Research On Data-Driven Performance Degradation Modelling And Remaining Useful Life Prediction For Mechanical Equipments

Mechanical equipments are mostly failure patterns of performance degradation, therefore it requires scientific and effective methods to analyze and model the performance degradation of the equipment, revealing the occurrence, development and evolution patterns of the degradation. Then corresponding maintenance strategies are taken to control, correct or even self-heal the performance degradation at the right time. Also, remaining useful life of the equipment can be predicted based on the available performance degradation information, providing important theoretical basis for operation and management maintenance decision-making.Data-driven approaches obtain the information relevant to performance degradation and residual life of an individual equipment by directly processing monitoring data of the equipment under its given operation conditions, and actual situations of the equipment can be better revealed. Moreover, for the large and complex mechanical equipments that are hard to build their physical failure models, data-driven methods can also be applied to study the performance degradation and remaining useful life of these equipments.Data-driven performance degradation modelling and remaining useful life prediction for mechanical equipments is studied in this thesis. And results of this thesis can not only be used to the mechanical equipments, but also have a broad application prospects on the bridge structures and the transmission towers, etc.Main works and achievements of the thesis are as follows:(1) An information exergy feature extraction approach based on singular value decomposition (SVD) is proposed. In view of engineering practice of multiple moments (or operational conditions) and multipoint sensors condition monitoring, this paper studies information exergy methods for integrated analysis of multi-moments (or operational conditions) and multi-sensors monitoring signals at the same time. And one SVD-based information exergy feature extraction method is proposed. Through the structure response tests of one truss structure with or without bolt loose, effectiveness of the information exergy analysis method for structural damage identification is verified. Then the information exergy feature extraction method based on SVD is used for bearing performance degradation assessment, results of inner raceway and rolling element test results show that the proposed method can obtain more effective results than existing ones.(2) A performance degradation feature selection methods based on multiple indices is presented. To solve the problem of lacking performance degradation feature evaluation and selection researches, this paper proposes four indices of correlation, monotonicity, predictability, and robustness to study the performance degradation feature evaluation and selection, reducing the workload of human observation and other related uncertainties. Through rolling bearing performance degradation experimental data, the performance degradation features of individual bearing are evaluated and selected, and effectiveness of the proposed indicators are verified. Further, these indicators are applied to study aeroengine performance degradation simulation data, the related sensor signal features are systematicly evaluated, and features that can better depict the performance degradation characteristic of all individuals under the same failure mode are selected.(3) A performance degradation modeling method based on functional principal component analysis (FPCA) and the normalized running time is proposed. In view that the current performance degradation models are mostly parametric ones, this paper studies the adaptive performance degradation modeling based on FPCA. To make that each individual performance degradation data are more evenly distributed in the same observation interval, and to reduce parameter estimation error of the FPCA model, running time is proposed to be normalized before modeling. Through the airoengine performance degradation simulation data, the performance degradation modeling method based on FPCA is verified. Then the proposed method is further used for DC cooling fan performance degradation test data processing. Results show that complex performance degradation trajectories of the DC cooling fan are adaptively modelled by the proposed method, and normalization of the running time is required.(4) A similarity-based remaining useful life prediction method by FPCA is presented. The highlights of similarity-based remaining useful life prediction methods is on abstract modeling of performance degradation trajectories and ascertainment of the time range used for similarity comparisons, but the related research is currently not developed enough. Therefore, on the basis of FPCA degradation model, this paper proposes a new similarity-based remaining useful life prediction method. The proposed method models performance degradation trajectories as functional data by FPCA, then the whole performance degradation trajectory of the test individual is used to adaptively generate reference performance degradation trajectories of the same length, avoiding ascertainment of the time range used for similarity comparisons. Through the aeroengine simulation data and DC cooling fans degradation test data, effectiveness of the proposed method is verified. The research results show that the proposed method can obtain accurate and reliable remaining useful life prediction, and prediction error can be reduced with the accumulation of condition monitoring data of the test individual.