Research On Gas Path Performance Degradation Prediction Method For Marine Gas Turbine Based On Deep Learning

Predictive maintenance is an effective means to improve the reliability and economy of marine gas turbines.It is of great significance for the reliable and efficient operation of the gas turbine to realize the evaluation and prediction of the degree of gas turbine degradation by digging out the degradation law contained in the historical operation monitoring data of the gas turbine.This paper takes marine three-shaft gas turbines as the research object,studies the gas turbine performance degradation simulation technology,introduces deep learning theory,designs marine gas turbine trend prediction and remaining useful life prediction algorithms,and completes the simulation test verification of related algorithms.The main research contents are as follows:(1)Simulation study on the gas path performance degradation during overhaul period of gas turbine: The gas turbine gas path performance degradation model is established,which includes gas turbine gas path mathematical models,gas path component degradation models,and gas path component degradation trajectory models,to realize gas turbine performance degradation simulations.The calculation model of gas turbine maintenance period and the of degradation impact factor are established to solve the boundary conditions calculation model setting of the gas path performance degradation model simulation under the combined effects of multiple continuous maintenance periods of recoverable degradation,unrecoverable degradation and permanent degradation.The Monte Carlo simulation method is introduced with comprehensively considered operating conditions,environmental temperature,initial degradation conditions,sensor noise and other factors,and 600 sets of gas path degradation simulation for marine gas turbine tests during the maintenance period are carried out.The gas path performance degradation data sets for marine gas turbine are composed of the design operating condition gas path performance degradation data set,the variable operating condition gas path performance degradation data set,and the gas path performance degradation data set with random failures was formed.It provides data support for fault diagnosis,fault prediction and health management research for marine gas turbine.(2)Research on the degradation trend prediction for gas path performance based on Long Short-Term Memory(LSTM): Through the standardization of gas path monitoring data,normalized preprocessing,and sliding window method overlapping sampling methods,the generalization ability of the degradation trend prediction algorithm for gas path performance is enhanced.The time sequence information of marine gas turbine gas path degradation monitoring data in multiple historical maintenance periods is deeply learned based on LSTM neurons,and it is realize the trend migration prediction of gas turbine gas path performance in a new maintenance cycle.Using the marine gas turbine gas path performance degradation data set,the LSTM-based gas path performance degradation trend prediction algorithm was used to predict the gas path performance degradation trend under design point and variable operating conditions.The results show that the proposed prediction algorithm realizes the trend prediction of monitoring parameters and the detection of sudden faults,with high prediction accuracy,and is not affected by environmental and operating conditions fluctuations.(3)Research on remaining useful life prediction of gas turbine based on Convolutional Neural Network(CNN): Aiming at the problem that the prediction accuracy of remaining useful life are affected by the redundancy of gas turbine monitoring parameters.The Speraman correlation coefficient and the mean impact value(MIV)model are proposed,and the sensitivity,correlation analysis and feature reduction methods of feature monitoring parameters are studied.Aiming at the trend prediction over-fitting problem that caused by the small amount of data,the study of feature parameter reconstruction based on the sliding window overlap sampling method was carried out.The two-layer convolution layer are used to extract the high-dimensional features of the gas path monitoring data,and a CNN-based remaining service life prediction althorithm is constructed to improve the convergence speed of the model while ensuring the prediction accuracy.Using the marine gas turbine gas path performance degradation data set,the CNN-based marine gas turbine gas path remaining useful life orediction algorithm is used to carry out simulation experiments on the remaining useful life of the gas path at the design point and variable conditions.The results show that the proposed method realizes the nonlinear mapping between the gas path degradation characteristic parameters and the remaining useful life,and predicts the remaining useful life of the gas turbine gas path as well.(4)Simulation test of gas path performance degradation prediction for marine gas turbine:The prototype of the gas path degradation prediction is developed,and the gas turbine gas path performance degradation prediction software is designed.The real-time degradation trend prediction of the gas path and the remaining useful life prediction are performed on the hardware in the loop simulation system.The feasibility and effectiveness of the designed marine gas turbine gas path performance degradation trend prediction and remaining useful life prediction algorithm are verified.