燃气轮机运行初期故障诊断与预警研究

Research on Initial Fault Diagnosis and Early Warning of Gas Turbine

为了在早期燃气轮机机组运行过程中发现潜在的机组故障,提出了使用门控循环单元(GRU)构建基线模型的参数趋势分析方法。针对早期燃气轮机缺乏历史故障数据,难以对机组进行具体的故障定位的问题,考虑到机组发生故障会使机组性能发生变化,进而促使燃气轮机的运行参数发生变化,将神经网络建立的基准模型与运行参数实际值的参数偏差作为监测对象,设立偏差阈值,从而在燃气轮机运行异常时发出预警。该方法在初期燃气轮机的故障预警中是一个简单有效的系统异常分析方法,研究内容可以为后续的故障分类提供指导。

In order to find out potential faults in the early operation stage of gas turbine units,a parameter trend analysis method is proposed to build baseline model using gated recurrent unit(GRU).Because of the lack of historical fault data of early operation stage of gas turbine,it is difficult to verify the specific fault in the unit.Considering that the failure of the unit will cause performance change of the unit,and then make the operation parameters of the gas turbine change,therefore,the parameter deviation between the benchmark model established by neural network and the actual value of the operation parameters is taken as the monitoring objective,and then,the deviation threshold is set so as to warn in case of abnormal operation in gas turbine.This method is simple and effective in the early fault warning of gas turbine.The research content can provide guidance for the subsequent fault classification.