基于油液光谱分析和粒子滤波的发动机剩余寿命预测研究

Research on Engine Remaining Useful Life Prediction Based on Oil Spectrum Analysis and Particle Filtering

油液光谱分析是机械磨损状态监测、故障诊断与故障预测的重要技术,基于光谱数据的机械状态剩余寿命预测有利于实现机械系统的最优维修决策。由于机械设备越来越复杂,其健康状态的退化过程很难用线性模型来表示,而粒子滤波(particle filter,PF)对非线性非高斯系统的处理能力,与经典Kalman滤波相比具有明显的优势,文章将PF预测方法运用于光谱分析,提出了基于PF和油液光谱分析技术的设备剩余寿命预测方法。在预测模型中实现了根据设备后验分布的估计值预测其先验分布概率,建立了基于PF的多步向前长期预测模型。最后,对某发动机实际的光谱分析数据进行了预测和分析,并与传统Kalman滤波方法的预测结果进行了比较,结果充分表明了本方法的有效性和优越性。

The spectrometric oil analysis（SOA）is an important technique for machine state monitoring,fault diagnosis and prognosis,and SOA based remaining useful life（RUL）prediction has an advantage of finding out the optimal maintenance strategy for machine system.Because the complexity of machine system,its health state degradation process can’t be simply characterized by linear model,while particle filtering（PF）possesses obvious advantages over traditional Kalman filtering for dealing nonlinear and non-Gaussian system,the PF approach was applied to state forecasting by SOA,and the RUL prediction technique based on SOA and PF algorithm is proposed.In the prediction model,according to the estimating result of system’s posterior probability,its prior probability distribution is realized,and the multi-step ahead prediction model based on PF algorithm is established.Finally,the practical SOA data of some engine was analyzed and forecasted by the above method,and the forecasting result was compared with that of traditional Kalman filtering method.The result fully shows the superiority and effectivity of the new method.