基于自编码神经网络的汽轮机故障诊断  

Fault diagnosis of steam turbine based on autoencode neural network

作　　者：赵朋 李学敏[1] 范世望 余洁 ZHAO Peng;LI Xue-min;FAN Shi-wang;YU Jie(Huazhong University of Science and Technology,School of Energy and Power Engineering,WuHan 430074,China;Shanghai steam turbine works Co.,Ltd,ShangHai 200240,China)

机构地区：[1]华中科技大学能源与动力工程学院,湖北武汉430074 [2]上海汽轮机厂有限公司,上海200240

出　　处：《能源工程》2021年第3期14-19,24,共7页Energy Engineering

基　　金：国家重点研发计划项目(2018YFB606101)。

摘　　要：汽轮机正常运行工况的数据比较容易采集,而故障数据的采集存在一定的困难。针对这种情况,采用自编码神经网络对汽轮机高压缸内三个级组的运行数据进行建模,用正常运行工况下的数据训练模型之后对未知工况数据进行重构分析。通过分析,重构前后压降、温差以及效率的变化情况与常见通流故障结垢、磨损、等发生时相关参数的表现,最终分析得出高压缸的一段抽汽到高压排汽这一级组发生了磨损故障的结论。The normal operation condition data of steam turbine is easy to collect,but it's difficult to collect data under unnormal operation conditions.In view of this situation,the autoencode neural network is used to model the operation data of three stages in the high pressure cylinder of steam turbine,and the data of unknown working conditions are reconstructed and analyzed after training the model with the data under normal operation conditions.By analyzing the changes of pressure drop,temperature difference and efficiency before and after reconstruction,and the performance of the relevant parameters when the common flow fault occurs,such as scaling fault,wear fault,etc.,the final conclusion is drawn that the stage group from the first stage extraction of the high pressure cylinder to the high pressure exhaust has occurred the wear fault.

关 键 词：汽轮机 自编码神经网络 重构分析 通流故障 

分 类 号：TK26[动力工程及工程热物理—动力机械及工程]