Research On Multiple Model-based Gas Path Fault Diagnosis For Marine Gas Turbine

The multiple model method can combine the parameter estimation with the pattern recognition to help overcome the problems of multiple faults diagnosis,under-determined estimation and the coupling between gas path fault and sensor fault in model-based gas path diagnosis and improve the performance of model-based gas path diagnosis methods.In this thesis,the multiple model method was introduced into the gas path fault diagnosis filed and systematically studied.This thesis mainly studies model set establishment,filter design and simultaneous diagnosis of sensor fault and gas path fault in multiple model based gas turbine fault diagnosis method,further improve the performance of the model-based diagnosis method and complement current diagnosis theory.The main studies were as follows:(1)The nonlinear model and linear state-space model for gas path fault diagnosis were established.The components' map fitting method was studied and a fitting method based on curve transformation was proposed for the compressor map,which can achieve higher accuracy with fewer optimization coefficients.Then the analytical nonlinear model was established based on the fitting results.Based on the nonlinear model,a piecewise linear model was established and a complete and non-redundant linearized points selection method based on gap metric was proposed,which can achieve higher accuracy with fewer models.(2)The application of multiple model method to the linear gas path fault diagnosis was studied.The problems faced by the multiple model method in the detection and isolation of gas path fault were analyzed,and a parameter augmentation method was proposed,which can quickly establish various hypothetical models based on the normal model.A hierarchical fault detection and isolation framework was developed for multiple faults.And a model set adaptive generation method was proposed,which can adaptively generate the next level model set according to the previous level detection result.Several case studies were conducted and the results show that the method has 95.5%,94% and 95% accuracy for single fault,dual gradual faults and dual abrupt faults,respectively,and can still accurately detect and isolate faults under different sensor number and outliers exist.In addition,based on the detection results,a generalized likelihood ratio estimation method was proposed,which realizes accurate estimation of the fault amplitude under under-determined condition.(3)The application of multiple model method to nonlinear gas path fault diagnosis was studied.Aiming at the filter design problem,a multiple model method based on strong tracking extended Kalman filter(STEKF)was proposed.And a method based on analytical linearization was proposed to obtain the Jacobian matrix,which significantly improves the computational efficiency.The performance of nonlinear multiple model method which based on common nonlinear Kalman filters was compared and analyzed.The results show that the method based on STEKF has the computational cost close to the method based on extended Kalman filter(EKF)but has strong robustness to uncertainty,and has 96% and 99% accuracy for dual abrupt faults,92% and 94% accuracy for dual gradual faults in steady state and transition condition,respectively.In addition,based on the detection results,a fault amplitude estimation method based on STEKF was proposed,which realizes an accurate estimation of the fault amplitude under under-determined conditions.(4)The multiple model based sensor and gas path coupling fault diagnosis method was studied.The nonlinear multiple model method was first used to detect and estimate the amplitude of the sensor fault.Considering the error between the hypothetic model and the actual condition is easy to cause incorrect detection,a detection-estimation framework was developed,which realizes accurate detection and estimation of gradual and abrupt faults.For the coupling problem between sensor fault and gas path fault,a fault validation method based on Chi-square test was proposed,which can effectively confirm the actual faults.Finally,a detection-estimation-validation framework based on multiple model method was developed and analyzed.The results show that it can achieve the simultaneous and accurate diagnosis of sensor and gas path faults and has 97% and 94% accuracy for sensor and gas path fault in the case of single coupling fault,and the estimation error of both faults is less than 0.04.(5)The hardware-in-loop(HIL)simulation verification technology of multiple model based coupling fault diagnosis method was studied.Based on the NI CompactRIO real-time controller and Labview software,a modular fault diagnosis system based on multiple model method was developed and the HIL simulation verification platform was set up.The real-time fault simulation results show that the proposed method can accurately detect and estimate the coupling fault in the HIL simulation environment,which lays a foundation for its practical application in marine gas turbine fault diagnosis.