【作者】 陈海波；

【导师】 付旭云；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2019， 硕士

【摘要】 民航发动机是大型复杂装备,具有可靠性要求高、使用寿命长、维修成本高、维修复杂等特点。合理的安排发动机全寿命历次维修时机和历次维修策略,不仅能够有效提高发动机的运行可靠性,同时能够大大降低发动机的维修成本。因此,民航发动机全寿命维修决策方法研究具有重要的理论意义和经济价值。目前民航发动机维修决策方面的研究大多数是将发动机的维修时机、单元体维修等级、寿命件更换进行拆分单独研究,而实际工程中三者是相互影响的整体。针对先前研究的不足,本文综合考虑民航发动机维修过程的维修时机、单元体维修等级、寿命件更换,建立了三种不同民航发动机维修决策模型,并采用不同的方法进行求解。民航发动机全寿命维修决策包括确定历次维修时机、历次维修时单元体维修策略和寿命件更换策略。直接将这三者作为决策变量建立民航发动机维修决策模型,存在解空间规模非常大、难以求解等问题。本文对民航发动机维修决策问题解耦。以民航发动机维修时机作为决策变量,以全寿命总维修成本最小为优化目标,建立一种基于粒子群优化的民航发动机维修决策模型。在确定维修时机条件下,提出一种寿命件更换规则求解寿命件最优更换策略及其成本,采用动态规划算法求解单元体最优维修策略及其成本。最后采用粒子群优化算法对本文建立的民航发动机维修决策模型进行求解。采用数值实验和航空公司实际数据对本文提出的方法进行验证和评估。结果表明,提出的方法适用于民航发动机维修决策问题。针对大规模机队的维修决策问题,粒子群优化算法存在着求解速度慢、效率低、求解结果不够好等缺点。增强学习能够学习到的知识存储,具有求解速度快的特点,因此将增强学习中的Q学习算法应用到民航发动机维修决策问题。首先以发动机的每一飞行循环进行一次维修决策,将发动机全寿命维修过程看作马尔科夫决策过程,建立基于Q学习的民航发动机全寿命维修决策模型。分别对模型中的状态、动作、回报和Q表进行设计。然后训练Q学习算法对模型进行求解,最后采用数值实验对提出的方法进行评估验证。结果表明,增强学习应用到民航发动机维修决策问题具有可行性,但是,基于Q学习的民航发动机维修决策方法只适用于寿命件和单元体数量较少的民航发动机维修决策问题。针对当寿命件和单元体数量较多时,状态空间增大,Q学习算法出现“维度灾难”问题,将DQN算法应用到民航发动机维修决策问题。结合民航发动机维修时机确定规则、寿命件更换和单元体维修规则,对状态空间和动作空间进行缩减,建立基于DQN的民航发动机维修决策模型。为了加速DQN算法的收敛,提出一种DQN预训练方法。基于规则产生一定数量的较好数据对DQN网络进行预训练。最后采用数值实验和民航公司的实际数据对提出的方法进行验证。结果表明,面对大规模的民航发动机维修决策问题,提出的方法能够快速的求解出较好的民航发动机全寿命维修决策。最后基于本文提出的民航发动机维修决策方法,结合民航公司对发动机维修决策的实际需求,设计并开发了民航发动机维修决策原型系统,为民航公司发动机维修决策工作提供技术支持。更多还原

【Abstract】 Civil aero-engine is a complex equipment,with high reliability requirements,long life,high maintenance costs,maintenance complex and other characteristics.Reasonable arrangement of the maintenance occasion and the mainenance strategy of civil aero-engine in life cycle,not only can effectively improve the reliability of the aero-engine,but also can greatly reduce the maintenance costs of the aero-engine.Therefore,the research on the maintenance decision method of civil aero-engine in its life cycle has important theoretical significance and economic value.At present,most of the research on the maintenance decision of civil aero-engine is to separate the maintenace occasion,the maintenance level of aero-engine module and the life limit parts replacement,while it influence each other in the actual engineering.In view of the shortcomings of previous research,this paper comprehensively considers the maintenance occasion,the maintenance level of aero-engine module and the life limit parts replacement strategy,and establishes three different civil aero-engine maintenance decision models,then uses different methods to solve.The civil aero-engine life cycle maintenance decision includes maintenance occasion decision,the life limit parts replacement strategy and the maintenance strategy of aero-engine modules.Aiming at the problem that the solution space of aero-engine maintenance decision model which directly established based on maintenace occasion,aero-engine module maintenance level and life limit parts replacement strategy is too large to get a good solution.This paper decoupling civil aero-engine maintenance decision problems.A civil aero-engine maintenance decision model is established with the maintenance occasion as a decision variable and with the minimum total maintenance cost in life cycle as the optimization goal.Under the condition of determining the maintenance occasion,a life limit parts replacement rule is proposed to solve the optimal life limit parts replacement strategy and cost,a dynamic programming algorithm is used to solve the optimal maintenance strategy and cost of the aero-engine module maintenance.Finally,the particle swarm optimization algorithm is used to solve the civil a ero-engine maintenance decision model.The simulated data and real-world data is used to evaluate and validate the proposed method.The results show that the proposed method can effectively solve the civil aero-engine maintance time decision problem.For a large number of aero-engine maintenance decision problems,the particle swarm optimization algorithm has the disadvantages of slow speed,low efficiency,and results insufficient.Reinforcement learning can storage the the knowledge,and it has the characteristics of fast solution speed.Therefore,the Q learning algorithm in reinforcement learning is introduced into the civil aero-engine maintenance decision problem.Firstly,the aero-engine maintenance process is regarded as a Markov decision process with each flight cycle to decision,the civil aero-engine life cycle maintenance decision model is re-established,and the state,action,reward,Q table in the model are designed respectively.Then,the Q learning algorithm is trained to solve the model.Finally,the numerical experiment method is used to evaluate and validate the proposed method.The results show that the maintenance decision method of civil aero-engine based on Q learning is suitable for the maintenance decision of civil aero-engine with a small number of life limit parts and engine modules,and the feasibility of reinforcement learning application to the maintenance decision problem of civil aero-engine is verified theoretically.In view of the fact that when the number of life limit parts and aero-engine modules is large,the state space increases,and the Q learning algorithm has the curse of dimensionality.The deep Q learning algorithm is applied to the civil aero-engine maintenance decision problem.Combining the rules of civil aero-engine maintanence occasion,the life limit parts replacement and the engine modules maintanence strategy to reduce the state space and action space.Then establish the maintanence decision model of civil aero-engine based on deep Q learning.In order to accelerate the convergence of the algorithm,a deep Q learning pre-training method is proposed,which generates a certain amount of better data based on rule s to pre-train the network.Finally,the simulated data and the real-world data of airlines are utilized to verify the proposed method.The results show that in the face of large-scale civil aero-engine maintenance decision problems,the proposed method can quickly get a good solution.Finally,based on the aero-engine maintenance decision method studied in this paper,combined with the needs of civil company,a prototype system for aero-engine maintenance decision was designed and developed,and it can pro vide technical support for civil aero-engine life cycle maintenance decision.更多还原