دانلود Fault Tolerance Considerations for Long Endurance AUVs

Abstract

Autonomous Underwater Vehicles (AUV) work in a harsh and uncertain environment which imposes challenges on their energy, navigation, and communications. Given the environment, there is little bandwidth to communicate solutions to an on-board fault or failure.  The AUV application discussed is for Naval Mine Countermeasures (NMCM) survey and minehunting missions.  For such missions,  operational availability, reliability demands, and system safety are  of  high  importance. To  address  this,  an  on-board  Fault  Detection,  Isolation  and  Recovery  (FDIR)  system  is  provided by  the  manufacturer  for  basic  faults  like  slow  leaks,  over-depth,  and  time-outs  due  to  unreceived  operator  commands .With  that,  most  AUVs  can  implement  a  scripted  mission  but  are  generally  unable  to  recover  from  more  complex  failures  like low energy, or reduced functionality in hydroplanes. These two cases are presented here as implemented examples.  The  examples  show  that  an  autonomous  on-board  recovery system could be devised and implemented for timely recovery from  these  types  of  failures.    With  such  measures,  the  AUV can be adaptive and as fault tolerant as possible to unexpected changes  in  itself,  the  environment  and  the  mission.    The recovery  employed  machine  learning  to  gain  insight  into  the best  solution  for  a  specific  failure  and  the  reason  for  failure  from observations on faults/failures.  Further, dynamic Bayesian  networks  (DBN)  are  proposed  as  a  novel  FDIR approach towards AUV reliability for long endurance NMCM missions.    DBN  are  suited  to  address  partial  observability,  uncertainties  inherent  in  the AUV subsystems’  evolution,  and  the subsystems’ interaction with the harsh and uncertain environment.    This  makes  advanced  reactive  and  preventive  fault/failure recovery possible.