On Fault Diagnosis For Condenser Of Power Plant Based On Neural Network

In recent years, it has been an important object for the scientist and technician to research the detection and diagnosis to the performance of power plant. Some kinds of faults or malfunctions of equipment in power plant may cause disasters. So, it is very meaningful to establish the fault detection and diagnosis system in power plant. Two ways of fault diagnosis using artificial neural network 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 SOM neural network, the other is based on fuzzy BP neural network.On the base of fault simulation model of the condenser in a power installation which is built, according to the characteristics of condenser, the paper dynamically simulate nine kinds of faults frequently occur in the condenser operation. The fault sets, fault symptoms and fault feature data are summarized.A method based on self-organizing neural network is applied to realize fault diagnosis of condenser. The structure of SOM network and learning algorithm is introduced. The most distinct feature of SOM neural network is that its training is an unsupervised process. It also has some advantages such as fast learning, lateral association and simple structural. When one kind of vector input the neural network, a nerve cell of the output layer will reach the maximum value, the other nerves of output layer reach the minimum value. So, according to the position of nerve cell with maximum value, SOM can judge the fault represented by input vector. Therefore, when we use enough different types of fault swatches to train the neural network, the every type of fault has a certain mapping area in the output planar plane, so all faults can be correctly mapped in the output plane of SOM neural network. SOM has its particular strength in pattern recognition and fault diagnosis, that is, the fault diagnosis result is simple and intelligent and can intuitively reflect the possibility interrelated with fault symptom. An instance of condenser fault diagnosis is presented in Matlab environment. The simulation result indicates that the method is effective.BP neural network is one kind of well developing and widely using networks. It has many advantages such as self-adaptability, nonlinear recognition, strong anti-noise capability and fault tolerance. But it also has some disadvantages such as lower learning efficiency, slow training speed, local minimum and so on. To overcome the defect of the basic BP algorithm, a new faster learning algorithm is proposed. The training process is divided into two steps. First, it use BP algorithm to approach the optimal point, then change the descent gradient to speed up the convergence. Simulation shows that the proposed method is efficient and fast.Fuzzy BP neural network is applied to condenser fault diagnosis. The input of fuzzy BP neural network is the fuzzy fault symptoms, and the output is the fault type grade of membership, and the prediction function of if-then rules is accomplished by the self-learning of neural network. Based on the fuzzy fault symptoms, fuzzy BP neural network model is constructed, including the network structure, network parameters and membership function and so on. The fault swatch is used to train the BP network, and then make use of the knowledge of repository to diagnosis the condenser fault. Fuzzy BP neural network greatly improve the performance of neural network, enhance the fault diagnosis precision. The simulation results show that the method has a good effect on fault diagnosis.Based on Matlab and Visual C++, the simulation system for the power plant fault diagnosis is constructed. Visual C++ is an object oriented method. It is close to the human thinking manner, and it is easy to make structured and easily transplanted programs. Visual C++ provides many kinds of convenient and simple developing approaches for database and class libraries with strong function, enhances the developing flexibility of database. The fault diagnosis system for power plant adopt mutual interface between man and computer, interface is friendly and simple.