Research On The Performance Degradation Prediction And Maintenance Decision For Turbomachinery

This thesis was supported by the major science and technology project "advanced manufacturing intelligent service" of Sichuan Province(Grant No.2019YFG0397).With the continuous development of manufacturing industry,turbine machinery is widely used in industrial compressors,fans,gas turbines,aeroengines and other mechanical devices.The structure of turbomachinery system is complex and the components are interrelated and coupled.During the operation,the performance state of the equipment will gradually decline until the failure occurs,which leads to the shutdown of the equipment and seriously affects the normal operation of the equipment.Therefore,in this thesis,the performance of complex turbomachinery was evaluated based on data fusion technology.The Long Short-Term Memory network model with attention mechanism was established to improve the accuracy of prediction of equipment performance decline.The preventive maintenance strategy to minimize the average maintenance cost of the equipment was formulated to ensure the performance reliability of the equipment.Firstly,aiming at the deficiency of single performance index to represent the degradation state of equipment performance,based on the information entropy theory,the optimal selection of sensor combination was carried out with the objective of maximizing joint entropy mutual information and minimizing total correlation.Principal component analysis was used to reduce and fuse the multi-dimensional performance indexes of equipment,and the contribution rate of each principal component was taken as the weight.Combined with fuzzy membership degree,the performance state evaluation model of equipment was established to realize the comprehensive performance evaluation of equipment.Then,considering the time-dependent relationship between the time series of equipment performance,and the different contribution rate of different performance indicators to the future decline state of equipment,a Long Short-Term Memory network model with attention mechanism was proposed to predict the performance decline of equipment.And the model could automatically learn to assign feature weights to improve the prediction accuracy.The error analysis of equipment performance degradation prediction results was carried out by using relevant evaluation standards.The simulation results showed that the proposed method was feasible and effective.Finally,based on the results of equipment performance status evaluation and performance degradation prediction,performance degradation factor was introduced to characterize the impact of different performance status on equipment maintenance.The equipment maintenance decision-making model was established with the detection interval and preventive maintenance threshold as the optimization variables,the minimum average maintenance cost as the objective.The genetic algorithm was used to solve the problem,and the optimal maintenance strategy was formulated.Combined with examples,this method had certain guiding significance and practical value for turbine machinery maintenance.