Research On Aeroengine Fault Diagnosis And Performance Parameters Prediction Based On Particle Swarm Optimization

Aeroengine is the core component of aircrafts.It is a difficult task to diagnose and prevent the faults of aeroengine due to the uncertainty and harsh of its operation environment,as well as its complex structure.With the rapid development of modern sensing and signal processing technology,condition monitoring and fault trend analysis of aeroengines become feasible.The diagnosis and prognostics technology based on intelligent algorithm has been widely adopted,and they has fundamentally changed the maintenance paradigm of aeroengine.Aiming at the diagnosis,prognostics,and data processing of aeroengines,the diagnosis and prognostics methods based particle swarm optimization(PSO)are proposed in this dissertation.The granular computing(GrC)is adopted to compress the similar data for reducing the time overhead of algorithms.It is a challenge to use PSO as a diagnosis and prognostics method.The main contributions of this dissertation are summarized as follows.(1)As a classical swarm intelligence algorithm,the PSO has been widely used in the optimization field,but it may easily trap into local suboptimal areas due to the premature convergence.Actually,the premature convergence is caused easily due to the lack of population diversity,inefficient interactive model,unbalance of search process,simplification of update strategy.Inspired by human learning behavior,the PSO is modified to overcome the premature convergence.By dividing the particles into tutors and students,a novel multiple extremum learning PSO(MELPSO)is proposed,which can imitate the strategies of collective learning,private tutoring,and research behavior,so that it has the characteristics of adaptability,interaction,dynamic and diversity in the learning process,consequence.The global optimal solution can be found with a higher probability,the stability and robustness are also enhanced.(2)According to the fault mode distribution of aeroengines,the recognition method based on distance is designed,and the optimization performance of MELPSO can find a single optimized classification point for each class.The optimized classification point meet the three-optimization objectives,including the shorter intra-class distance,longer inter-class distance,and maximum classification accuracy of training samples,by which the unknown fault modes can be recognized by the nearest distance to the optimized classification points.However,the recognition principle based on single optimized points cannot recognize the data sets with nonlinear separatrix accurately,some appropriate points for each class need to be determined rather than only finding a single optimized point,thus,and an adaptive optimized classification strategy is proposed.Meanwhile,the concept of priority levels is proposed for the multi-objective optimizations,and it can ensure that the prime objective is achieved firstly.The improved recognition strategy has a better performance for recognizing the sets with nonlinear separatrix.(3)Massive monitoring data can accurately describe the operation states of the mechanical equipment,but a large number of data are the same or alike,which reflect the same fault model.However,with the increase of the sample size,the computational efficiency can be reduced.Accordingly,the granular computing based graph partition is proposed into compress fault data,the granularities are generated by extracting and partitioning the trivial subgraph set and complete subrgraph set.Meanwhile,in order to overcome the influence of attributes with different dimensions,the dimensionless similarity is proposed.This method can compress the data to meet the similarity threshold,and it can also reduce the sample size and computational demand.Moreover,this method can guarantee the spatial distribution of compressed data is closer to that of original data,so that the influence of data information loss on the accuracy of fault diagnosis can be decreased(4)The main performance parameters of aero-engine are main indexes to judge the health state,and they are easy to be influenced by factors such as operation environment,mechanical factors,and flight condition and so on.They always have the characteristics of non-linearity,non-closure,and uncertainty.By the means of the global optimization capability of MELPSO,a function with nonlinear mapping can be worked out,and then the prognostics method based on time series is constructed to predict the known data.Furthermore,in order to the defect of the time series with output deviation,the impact factors on the performance parameters are determined,and a prognostics method based on impact factor is further constructed,it can improve the accuracy of prediction and the stability of output.