【作者】 董学育；

【导师】 陈来九；

【作者基本信息】 东南大学 ， 热能工程， 2001， 博士

【摘要】 过程故障检测与诊断技术的工程应用正日益为人们重视,已有多种故障诊断方法如:参数估计方法、应用模式识别技术以及基于规则推理的诊断技术等,在不同领域得到实际应用。近年来人工神经网络已经被成功应用于系统辨识和模式识别,在故障检测与诊断方面也有成功应用的报道。 电站设备或系统的安全与经济运行一直为人们所关注,某些设备或系统的故障可能会导致灾难性后果,因此,建立电站故障检测与诊断系统具有十分重要的意义。电站又是一个复杂的大系统,发生在电站中的故障的类型是多样的,不可能有一种通用方法,实现对所有故障的诊断。应该根据电站中各种故障的不同性质,进行有针对性的研究,提出实用性较强的诊断方法。也就是说,电站设备或系统的故障诊断方法可以是多样的,重要的是寻求新型实用的方法。本文在分析了各种传统的故障诊断方法以及人工神经网络原理的基础上,提出了两种可应用于电站设备或系统的神经网络故障诊断方法,即基于神经网络系统辨识能力的方法和基于神经网络模式识别能力的方法。 本文的主要工作和成果是: (1) 利用人工神经网络的非线性动态系统辨识能力,建立测量参数的神经网络预测模型,根据实际测量值与神经网络模型的预测值之间的残差,可实现故障的检测。此外,利用神经网络预测值,可以暂时替代发生故障的测量。在用其它方法排除测量故障原因后,根据残差信号,可实现对过程的故障检测。 (2) 利用神经网络辨识能力,建立电站中设备的静态特性模型,通过残差分析,可确定设备性能下降故障。 (3) 利用BP网络的模式识别能力,对设备性能下降的程度进行诊断。这将为建立电站性能监测与诊断系统提供新的方法,为在电站实现状态预测性检修创造条件。 (4) 利用竞争型神经网络,实现对凝汽器系统的故障诊断。这一方法可以用于那些可以由故障征兆描述的故障的分类诊断。更多还原

【Abstract】 The engineering application of the process fault detection and diagnosis technology has been attached more and more importance by people during recent years. There are many kinds of fault detection and diagnosis methods, such as estimation method, rule-base reasoning and pattern recognition techniques, which are applied in different areas. Lately the artificial neural networks have been used successfully in system recognition and pattern recognition tasks, and their suitability for fault detection and diagnosis problems has also been demonstrated.Operation with high safety and efficiency is always the focal point in power plant. Some kinds of faults or malfunctions of equipment or system in power plant may cause disasters. So, it is meaningful to establish the fault detection and diagnosis system in power plant. On the other hand, the power plant is a more complex system, so the fault types in power plant is much more. We can not find a general method to detect or diagnose all kinds of faults in the power plant. To put forth an effective diagnosis method, It is necessary to study the diagnosis techniques in different ways according to different kind of fault in power plant. That is to say, we can use different method to detect and diagnose faults in power plant, but the most important thing is to find the new and practical diagnosis method. Two ways of fault detection and diagnosis using artificial neural network for power plant are provided in this paper, which is based on the analysis in the conventional diagnosis techniques and the principle of artificial neural network. One is based on the ability of system identification of the artificial neural network, the other is based on the ability of pattern recognition of the network.The main works and achievements in this paper are as follows:(1) Based on the ability of the artificial neural network in nonlinear dynamic system identification, set up a prediction model using artificial neural network, we can detect the fault of sensor by using the difference signal between the real measurement value and the prediction value. In addition, we can use the prediction value to substitute for the error measurement value temporarily. After we can exclude the fault of the sensor by other methods, the process fault can be detected according to the difference signal.(2) Based on the ability of the artificial neural network in system identification, set up the static model for the equipment in power plant. According to the analysis of the difference signal, we can detect the lowering of the performance of the equipment.(3) It is possible to diagnose the degree of the performance deterioration of the equipment by using the ability of the artificial neural network in pattern recognition. The results have shown that this method is effective. It can provide a new way for the performance monitoring and diagnosing in power plant, and provide a foundation of predictive maintenance in power plant.(4) Using competitive neural network to diagnose the fault in condenser system. This method can be used to diagnose or classify the faults that can be described by fault symptom.更多还原