This paper aims to investigate the rub-impact fault of a gas turbine using the proper orthogonal decomposition (POD) and the smooth orthogonal decomposition (SOD). It is an attempt to combine POD and SOD for pattern recognition. First, simulated data are generated to investigate the sensitivity of POD and SOD to pattern changes. After that, field data from a gas turbine are presented. We consider two data sets: one related to the standard operation, and another to the rub-impact fault. The results show that POD and SOD can recognize the pattern changes, and identify the rub-impact fault. Moreover, the present work demonstrates that POD and SOD can complement each other. The proposed methodology offers the possibility to monitor the evolution of modal component patterns and might support the plant engineers to make decisions.

本文旨在使用适当的正交分解（POD）和平滑正交分解（SOD）研究燃气轮机的摩擦故障。是一种将POD和SOD结合起来进行模式识别的尝试。首先，生成模拟数据以研究 POD 和 SOD 对模式变化的敏感性。之后，展示了来自燃气轮机的现场数据。我们考虑两个数据集：一个与标准操作有关，另一个与摩擦故障有关。结果表明，POD和SOD能够识别模式变化，识别摩擦故障。此外，目前的工作表明 POD 和 SOD 可以相互补充。所提出的方法提供了监控模态组件模式演变的可能性，并可能支持工厂工程师做出决策。