Structural Health Monitoring for Rotating Machinery based on Operational Modal Analysis and Artificial Intelligence

现代飞机涡轮风扇发动机暴露在高温、复合载荷和环境影响下，需要大量的制造工作，采用最先进的材料以及涂层和喷丸等附加处理技术。随着发动机技术的发展，对机械测试设施采集数据的要求也在提高，要求在各种载荷情况下提高精度和全面确定结构响应。该研究旨在开发一种新的混合方法，用于在机械纺纱试验的背景下实时评估结构状况。本研究的主要前景是有助于提高效率和信息价值的机械试验。这对依赖于测试数据的其他领域提出了有益的影响，包括优化部件设计、潜在地降低维护成本以及提高燃气涡轮机发动机的操作安全性。该项目的目标可以细分为以下三个关键目标，这将在本研究过程中涵盖。第一个目标是建立一种方法，该方法允许使用操作模态分析（OMA）在测试操作期间准确估计旋转结构的动态（模态）特性。现有的方法不能充分覆盖旋转测试所特有的各个方面，例如，具有低振幅随机激励的高程度的谐波负载、变化的操作速度、由于轴承和齿轮中的摩擦而变化的温度等。因此，本研究将评估和解决目前在这一领域的局限性，测试系统的测量模态参数可用于调整相应的结构通常基于有限元（FE）的计算机模型，以实现真实的零件或组件的更接近的表示。这个过程，被称为模型更新，是特别可取的，因为有限元模型的机械结构已成为一个关键要素，在设计过程中，随着计算机辅助工程（CAE）的兴起。由于模型更新涉及迭代模拟，计算时间成为一个限制因素。因此，进一步的目标是调查与新的优化algorithm.Finally，第三个目标是涵盖人工智能（AI）的集成过程中的结构健康监测的测试系统，如旋转风扇，因为目前在这方面的研究主要限于土木工程应用的旋转结构模型更新的实施。为此目的考虑了两种主要方法。一方面，利用基于实际测量数据的机器学习，对标称（未损坏）条件进行自适应表征，有可能提高故障检测的可靠性。另一方面，AI驱动的模型可以在不同结构条件下的机械系统的FE仿真数据上进行训练。当应用于实际测试运行后，这样的AI模型可以允许实时识别和定位损坏，而不会阻碍原始FE模拟的计算需求。

Modern aircraft turbofan engines are exposed to high temperatures, combined loads, environmental influences and require great manufacturing effort featuring state-of-the-art materials with additional treatment technology like coating and shot peening. As engine technology evolves, requirements to data acquired by mechanical testing facilities rise as well, demanding increased precision and comprehensive determination of structural response at various loading scenarios. The proposed research aims to develop a novel hybrid method for real-time evaluation of the structural condition in the context of mechanical spinning tests. The main prospect of this research is to contribute to increased efficiency and informative value of conducted mechanical tests. This poses beneficial implications for further areas relying on test data, including the optimisation of component design, potential reduction of maintenance costs and increased operational safety of gas turbine engines. The presented aim of the project can be subdivided into the three following key objectives, which will be covered in the course of this research.The first objective is to establish a method, which allows to accurately estimate the dynamic (modal) properties of a rotating structure during test operation using Operational Modal Analysis (OMA). Existing methods do not sufficiently cover various aspects, which are specific to spinning tests, e.g. the high degree of harmonic loading with low amplitude random excitation, changing operating speed, varying temperatures due to friction in bearings and gears, etc. Therefore, the presented research will evaluate and address current limitations in this area.Measured modal parameters of the tested system can be used to adjust corresponding structural computer models, which are usually based on Finite Elements (FE), to achieve a closer representation of the real part or assembly. This process, known as model updating, is especially desirable since FE models of mechanical structures have become a key element in design processes with the rise of Computer Aided Engineering (CAE). Since model updating involves iterative simulations, computing duration becomes a limiting factor. Therefore, a further objective is to investigate the implementation of model updating for rotating structures in conjunction with novel optimisation algorithms.Finally, the third objective is to cover the integration of Artificial Intelligence (AI) into processes for structural health monitoring of a tested system, e.g. a rotating fan, since current research in this area is mostly limited to civil engineering applications. Two main approaches are considered for this purpose. On one hand, an adaptable characterisation of the nominal (undamaged) condition, utilising machine learning based on actual measurement data, has the potential to increase the reliability of fault detection. On the other hand, AI-driven models can be trained on FE simulation data of a mechanical system in different structural conditions. When applied to actual test runs afterwards, such AI models may allow to characterise and locate damage in real-time, without the impeding computational demands of the original FE simulations.