| With the fast development of sensor technology,the expansion of sensor application and the polularization of mobile terminal,we are ushered in the era of the civil aviation big data.Aeroengine,as the ―heart‖ of aircraft,its health management based on big data is a worthwhile project to research.How can we make good use of civil aviation big data for promoting the safety,reliability and economy of aeroengine? Furthermore,in terms of time-consuming or processing multi-type data,conventional methods used for the condition monitoring are not suitable for civil aviation big data analysis and processing.The big data of civil aviation has brought us challenges and opportunities.In this dissertation,some key problems of aeroengine condition monitoring based on big data have been thoroughly investigated.Our work is based on the state-of-the-art data-driven methods.The main contributions of this dissertation are summarized as follows.(1)Because of sensor fault or missing collection,the aeroengine monitoring data could be incomplete.So we must to reconstruct those data in order to make full use of that data.To handle the incomplete sensor data,we propose an on-line data reconstruction model based on the polar incremental matrix completion(PIMC)algorithm,which represent the evolving features of high dimensional data by low rank subspace.The model extracts the current data feature from the history data and updates the subspace to track the evolving features via new data.We conduct our method on both simulation data and real data,and the experimental results show that the proposed model is robust to random missing data and noise for aeroengine sensor data reconstruction.(2)To solve the practical problem of imbalanced data clustering,where the volume of the health data is larger than that of degradation data,we propose a new clustering algorithm—— feature weighting fuzzy compactness and separation cluster(WFCS)algorithm.In view of the contribution of features to clustering,the proposed algorithm introduces the feature weighting into the objective function and can handle high dimensional data.We also propose a protype for aeroengine condition monitoring based on WFCS.Our model is composed of two parts: one is offline-learning module,in which the module parameters are iteratively computed,and the other is online-monitoring module,in which the realtime data can be classified according to the parameters.The realtime data are inputted into the offline-module,and then the module parameters are updated.It can be shown from the experiments that the proposed protype works well for aeroengine condition monitoring with good robustness and generalization and has practical application value for uncertain aeroengine condition monitoring.(3)Aeroengine degradation pattern consists of normal performance degradation and fault-caused degradation,the performance of aero-engins changes even the degradation pattern is the same,For example,the degradation speed of different aeroengines varies because of different operational regimes.Therefore,it is difficult to precisely define the degradation model of an aeroengine.A large quantity of on-board degradation data gathered by airways contains lots of valuable information,and these degradation data constitue the sample database.But we have not taken advantage of these data for aeroengine degradation pattern recognition yet.We have proposed a multi-parameter fault diagnosis model based on the canonical time warping(CTW)algorithm,which can simultaneously project the multi-sourses data to a unified feature space and recognize the degradation pattern through matching data on the timeline.The proposed model increases the accuracy and flexibility of aeroengine degradation pattern recognition.The experimental results of simulation data and real data show that our model can discriminate the normal degradation and fault-caused degradation and can locate the fault in component level.(4)According to the real condition of airlines,we propose a system framework of engine health management and maintainence decision support system(EHM&MDSS).The framwork consists of three layers: physic layer,data layer and application layer.Furthermore,the data quality is very important for data-driven based engine health management,so we also propose the protype,methods and flow of data quality management under the EHM&MDSS system framework.Then we carry out a detailed system requirement analysis and design the system function modules.After many discussions with the airlines,they will accept this system framework.And this part of work has some guiding significance for the application of big data for the aeroengine health management and maintainence decision support. |
