【作者】 严金云；

【导师】 马宏忠；

【作者基本信息】 河海大学 ， 电力系统及自动化， 2006， 硕士

【摘要】 汽轮发电机作为旋转动力设备,其振动大小是衡量其运行状态的重要标志,利用它外部的振动信号来诊断内部的故障是现代诊断技术常用的方法。本文主要研究信号的采集、处理,并利用人工神经网络对汽轮发电机的振动故障进行诊断,为合理利用诊断资源,探索了一种远程故障诊断方式。 本文首先研究汽轮发电机振动的故障机理,介绍常用的基于故障机理及故障特征的在线诊断方法,并比较各种方法的优缺点。其次是故障信息处理技术的研究,作者从时域和频域两个角度对非整周期采样对振动信号测量精度的影响进行了分析,并给出减少同步误差对信号测量精度以及频谱分析的影响的方法。再次本文对多层前向神经网络在汽轮发电机振动故障诊断中的应用进行了研究,并结合遗传算法训练神经网络进行故障诊断。结果表明,训练过的神经网络对学习过的故障有较高的识别精度,并且具有一定的抗噪声能力。文中最后探讨了基于Internet的汽轮发电机故障监测与诊断系统,系统由数据采集与预处理、在线故障识别与控制、远程诊断三部分组成。因为时间与实验条件限制,本文仅对样本管理与神经网络训练、数据采集上报(仿真)、在线识别与控制给出了实现,验证了方案的可行性。为了保证在不同操作系统环境下的可移植性,系统采用Java语言进行开发。 本文的创新之处在于从时域和频域两个角度对非整周期采样对振动信号测量精度的影响进行了研究;在汽轮发电机故障诊断中引入了人工神经网络与遗传算法相结合的技术,通过提取故障特征,构造学习样本,并使用遗传算法对人工神经网络进行训练,达到良好的学习效果和较高的识别精度,并具备一定的抗噪声能力;通过网络技术,采用客户机/服务器(C/S)模式,在线监测应用可以同时支持多个监测客户端上报采集到的故障信息,进行在线识别后,根据识别结果下发控制指令;采用Java语言开发,具备良好的跨平台特性,所开发的应用程序在Windows XP、UNIX、Linux等平台都可以稳定运行。更多还原

【Abstract】 As the equipment that generates power by revolving, How much a turbo generator vibrates is an important symbol to judge its running status. Analyzing the external vibration signals of a turbo generator enables the diagnosis of internal faults, and this is commonly used in modern diagnosis technologies.This paper elaborates the technology of processing vibration signal and diagnosing the vibration faults of turbo generators based on the artificial neural network, and for taking advantage of diagnosing resources reasonably, this paper explores a remote diagnosis method to find out the internal faults of turbo generators.The first part of this paper introduces the fault mechanism of turbo generator’s vibration, as well as common on-line diagnosis methods based on fault mechanism and characteristics. Besides, it provides the pros and cons of each method. The second part elaborates the technology of processing fault information. In the aspects of time field and frequency field, the effect on the measure accuracy of the vibration signals from non-full-cycle sampling is elaborated. At the same time, it also details how to reduce the effect on the signal measure accuracy and spectrum analysis from the synchronous error. The third part describes how the multilayer forward nerve network is applied in the diagnosis of turbo generator’s vibration fault. What’s more, the genetic algorithm is adopted to train the nerve network conducting fault diagnosis. It is proved that when diagnosing vibration faults, a trained nerve network performs better as far as recognition accuracy is concerned, and has certain anti-noise ability. The fourth part introduces a turbo generator’s fault monitoring and diagnosing system based on the Internet. The system consists of three parts: data sampling and preprocessing, on-line fault recognition and control, and remote diagnosis.Due to the restrictions on time and experimental conditions, this paper only realizes sample management and neural network training, sample data reporting (simulation), and on-line recognition and control. In this way, the method is proved to be feasible. This system is developed in Java to ensure it is transplantable between different operating systems.It should be mentioned that the innovation of this paper is to describe the effect on the measure accuracy of the vibration signals from non-full-cycle sampling in the aspects of time field and frequency field, and it also introduce the technology of combining the artificial neural network with GA in the turbo generator’s fault diagnosis. By gathering the fault characteristics, constructing study samples, and training theartificial neural network with GA, the method enables good study effect and high recognition accuracy, and certain anti-noise ability. By using the pattern of client/server (C/S) of the network technology, on-line monitoring applications support multiple monitoring clients gather and report fault information simultaneously. After on-line recognition, control commands are sent according to recognition results. Java-based development enables the applications to run stably in Windows XP, UNIX, and Linux.更多还原