航空装备在交通运输和国防安全领域具有重要地位。滑动轴承是航空装备中广泛使用的一类核心零部件，具有回转精度高、安装空间小、噪声低等优点。磨损是滑动轴承的主要失效形式之一，对零部件乃至装备的运行安全有着直接影响。本文聚焦于滑动轴承磨损在线诊断与预测问题，结合经典磨损理论以及人工智能方法，建立物理模型与数据驱动并行结合的深度融合模型，基于参数化指标以及数字化磨损轮廓实现了滑动轴承在线磨损诊断与预测。具体研究内容和成果如下：针对滑动轴承磨损轮廓动态映射问题，提出了深度融合理论框架，构建了有限元-神经网络并行深度融合模型（parallel deep fusion model of finite element and neural network, PFENN）。对滑动轴承承载面上的磨损状态进行数学描述，定义了磨损状态下承载面节点的数据结构，给出了磨损轮廓映射算法，完成数据驱动模型与物理磨损模型的结合，实现了“状态监测数据-磨损状态参数-磨损轮廓”的映射过程。针对滑动轴承磨损在线诊断问题，提出了分别适用于小样本量和大样本量情况的数据驱动诊断方法。小样本量情况下，基于轴承振动信号构建了多域特征，提出了基于稀疏相关向量迭代指数退化（relevance-vectors-based iterative exponential degradation, RV-IED）的磨损诊断与预测方法；大样本量情况下，建立了多尺度门控卷积神经网络（multi-scale gated convolutional neural network, MGCNN），具备自适应特征处理能力，可实现“振动信号 → 最大磨损深度”的端到端在线诊断。针对滑动轴承磨损轮廓计算问题，提出了基于有限元-神经网络串行（sequential hybrid of finite element and neural network, SFENN）代理模型的磨损轮廓动态计算方法。基于 Archard 磨损理论推导了滑动轴承的磨损率模型，结合有限元方法实现应力-磨损迭代计算，由工况条件、材料性质等静态参数计算出对应的磨损轮廓；使用神经网络学习静态参数与磨损轮廓的非线性关系，构建出了 SFENN 代理模型，实现了新工况下磨损轮廓瞬时输出，解决了传统数值计算方法时间成本高的问题。设计并搭建了滑动轴承试验平台，开展所提方法的验证工作。在此基础上，基于迁移方法将 PFENN 模型应用于某型航空燃油泵滑动轴承，并开发了滑动轴承磨损监测系统。结果表明，本文所提方法可基于参数化健康指标和磨损轮廓动态映射反映轴承磨损状态，有效实现滑动轴承磨损在线诊断与预测。

Aircraft equipment holds a pivotal position in the fields of transportation and national defense. Sliding bearings, widely employed core components in aircraft equipment, offer advantages such as high rotational precision, compact installation space, and low noise. Wear represents a primary failure mode for sliding bearings, directly impacting the reliability and safety of components and equipment. This paper focuses on the online diagnosis and prognosis of sliding bearing wear, combines classical wear theories with artificial intelligence methods, and establishes a deep fusion model integrating a physical wear model and a data-driven model, which achieves online wear diagnosis and prognosis of sliding bearings wear based on parameterized health indicators and digital wear profiles. Specific research content and achievements include:In terms of the dynamic mapping of sliding bearing wear profiles, a deep fusion framework is proposed, leading to the development of a parallel deep fusion model of finite element and neural network （PFENN）. Mathematical descriptions and analyses are conducted on the degradation behavior and states of wear on bearing load surfaces. A wear profile mapping algorithm is defined, integrating data-driven and physical wear models to realize the mapping process of ”state monitoring data→wear state parameters→wear profiles.”In terms of the online diagnosis of sliding bearing wear, methods are proposed for small and large sample sizes respectively. For scenarios with small sample sizes, a wear diagnosis and prognosis method based on the relevance-vectors-based iterative exponential degradation （RV-IED） method is proposed, utilizing multi-domain features combining time-domain, frequency-domain, and nonlinear features extracted from bearing vibration signals. For large sample sizes, a multi-scale gated convolutional neural network （MGCNN） is developed, equipped with adaptive feature processing capabilities, achieving end-to-end online diagnosis from ”vibration signals → maximum wear depth.”In terms of the calculation of sliding bearing wear profiles, a dynamic wear profile calculation method based on the sequential hybrid of finite element and neural network （SFENN） surrogate model is proposed. Based on the Archard wear theory, a wear rate model for sliding bearings is derived. Iterative stress-wear calculations are performed using finite element method to compute wear profiles based on static parameters such as operating conditions and material properties. A neural network learns the nonlinear relationship between static parameters and wear profiles, constructing the SFENN surrogate model to achieve a real-time output of wear profiles under new operating conditions, addressing the high time costs associated with traditional numerical methods.A sliding bearing test rig is designed and constructed to validate the effectiveness of the proposed theories. Furthermore, PFENN is applied to a specific type of aircraft fuel pump sliding bearing using transfer learning method, and a sliding bearing wear monitoring system is developed. Results demonstrate that the proposed methods effectively enable online diagnosis and prognosis of sliding bearing wear based on parameterized health indicators and dynamic mapping of wear profiles.