Optimum Life-cycle Maintenance Policy Of Fatigue Structure Based On Reinforcement Learning

Life-cycle maintenance of ship is essential for ship structural safety during the service life.Preventive maintenance has been attracted growing interest in this area.Time-based maintenance(TBM)is a commonly used policy for preventive maintenance.TBM is conducted at regular intervals,regardless of the actual state of the structure.It is difficult to accurately determine the maintenance time interval of TBM,resulting in insufficient or excessive maintenance.With the rapid development of sensing technology,monitoring and diagnosis technology,the condition-based maintenance(CBM)can consider the actual operating state of the structure when performing maintenance.The advantage of CBM is that it is highly targeted,can make full use of the service life of the structure,and can also effectively prevent the occurrence of failures.CBM is an important measure to ensure that the ship structure meets the reliability requirements during the life-cycle of ship structure.This thesis proposes a novel method for CBM policy making of fatigue-sensitive ship structures.Reinforcement learning(RL)is a promising method of machine learning.As a decision-making optimization tool,solving the optimal CBM policy is a rational application scenario of RL.The purpose of introducing RL is to train an agent to give the optimal type of maintenance measures under different conditions.As an optimization problem of RL,CBM policy has been studied by researchers.However,in the traditional reinforcement learning-based methods,the modeling of the deterioration process normally relies on assumptions,thus limits the effective application and of RL in CBM policy making.This thesis investigates the various maintenance policy of fatigue-sensitive ship structure.First,the optimal TBM policy is solved based on the fatigue mechanism,and the influence of different inspection times and maintenance effects on the optimal policy is demonstrated.Then,a ship deterioration process is established through experience.The optimal CBM policy of the ship in multiple deterioration states is derived by RL.Finally,a reinforcement learning framework based on the fatigue damage mechanism is proposed to optimize the CBM policy.The obtained optimal CBM policy is compared with several different policy to show advantages and drawbacks of the method.The CBM policy obtained by the proposed framework has advantages in economy and safety.Proposed framework promotes the development of reinforcement learning applying in ship structures maintenance.