Research On The Data-driven Approaches Of Hybrid Fault Prognostics

With the rapid development of the modern science and technology industry represented by information technology,the engineering system in the aviation domain is becoming increasingly complex,and the level of intellectualization of a large number of complex systems is constantly enhancing.Considering the security,reliability and economy of complex systems,prognostics and health management(PHM)technology has gradually become the highlight of research in the field of aviation equipment comprehensive coverage.The fault prognostics technology of high-end complex aviation equipment has always been a hot research topic full of important theory and application value.Particle filtering(PF)technology has been successfully applied in the field of fault prediction because of its advantages in addressing the problem of nonlinear and non-Gaussian state estimation.There are at present two theoretical problems to be solved urgently in particle filter(PF)algorithm,namely,the appropriate design of importance probability density function and particle degeneracy problem.Aiming at these two issues,the importance probability density function of traditional PF algorithm is redesigned,and two sets of PF algorithms based on minimizing Hellinger distance(MHDPF)and Kendall correlation coefficient(KPF)are respectively proposed for the first time to cope with the problems of particle degradation and sample impoverishment.Furthermore,in order to improve the reliability and applicability of support vector regression(SVR)non-linear output prediction,the traditional SVR algorithm is enhanced,and the semi-parametric probability density estimation(semi-PDE)technique is applied to the SVR fitting deviation sequence.The SVR non-linear output prediction algorithm with error confidence is proposed,which provides more reliable incentive data for the further analysis of the subsequent PF algorithm.Next,in order to overcome the restriction of the single PF algorithm,a novel fault prognostics method based upon the combination of SVR algorithm and PF algorithm is developed from the perspective of data-driven approach,which is characterized by deep fusion and uncertainty expression,namely SVR-PF algorithm.It is applied to the fault prognostics of Electro-Hydraulic Servo Actuators(EHSAs)of civil aircraft.What's more,a fault prediction method based on the fusion of MHDPF and Paris physical model,namely MHDPF-Paris,is also proposed to predict the remaining useful life(RUL)of metal structural components with fatigue damage.Finally,the proposed two fusion fault prognostics approaches are tested and validated grounded on the actual EHSA test platform and the 2A12-T4 aluminum alloy sheet.Compared with the existing fault prognostics methods,the experimental results are satisfactory.