Degradation Tracking And Prognostics Of Rotary Machine Components Using Cross Recurrence Analysis

In modern industry,rotary machines,with bearings and gears as their key rotary components,play an important role in daily living and production.Since working environments of them are tough,such as heavy load and high speed,machine degradation and failures which can lead to economic losses and disastrous accidents often occur.Therefore,the development on condition based maintenance(CBM)of rotary machines has received considerable attention in recent years.As compared to traditional approaches focusing on breakdown maintenance and preventive maintenance,which only conduct fault identification and classification after failures occur or within periodic machine halt,expensive downtime and maintenance costs always lead to low productivity.The strategy that combines degradation trend tracking and prognostics involved in CBM can overcome those shortcomings and help achieve maximum productivity.For existing prognostic approaches,studies on nonlinear degradation monitoring,real-time online ?first failure point‘ prediction,and deep learning based remaining useful life(RUL)prediction are still limited.Therefore,based on the phase space reconstruction theory,the degradation monitoring and fault prediction of rotary machine components are investigated in this thesis.The main research contents can be summarized as follows.(1)Systematically studying the phase space reconstruction theory which based on Takens embedding law,and investigating the selection methods of two important parameters,time delay and embedding dimension.The research shows that phase space reconstruction using mutual information(MI)and false nearest neighbors(FNNs)for embedding parameter selection can reconstruct the phase space from a one-dimension measurement series and well preserve the basic structures and dynamic properties of the original system.(2)Investigating the cross recurrence quantitative analysis(CRQA)method which based on the phase space reconstruction,and extracting three CRQA variables to diagnose rotary component faults.The comparison experiments with different working condition and fault types indicate that CRQA has better performance compared to traditional recurrence quantitative analysis(RQA)in both early failure detection and robustness.(3)Proposing an improved CRQA method for bearing degradation evaluation.CRQA entropy is extracted as the health index(HI)to establish mechanical degradation trend.Due to the fact that traditional process for calculating the hyper parameters of CRQA requires a large amount of time to reconstruct phase space,this step is replaced by calculating divergence rate of each signal sample to cut down the time consumption.After establishing degradation trend based on the extracted variable,nonlinear auto-regressive neural network(NARNN)is developed to predict future degradation trend.Furthermore,a new failure detection method is investigated to indicate the ?first failure point‘ during degradation tracking.The method applies temperature signals as auxiliary information to calculate adaptive threshold and classify extracted features into different health status.Experiments on bearing vibration signals have verified that the improved CRQA method can reduce time consumption by more than 90%.In addition,defect characteristic frequencies extracted using wavelet analysis have validated the detection accuracy.(4)Proposing an improved gated recurrent unit(GRU)network for rotary machine RUL prediction.Utilizing degradation curves established by improved CRQA recurrence rate(RR)and percent determinism(DET)as the inputs of GRU.Considering risks caused by early detection are much smaller than caused by late prediction,penalty factors are introduced to modify the traditional mean square error(MSE)loss function,and enforcing the network to learn more information about early detection.