Machinery Diagnostics Using Mechanistic and Data-Driven Models

所有的机器都容易发生某种故障。机械诊断允许在灾难性故障发生之前进行维护，方法是观察指示即将发生的问题的机器行为的某些方面，然后使用信号处理和特征提取来确定是否存在故障。一些技术还可以预测机器在特定条件下的剩余使用寿命。不幸的是，有限的可观察性和特定于机器的条件监控方法使得数据驱动模型难以应用于一般应用程序。新的传感器技术和嵌入式传感器网络标准有望显著提高机械故障的可观察性，无论是对新设备还是对现有组件和系统的改造。新一期基于Discovery-Grant的研究将检验这样一种假设，即提高损伤机制的可观测性将使系统可靠性得到改善。将为旋转机械和过程设备中冲击故障的实验室故障诊断案例研究评估故障分类的单独方法和组合方法。基于物理的建模将用于评估如何敏感地测量滚动元件轴承和工艺设备中易受冲击磨损的故障。实验室系统将向公众开放，这样其他人就可以使用基准数据集来测试他们方法的性能。慢性冲击事件的故障模型适用于各种旋转设备和过程单元，适用于加工厂、风力涡轮机和车辆。参数故障识别模型将对工业界和其他研究人员有直接的好处。故障严重性排序的比较方法将产生更好的方法来预测设备寿命和提高维护效率。机器损坏案例的数据集将使研究人员能够明确地比较诊断技术，为最佳实践做出贡献。改进的故障诊断将有助于提供更耐用的可修复设备和更长使用寿命的产品，这对更可持续的技术未来至关重要。

All machines are prone to some kind of failure. Machinery diagnostics allows maintenance to be done before a catastrophic failure occurs, by observing some aspect of machine behaviour that indicates an impending problem and then using signal processing and feature extraction to determine whether a fault is present. Some techniques can also predict the remaining useful life of the machine in certain conditions. Unfortunately, limited observability and machine-specific approaches to condition monitoring make data-driven models difficult to apply in general applications. New sensor technologies and embedded sensor network standards hold promise to significantly improve the observability of faults in machinery, both for new equipment and for retrofits to existing components and systems. The new instalment of Discovery-Grant based research will test the hypothesis that improving observability of damage mechanisms will enable system reliability improvements. Individual methods and combined methods of fault classification will be assessed for laboratory-based fault diagnosis case studies of impact faults in rotating machinery and process equipment. Physics-based modeling will be used to evaluate how to measure faults sensitively in rolling-element bearings and process equipment subject to impact wear. The laboratory systems will be publicly available so that others can test the performance of their method using benchmark datasets. Fault models of chronic impact events apply to a wide range of rotating equipment & process units, for process plants, wind turbine generators, and vehicles. Parametric fault identification models will be of direct benefit to industry and to other researchers. Comparative methods for ranking fault severity will yield better methods to predict equipment life and improve maintenance effectiveness. Datasets for machine damage cases will allow researchers to compare diagnostic techniques definitively, contributing to best practices. Improved fault diagnostics will contribute to more durable repairable equipment and products with longer service life, which is crucial for a more sustainable technological future.