应用交互式多模型算法的设备剩余寿命预测

针对目前异常检测与剩余寿命预测联合研究中存在的问题,基于交互式多模型算法,提出了一种考虑退化模式动态转移的设备剩余寿命预测模型。首先根据模式动态转移的状态空间模型对设备进行退化建模,然后使用IMM算法对设备的隐含退化状态与退化模式后验概率进行联合估计,根据后验概率判别异常点,并采用期望最大化(Expectation Maximum,EM)算法对模型参数进行在线估计与更新,最终实现设备的实时剩余寿命预测。数值分析结果表明:该模型能够准确地检测异常点、降低剩余寿命概率分布的不确定性与提高剩余寿命的预测精度,为实现设备的视情维修提供决策依据。

A Residual Useful Lifetime Prediction Based on Interacting Multiple Model Algorithm

Aimed at the problems that at present the anomaly detection and the residual useful lifetime (RUL) prediction exist in the integrated study, a RUL prediction model, i.e. a dynamic transition of the degradation model,is presented based on IMM algorithm. The presented model overcomes the disadvantages of the tradition models that the anomaly points at single stage are on no consideration by the prediction model, and can predict the RUL in real time. The numeric results show that the presented model can detect anomaly accurately, can reduce the uncertainty of RUL probability distribution, can improve the precision in RUL prediction, and can provide a basis for a decision making in completing maintenance work in accordance with the concret conditions.