markov chain consists of two separate tasks. The first defines the structure that encodes the sequence of events. The second task assigns probabilities to each possible transition. Our model comprises eleven discrete states, and includes distance-dependent underway survival statistics. The integration of the Markov model with underway survival statistics allows us to quantify the likelihood of success during each state and transition and consequently the likelihood of achieving the desired mission goals. To illustrate this generic process, the fault history of the Autosub3 autonomous underwater vehicle provides the information for different phases of operation. The method proposed here adds more detail to previous analyses; faults are discriminated according to the phase of the mission in which they took place.

Risk modeling of autonomous underwater vehicle operation focusing on the human operator

“…The reliability was derived by expert judgement based on the fault logs of the vehicle. Brito and Griffiths (2011) used a Markov model to assess the reliability of the AU-TOSUB 3. The aim was to identify the probability of critical states that might lead to loss of the vehicle.…”

“…Manley (2007) states that mission files often contain errors due to typographic, sign or geographic datum errors, and wrong use of the mission programming software, which operators use for mission planning and preparation. Brito and Griffiths (2011) present, along with their Markov model for assessment of critical states of AUV operation, some incident data for the AU-TOSUB 3 AUV; nine out of 28 failed or preliminary aborted missions can be attributed to human errors or influences. If the operator introduces errors in the mission plan, the AUV might not detect these errors and follow a path, which is potentially dangerous for the mission or AUV, e.g., passing fishing vessels (Kirkwood, 2009) or heading towards shallow water due to a wrongly implemented waypoint or drift.…”

A Behavioral Probabilistic Risk Assessment Framework for Managing Autonomous Underwater Vehicle Deployments

“…AUVs are being increasingly used for polar missions, including for commercial (Kleiner et al 2011) and geopolitical purposes. The science community has recognized that there is a substantial risk of not completing missions and a risk of losing the vehicle when operating in polar seas (Griffiths et al 2003), and it has developed mathematical methodologies to assess and manage those risks (Brito et al 2010). However, when AUVs are to be used for data gathering in support of geopolitical or commercial missions, the risks of not completing missions or campaigns are likely to have greater impact and repercussions than for science missions.…”

“…Individual expert judgments can be aggregated mathematically or behaviorally to reach this group judgment. Previously, expert judgments have been mathematically aggregated using the linear opinion pool, where experts have been kept separate during the elicitation (Clemen and Winkler 1999;Griffiths et al 2009;Brito et al 2010). There are different schools of thought as to what is the best aggregation method; while some postulate the use of mathematical methods (e.g., Mosleh et al 1987) others postulate the use of behavioral methods (e.g., Phillips 1999).…”

“…In previous AUV risk assessments, for particular faults or incidents, experts provided judgments of a probability of fault leading to loss that spanned three orders of magnitude (Brito et al 2010). Despite capturing each expert's estimate of uncertainty using quartiles, the large spread for some assessments remained a concern.…”