| As one of the important power plants in modern society, gas turbine has been widely used in the industry. Due to the complex structure and poor working conditions of the gas turbine, its components are prone to various failures. The failures will directly affect the normal operation of the whole device, which causes significant economic losses, and even endangers the personal safety. Therefore, the fault diagnosis of gas turbine has great significance.Turbine blades are the key component of gas turbine. They operate in the high temperature and high pressure environment for a long time that causes faults easily. Whether the turbine blades can work properly or not directly affects the stability of the entire device. Thus, choosing the turbine blades as the important part of gas turbine is significant in the practical engineering based on the fault diagnosis technology.This thesis investigated and analyzed large number of fault diagnosis technologies of dynamic system. According to the characteristics of turbine blades, effectiveness of the artificial neural network was discussed in fault diagnosis. Besides, a fault diagnosis of the gas turbine system based on recurrent neural network was implemented. The main work and contribution in this thesis include:1) The failure reasons of gas turbine and its key components are studied. According to the related documents and the actual operation of gas turbine, the principles of failures and harms are summarized. By researching of failure mechanism of the turbine blades, the fault tree analysis of turbine blades is established. The hierarchical relationship of fault events is clear by analyzing the failure mode of gas turbine based on fault tree.2) A new fault diagnosis method based on improved Elman neural network is presented to improve the accuracy of diagnostic results. The faults of turbine blades are diagnosed by static network and dynamic network of artificial neural networks, respectively. The result based on Elman neural network is more accurate because it has better dynamic characteristics. Furthermore, in order to enhance the accuracy of diagnostic results, new adjustable weights are added to optimize the structure of Elman network.The improved Elman neural network model, namely OHF Elman network, introduces transfer data between each layer at the previous time related to the expected output of the current time. Experimental results show that the OHF Elman network has a better accuracy of diagnostic results.3) A novel model of fault diagnosis based on RS-OHF Elman network is proposed to enhance the efficiency and accuracy. Because of the RS, the dimensions of input feature parameters are reduced so as to eliminate redundant information. The phenomenon that the large input parameters lead to reducing the training rate and accuracy is avoided. So the learning efficiency of samples based on neural network can be improved because the training time and calculation amount are decreased. At the same time, the noise that effects the diagnostic model is suppressed by the neural network for improving the learning and generalization ability of the network models. They combine and work together to enhance the level and accuracy of the fault diagnosis.4) A fault diagnosis of the gas turbine system is designed and implemented based on recurrent neural network. The collecting data of the gas turbine can be displayed and stored in the system. What’s more, the fault of turbine blades can be identified through the diagnosis module. The system can provide technical support to the staff so as to save the cost of maintenance. |
