Research On Performance Assessment And Degradation Prediction Of Aeroengine

Airlines try to guarantee that all the on-wing aeroengines are safe and reliable, and decrease the engine maintenance cost. The performance assessment and degradation prediction of aeroengines can support the maintenance decision making of Airlines. Due to complexity of the aeroengine, the exact physics-based model is difficult to be constructed; thus, the data-driven approach based on the aeroengine monitoring data is a solution for performance assessment and degradation prediction of aeroengine. The dynamic operational conditions of the aeroengine and the complicated degradation mode pose great challenges on the performance assessment and degradation prediction of the aeroengine, whose results influence the maintenance decision. Thus far, the multi-regime and multi-parameter performance assessment is one problem; the complicated degradation property brings difficulty to the prediction model. Therefore, the research on performance assessment and degradation prediction of the aeroengine is important.In this thesis, based on the data-driven approaches, several aspects of performance assessment and degradation prediction of the aeroengine are explored, including performance degradation assessment, performance prediction, parameter prediction, the remaining on-wing life prediction, and the uncertainty assessment.The performance degradation assessment of the aeroengine is studied. Due to the massive monitoring parameters and their complex relationships, the tree-formed representation of aeroengine monitoring data is proposed, and massive monitoring parameters are organized according to their functionalities and memberships. Then, to overcome the problem of how to measure the difference between two monitoring data, based on the tree-formed representation, the distance between two monitoring data trees are defined. In addition, it is also proved to be a generalized distance. Last, the performance degradation assessment approach is presented based on the defined distance. When combined with local model approach, the approach can handle the discrete multi-regime performance degradation assessment of aeroengine.The short-term performance degradation prediction of the aroengine is researched. Conventional prediction models neglect the time accumulation effect of degradation process, and do not consider the trend of degradation. To overcome these two drawbacks, the process fuzzy rule-based model(PFR) is proposed. The inputs and outputs of PFR model are all continuous functions, which are used to consider the time accumulation effect and degradation trend, respectively. Based on the derivation, the analytical solution of PFR model is presented, and the rapid parameter identification algorithm is proposed; thus, no iteration is required in the training phase of PFR model, making the PFR model efficient and stable. The historical performance degradation states are adopted to train the PFR model, which then can predict the future performance of the aeroengine. The results show that PFR model is accurate, stable, and efficient.The short-term prediction of performance parameter is explored. In view of the problem that the performance parameters of the aeroengine are influenced by regime parameters, the double-reservoir echo state network(DRESN) model is adopted. The historical performance parameters and corresponding regime parameters are aggregated to predict future performance parameters; thus, the influence of regime parameters is considered effectively in the prediction model. The analytical solution of DRESN model is derived, and the rapid training algorithm is presented, making the prediction model accurate and efficient.The aeroengine remaining on-wing life prediction is studied. To improve the accuracy of remaining on-wing life prediction, according to the circle of case-based reasoning(CBR), the ensemble learning model and similarity-based approach are combined, and the neighborhood enhanced and two-layer random forests model is proposed. First, the neighborhood enhancement approach is proposed, the historical engines that have similar degradation processes are chosen to enhance the training data, making the training data focus on similar degradation processes. Then, the two-layer model scheme is proposed to improve the accuracy of random forests. Last, these two approaches are combined to predict the remaining on-wing life.The prediction interval estimation of prediction models is investigated. To evaluate the uncertainties of the above three point estimation models, the prediction interval estimation approach is adopted. Based on the residuals of point estimation models, the prediction interval upper limit and lower limit estimation models are constructed using extreme learning model, which then can estimate the prediction interval.The research in this thesis can enrich the methodology of the aeroengine PHM, and it can improve the reliability of the Airlines. In addition, the methods in this thesis can also be applied to other complex equipment.