基于相关向量机的发动机剩余寿命预测

Remaining Useful Life Prediction for Aeroengine Based on Relevance Vector Machine

针对民航发动机寿命预测中监测参数较多筛选困难的问题,提出一种基于信息融合与相关向量机(relevance vector machine,RVM)的发动机剩余寿命(remaining useful life,RUL)预测方法。首先通过核主元分析(kernel principle component analysis,KPCA)方法从发动机多维监测数据中提取退化特征信息;然后利用非线性模型将主元序列融合成反映发动机退化趋势的健康指数序列;最后采用相关向量机以历史失效数据为训练样本建立预测模型,对现有的发动机健康指数序列进行外推预测得到当前样本的寿命预测值。通过美国国家航空航天局(National Aeronautics and Space Administration,NASA) Ames研究中心公开的涡轮风扇发动机仿真数据验证了该方法的有效性,其预测性能优于常用的支持向量机(support vector machine,SVM)模型和过程神经网络模型。

Aiming at the difficulty of screening the monitoring parameters in civil aero-engine life prediction, a method based on information fusion and relevance vector machine(RVM) was proposed for predicting the remaining useful life(RUL) of a civil aero-engine. First, the kernel principle component analysis(KPCA) in feature extraction information from the engine multi-dimensional monitoring data was applied. Secondly, a non-linear model to fuse the principal component into a series of health index was adapted, which reflected the engine degradation tendency. Finally, the RVM was used to build a prediction model with the training samples of historical failure data, after which the RUL prediction values of the current samples were obtained by extrapolating the existing engine health index series. The proposed method was evaluated by using the turbo engine degradation simulation data published by the Ames Research Center of the National Aeronautics and Space Administration(NASA). The results show the proposed method is superiority over those of support vector machine(SVM) model and process neural network model.