航空发动机作为飞机的主要动力来源，对于确保飞行安全至关重要，其核心技术也是大国博弈的焦点。由于航空发动机是一种复杂精密部件，工作条件相对苛刻，这使得发动机的部件很容易发生故障，一旦出现任何问题，都有可能造成严重的后果，甚至有可能造成人员伤亡。根据中国民用航空局2021年发布的《中国民航航空安全报告》，在2011-2020年间发动机空中停车总计159次[1]。当前针对航空器的维修工作大部分还停留在定时维修与事后维修，该方式耗费人力、物力和财力，与现在快速发展的社会节奏显然不太适应，基于设备状态的视情维修受到了越来越多的关注。从视情维修的基础上衍生出的航空发动机预测与健康管理（PHM）已经成为了航空领域的一门新兴技术，它的主要内容是，通过剩余寿命预测判定发动机的健康状态，进而为维修决策提供基础；同时，当前针对航空发动机寿命预测的研究都是基于理论层面，缺乏实际工程经验，在整个研究过程中存在不完善的地方。基于此，本文将开展以下工作：（1）针对航空发动机的传感器在恶劣工作条件下采集到的大量且被干扰的数据问题，通过分析当前数据筛选方法的局限性，提出基于重要性排名的特征融合方法，实现了提取出有效的特征参数，为寿命预测模型的数据输入打下了基础。（2）针对神经网络寿命预测模型训练效果不佳的问题，提出了基于智能优化算法融合的神经网络寿命预测模型，实现了神经网络预测模型中最优参数的选择，得到了较好预测结果的同时也获得最优退化模型，为实现发动机的寿命预测以及维修决策打下基础。（3）针对剩余寿命预测结果评价标准不全面的问题，参考当前研究领域内对预测结果评价的指标，从实际工作经验出发，提出预测准确率指标，实现了理论与实际工作相结合，达到了预测结果评价的完整性效果，并通过评价指标实际验证了预测模型的可行性。（4）针对传统发动机健康等级无法直接实现维修决策的问题，将预测结果与发动机健康等级进行融合，提出新的发动机健康等级划分标准，实现了健康的维修决策，初步达到了发动机健康管理的效果，并通过实例验证新标准的合理性。
As the main source of power for aircraft, aero engines are vital to ensuring flight safety, and their core technology is the focus of major power games. As an aero-engine is a complex and precise component, the working conditions are relatively harsh, which makes the components of the engine prone to failure, and if any problem occurs, it may cause serious consequences, and may even cause casualties. According to 《China Civil Aviation Safety Report》,published by the Civil Aviation Administration of China(CAAC) in 2021, there were a total of 159 In-Fight-Shut-Down between 2011 and 2020 [1].Most of the current maintenance work for aircraft is still based on reliability analysis and situational maintenance, which cost large human, material and financial, and is clearly not compatible with the current fast-paced society, so the Condition Based Maintenance based on equipment status has received increasing attention. The Prognostics Health Management (PHM) derived from condition based maintenance has become an emerging technology in the aviation field. Its main content is to determine the health status of the engine through residual life prediction, thereby providing a basis for maintenance decision-making; At the same time, current research on predicting the lifespan of aviation engines is based on theoretical aspects, lacking practical engineering experience, and there are shortcomings throughout the entire research process. In view of the current situation, this thesis conducts the following research:(1) Aiming at the problem of large and disturbed data collected by sensors of aero engines under harsh operating conditions is addressed. A feature fusion method based on importance ranking is proposed to achieve the extraction of effective feature parameter，which provides the basis for data input to the life prediction model.(2) Aiming at the problem of poor training of neural network life prediction models, a life prediction model based on the fusion of intelligent optimization algorithms is proposed. The selection of optimal parameters in the neural network prediction model is achieved，and good prediction results are obtained while also obtaining the optimal degradation model. The basis is laid for the implementation of engine life prediction and maintenance decisions, and the validity of the fusion model was also verified experimentally.(3) Aiming at the problem that the evaluation criteria of the remaining life prediction results are not comprehensive. The result evaluation accuracy index is proposed to achieve the completeness of the prediction result evaluation, and the feasibility of the prediction model is verified by the evaluation indicators.(4) Aiming at the problem that traditional engine health classes cannot achieve maintenance decisions scientifically. A new engine class classification standard is proposed to achieve the healthy maintenance decision and the rationality of the new standard is verified by example.
