Remaining Useful Life Prediction Of Aero-engine Based On Performance Degradation Data

As the core of airplanes, engines are undoubtedly the key point of aviation safety. With the accurate reliability assessment on every stage of engine design, manufacture, working and maintenance, effective guidance on application and maintenance management of aero-engines is definitely available, which means quite a lot for the flight safety and cost saving.With the rapid development of fault diagnosis, prediction and health management, along with the application of safety design and high-grade high-precision advanced technology on manufacture, reliability of the equipment is getting higher and higher, which makes it difficult to obtain adequate disabled data within a short time. Therefore, it’s not easy to realize life prediction by using traditional reliability method. However, with the continuous development of sensor technology, data acquisition and management, plenty of performance data on life information is accumulated in the process of application with aero-engines. By using this type of performance degraded data, reliability analysis and life prediction on equipment with high reliability and long life become a trend.In this thesis, firstly, under the circumstance of one-parameter, we assuming that the degradation of aero-engines is subordinated to random distribution with normal error term. The prior distribution is obtained with the historical data of the same kind, then the model parameters is real-time updated with the current monitoring data using Bayesian method. With Markov Chain Monte Carlo simulation method, the posteriori distribution is obtained. Finally the remaining useful life of the engine is predicted under a certain failure threshold.A data processing method on multidimensional performance monitoring data in several operating modes is discussed. This method includes best relay selection of parameters and dimensionality reduction of data. Using the principal component analytical method, data dimensions are reduced, disabled data space is structured. On this basis, the Euclidean distance between the projection of disabled data space and the disabled data center is defined as the current performance state with new parameter, by which the feature extraction of multi-function parameter to one-dimensional parameter is achieved.Once the state parameters are obtained, two aspects of “life prediction based on similarity” are discussed below:(1) the performance parameters are multidimensional, reference samples are adequate and disabled data is available. The data is reprocessed by Kalman filtering method, interference information is discarded, a method based on the theory of similarity life prediction is developed, then the prediction effect verification is realized by using the simulation data of the aviation engines, problems could be found. After that, causes are analyzed, primary model is modified. Finally verification and results comparison are carried out, which indicated that the precision and practicability of the modified method are much better.(2) the performance parameters are multidimensional, reference samples are inadequate and not disabled. A life prediction method based on the combination of support vector machine and similarity is introduced. Degraded tracing models are structured using training samples, failure time of the reference samples is predicted. Then according to the running time of the samples in service predicted by the support vector machine, the similarity between the reference samples and the test sample is calculated, and the remaining life of the engines is precisely predicted.