Research On The Methods Of Health Assessment And Prognostics For Aero Engine On The Module Level

With the rapid development of civil aviation industry, the civil aircraft engine safety andbenefit/cost issue become an urgent problem. The next-generation air transportation system isproposed with the targets of “checkpoints shifted front for safety” and “continued safety”. Thedomestically produced civil and military aero engine has a need on the Engine Health Management(EHM) technologies. So there is an urgent need to strengthen the research on the key technologies toimprove the EHM capability of the civil aircraft engines and to support the development anddeployment of the EHM system for the domestically produced aero engines. Based on the literaturesurvey and a full study of the concept and technical architecture of the EHM, considering theproblems in the practice, the main topics of the thesis focus on the gas path components conditionmonitoring, health status assessment as well as remaing useful life prognostics methods. On one sideit can provide technologies and methods for the improvement of the civil aircraft engine EHMcapability, on the other side it can provide experiences and mature technologies for the domesticallyproduced engine’s EHM system. The main topics of the thesis are as follows:(1) Gas path components condition monitoring:In order to detect the anomaly as soon as possible from the delta parameter parameters with largenoise and scatters, which maybe cover the fault signature, then to trigger a timely warning at early stageof the fault, a Bayesian factor-based method is proposed to monitor and analyze parameter delta series.Since the “generic” baseline model embedded in the OEM software cannot capture the characteristicsof the individual engine under the real operating conditions, the Multi State Estimation Technique(MSET) is proposed to build the individual baseline model for the specific engine, which can capturethe characteristics of the individual engine. Based on the individual baseline model, more accurate deltadata can be obtained which can advance the fault warning in time. In the application of the new gaspath condition monitoring techniques, the Exhaust Gas Electrostatic Monitoring Signal (EGEMS) isthought as a new gas path performance parameter, and two EGEMS baseline model building methodsare proposed. One is based on one parameter-the fuel flow rate, and the other one is based on multiparameters which are correlated using the MSET. Based on the developed baseline model, the deltabetween the real RMS value of the EGEMS and baseline value is monitored in real time to monitor thegas path component condition and to trigger a warning once some fault occures.(2) Gas path components health assessment:A study of in-field engine components health assessment techniques based on adaptive engineperformance model is carried out. Considering the fact that less measured performance parameters thanthe health parameters to estimate and the influence of the measurement noise, the health parametersestimation problem is changed into an optimization problem. An exclusive method based faultdetection and assessment framework is proposed, in which the healthy module is excluded one-by-one to reduce the number of the health parameters to estimate until they are less than the measuredparameters. Considering the fact that the gradual performance deterioration in field is inevitable, ahealth parameter estimation framework is proposed to incorporate the gradual performancedeterioration information. In this framework, the gradual performance deterioration is tracked to get thedegradation state of each module before the fault, then the information is incorporated when isolatingand assessment the fault to improve the health assessment results. A information fusion based gas pathanalysis framework is proposed to tackle the issue of lack of enough measured gas path parameters. Inthis framework, an information fusion mechanism based on the Bayesian network is developed toincorporate the the diagnosis information from multi sources, in which the qualitative information isincorporated by the fault mode prior probability table and the quantitative information is incorporatedby the prior distribution of the health parameter, then the information is fused using Bayesian rules toimprove the accuracy and precision of the estimation results of the health parameters.(3) Modeling methods for remaining useful life prognostics:The methods on remaining useful life prognosis for individual system under real operatingconditions are discussed in depth. The state space-based degradation model combined with Bayesianstate estimation theory is proposed for system remaining useful life prognostics and in-servicereliability estimation. The EGTM parameter is used as a degradation parameter to quantify thedegradation state of the engine. Then linear Gaussian state space model is adapted to describe thedegradation trajectory based on the observed EGTM data, and then the conjugate Bayesian inferenceis carried out to estimate the degradation state and further to make a prediction of the failure time. Thestate space based degradation model differentiates the noisy observation from the true degradationstate, which is closer to the actual case. The state space degradation model does not need to makestationarity assumption, so it can effectively manage the situation when there is a sudden change inthe health state due to fault or maintenance. Considering the problems that a single parameter cannotcharacterize the health state of a complex system, a fusion mechanism is developed to fuse multiparameters to get a health index to characterize the health state of the gas path component, based onwhich a state space degradation model is established to describe the degradation path and predict theremaining useful life. For the crack growth failure of the critical components, a fusion framework isproposed to integrate the damage monitoring data and physics of failure mechanism for remaininguseful life prediction. A state space based crack growth model is developed based the Paris crackgrowth model, then the damage monitoring data is integrated using the Bayesian rule. By integratingthe monitoring data the prognostics uncertainty can be continued reduced.