基于振动样本民航发动机故障诊断方法及应用分析-civil aviation engine fault diagnosis method and application analysis base on vibration sample.docxVIP

问题，提出了一种基于等高线图和交叉验证的参数优化算法，利用等高线图强大的全局搜索能力和可并行性，实现了对支持向量机分类性能有重要影响的参数进行快速有效的搜索，使支持向量机参数选择问题得到解决。精确的振动趋势预测对发动机的使用至关重要。以神经网络为代表的人工智能方法由于网络结构设计上的缺陷，使网络的推广能力受到制约，预测结果与真实值之间存在较大误差。过程神经网络充分考虑了输入过程中的时间累积，在网络模型中增加了反映时间累积过程的参数，其输出依赖于输入过程的时间累积而不单纯是单点式的瞬时对应。基于过程神经网络理论，本文实现了对民航发动机短期趋势的预测，并给出具体应用方案。针对工程实际的中长期预测问题，基于光滑化方法改进了支持向量机回归算法，将二次优化问题从有约束转换为无约束优化问题，采用历史数据的振动样本验证了该方法，结果证明即使预测较长区间的振动值，预测结果仍具有相当高的精度。关键词：振动样本；民航发动机；故障诊断；视情维修；独立分量分析；核主元分析；支持向量机回归AbstractStructureofaeroengineiscomplicatedandtheworkingconditionisverybad,whicheasilycausesvariousmechanicalfailures.Therefore,accuratefailurediagnosisandtimelytroubleshootingareofgreatsignificancetoairsafety.Traditionalperiodicmaintenancemethodisoflowefficiencyandhighcostwhileconditionbasedmaintenancehastheadvantagesofconvenientoperation,highdiagnosisrateandlowcostaswellastheadvantagethatseriouspotentialaccidentscanbetimelynoticed,whichis,therefore,applicableforaeroenginemaintenancesupport.Implementationofconditionbasedmaintenancerequiresthatthefailurediagnosissystemofcivilaero-enginehasstrongcapacitiesforhealthstatusanalysisandforselfdiagnosisprediction.Failurediagnosismethodsrelevanttoconditionbasedmaintenanceforaeroengineiscarriedoutresearchbasedonthevibrationsamplecollectedbytheairline.Coaxialconfigurationwithmanyrotorsinaeroenginedeterminesthatthevibrationsignalmeasuredbytheenginesensorismixedbymanyvibrationsignals,themixingmethodbeingunknown.Itiscrucialforfailurediagnosistoseparateindependentvibrationsignalofeachrevolvingshaftandfilteruselessinterference.Therefore,thebasicmethodforseparatingthevibrationsignalsoftheengineisresearchedinthispaperinthefirstplace.Specifictothedisadvantagesoftraditionalseparationmethod,itisproposedtoapplynaturalgradientlearningtotherecursivealgorithmofindependentcomponentanalysisobjectivefunction.Andalso,thisapplicationisadoptedforseparatingandprocessingtypicalmechanicalsimulationsignalsandengineboxsamplesignals,givingspecificseparationsteps;besides,thefeasibilityandstabilityofthisalgorithminuseareanal