A hybrid model of an artificial neural network with thermodynamic model for system diagnosis of electrical power plant gas turbine

Abstract In this paper, the diagnosis system of power plant gas turbine has been developed to detect the deterioration of engine performance. This system can be analyzed the gas path measurement to predict the deterioration of engine main component by using artificial neural network. The deterioration performance data of gas turbine was generated by using the thermodynamic model. So, the artificial neural network model was built to predict the deteriorated characteristics of gas turbine. Thermodynamic model was used to simulate gas turbine performance as well as the deterioration of engine components (compressor, combustion chamber and turbine) which were represented by changing component characteristic parameters (efficiency and flow capacity). On one hand, the probability of these deteriorated components was simulated to generate deteriorated data (measurement parameters and deterioration degree of each component). On the other hand, the neural network was trained with deterioration data and the best structure of neural network (number of hidden layers, number of neurons in hidden layer and transfer function) was selected based on the minimum value of the mean square error. The different deterioration data (testing data) was generated in thermodynamic model to test the effectiveness of the neural network. The comparison between the mean square error value of single and multi-neural network output parameters at training and testing data were achieved. In final, the testing with the real engine data were achieved.

[1]  Khashayar Khorasani,et al.  Fault detection and isolation of a dual spool gas turbine engine using dynamic neural networks and multiple model approach , 2014, Inf. Sci..

[2]  Mohsen Assadi,et al.  Development of an optimized artificial neural network model for combined heat and power micro gas turbines , 2013 .

[3]  Jinde Cao,et al.  A Novel Recurrent Neural Network With One Neuron and Finite-Time Convergence for $k$-Winners-Take-All Operation , 2010, IEEE Transactions on Neural Networks.

[4]  Sanford Gordon,et al.  NASA Glenn Coefficients for Calculating Thermodynamic Properties of Individual Species , 2002 .

[5]  Mohammad Bagher Menhaj,et al.  Artificial Neural Network–Based System Identification for a Single-Shaft Gas Turbine , 2013 .

[6]  Ezenwa A. Ogbonnaya,et al.  Gas Turbine Performance Optimization Using Compressor Online Water Washing Technique , 2011 .

[7]  Jinde Cao,et al.  A Recurrent Neural Network Based on Projection Operator for Extended General Variational Inequalities , 2010, IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics).

[8]  Hassan B. Kazemian,et al.  A neural network based solution for automatic typing errors correction , 2010, Neural Computing and Applications.

[9]  Michele Pinelli,et al.  Computational Fluid Dynamics Simulation of Fouling on Axial Compressor Stages , 2010 .

[10]  Nurul Musfirah Mazlan,et al.  Computational Exploration of a Two-Spool High Bypass Turbofan Engine's Component Deterioration Effects on Engine Performance , 2014 .

[11]  Jinde Cao,et al.  A Novel Recurrent Neural Network with Finite-Time Convergence for Linear Programming , 2010, Neural Computation.

[12]  Stephen Ogaji,et al.  Evolution strategy for gas-turbine fault-diagnoses , 2005 .

[13]  Thomas Palmé,et al.  Gas turbine sensor validation through classification with artificial neural networks , 2011 .