Mission Abort Policy Considering Imperfect Alarms and Two Abort Options
针对警报系统存在漏报和误报的情况，提出部分可观测马尔可夫决策过程框架，动态更新系统缺陷概率，并证明最优中止策略遵循控制限策略，以最小化任务和系统故障的期望总成本。
ABSTRACT A mission may be aborted during operation to increase the likelihood of system survival when an alarm system indicates that the system has entered a defect state. However, due to measurement and environmental noise, missed and false alarms frequently occur during online condition monitoring, complicating the decision‐making process for mission abort. Additionally, once a mission is aborted, various rescue procedures can be initiated. This study addresses these challenges by proposing a partially observable Markov decision process (POMDP) framework aimed at minimizing the expected total cost associated with mission and system failures. Within this framework, the probability that the system is in a defect state is dynamically updated using real‐time signals from the alarm system. We demonstrate that the optimal mission abort policy follows a control limit policy, where a mission should be aborted when the defect probability exceeds a certain threshold during each period. To illustrate the effectiveness of the proposed policy, we present a case study on an unmanned aerial vehicle (UAV).