Fault Detection Scheme Optimization And Health Prediction For Hidden-state Equipment Based On Degradation Characteristics

Along with the rapid trend in multi-functional,high-integration and intelligent technologies,most modern equipment present hidden-state characteristics,which means the degradation signal cannot be fully observed.To improve the stability of the equipment functioning and efficiency of the production,investigations of the equipment related to degradation characteristics,degradation modeling,fault detection and health prediction need to be done.To this aim,the dissertation systematically studies the whole fault detection and health prediction program of the equipment on three-state,multi-state and continuousstate characteristics from degradation characteristics to fault detection and health prediction.This work provides a comprehensive study on hidden-state equipment.The health measures predicted in this work includes mean remaining useful life,remaining useful life distribution and conditional reliability,which correspond to three important indices: residual functioning time of equipment,future functioning condition of equipment and probability of future functioning of equipment,respectively.The content and main contributions of this work are as follows:(1)The degradation process of equipment is categorized as three-state mode,multistate mode and continuous-sate mode.The deissertation conducts the research from the three-state mode to continuous-state mode sequentially.(2)Fault detection and health prediction for three-state equipment with multivariate observations.The approach presented here is based on hidden semi-Markov modeling using the optimal Bayesian control and estimation technique.Equipment condition is modeled using a continuous time semi-Markov chain with three states,i.e.,unobservable healthy state 1,unobservable warning state 2 and observable failure state 3.Model parameter estimates are calculated using the expectation-maximization algorithm with explicit update formulas.The optimal control policy for the three-state model is represented by a Bayesian control chart for a multivariate observation process,and the decision variables are optimized by an efficient algorithm.Formulas for the health prediction of the equipment including conditional reliability,remaining useful life distribution and mean remaining useful life are also derived based on Bayesian approach.A comparison with other approaches is also given,which illustrates the effectiveness of the proposed approach.(3)Fault detection and health prediction for three-state equipment under competing risks of dependent failure with multivariate observations.The equipment degradation process is modeled using a continuous time stochastic process with three states,i.e.unobservable healthy state 1,unobservable warning state 2 and observable failure state 3.To model the dependence of two failure modes,we assume that the joint distribution of the time to catastrophic failure and sojourn time in healthy state follows Marshal-Olkin bivariate exponential distribution.Model parameter estimates are calculated using the expectation-maximization algorithm with explicit update formulas.To avoid unnecessary sampling cost and to effectively detect impending failure,a two-level control policy,in which longer sampling interval is applied for healthier state and shorter sampling interval is used in severe degradation,is proposed in Bayesian control chart framework for a multivariate observation process under dependent failure modes.The decision variables are optimized by an efficient algorithm.Formulas for the conditional reliability,remaining useful life distribution and mean remaining useful life are also derived based on the Bayesian approach.A comparison with other approaches is also given,which illustrates the effectiveness of the proposed approach.(4)Fault detection and health prediction for multi-state equipment with multivariate observations.The approach uses proportional hazards model incorporating the multi-state continuous time Markov chain with non-constant degradation to model the hazard rate of the equipment.An integrated model for health prediction and fault detection is proposed,where a matrix-based approximation method is employed to compute health measures of the machine,such as condition reliability,mean residual life,residual life distribution,and a semi-Markov decision process algorithm with matrix form is developed to optimize the two-level fault detection policy.The matrix-based approximation method has addressed the long-standing computing problem of proportional hazards model,and the health measures and decision variables in the integrated model for health prediction and fault detection can be obtained by simply operating the matrices.A comparison with other methods is also given,which illustrates the effectiveness of the proposed approach.(5)Fault detection and health prediction for continous-state equipment.The approach uses proportional hazards model incorporating the Gamma process with non-constant degradation to model the hazard rate of the equipment.Then,an integrated model for health prediction and fault detection based on Gamma process is proposed.New derived matrixbased approximation method is applied to obtain the health measures and decision variables.A comparison with other models is also given,which illustrates the effectiveness of the proposed approach.Although we analyze degradation data coming from the CNC equipment,our method can be applied to all types of multivariate data such as oil data from oil pumps and vibration data from wind turbines.In addition,the integrated model for health prediction and fault detection presented in this work can be applied a much wider class of stochastic degradation process.For example,our model can be applied to Wiener process model.