非均匀监测条件下滚动轴承剩余寿命预测方法

A Remaining Useful Life Prediction Approach with Nonuniform Monitoring Conditions for Rolling Bearings

退化过程的建模依赖时间和幅值的离散化,现有的退化建模方法大多基于时间均匀离散的假设。然而,在设备的日常运维中,由于传感器故障或操作人员失误等因素,可用监测数据会出现非均匀分布的情况(如旋转机械的关键部件滚动轴承的状态监测数据),导致退化模型进行参数更新和剩余寿命预测时存在额外偏差。针对该问题,提出一种非均匀监测条件下的滚动轴承剩余寿命预测方法。首先,构造监测间隔相关的布朗运动项,以精确刻画非等间隔退化模型中随机参数的时间不确定性。在此基础上,提出了一种基于期望最大化算法和平方根容积卡尔曼滤波的非等间隔退化模型的参数动态推断方法,实现了滚动轴承退化状态与剩余寿命的自适应估计。通过滚动轴承全寿命试验实例,验证所提出剩余寿命预测方法的有效性。结果表明,在非均匀监测条件下,与其他指数退化模型相比,所提出的方法具有更高的预测准确度和更优的拟合性能。

The modeling of degradation processes relies on the discretization of time and amplitude,and most of the existing degradation modeling methods are based on the assumption of uniform dispersion of time.However,in the daily operation and maintenance of the equipment,due to factors such as sensor failures or operator errors,the available monitoring data can be nonuniform(such as condition monitoring data of rolling bearings as key components of rotating machinery),which results in additional deviations in the degradation model when updating parameters and predicting the remaining life(RUL).Aiming at this problem,a RUL prediction approach with nonuniform monitoring conditions for rolling bearings is proposed.Firstly,the Brownian motion term associated with monitoring intervals is constructed to accurately characterize the temporal variability of random parameters of the degradation model with nonuniform intervals.Then,based on the expectation maximization(EM)algorithm and the square-root cubature Kalman filter(SCKF),a dynamic parameter inference method of the degradation model with nonuniform intervals is proposed,and the degradation state and RUL adaptive estimation of rolling bearings are accomplished.The effectiveness of the proposed approach for predicting the RUL is verified by means of rolling bearing full life test examples.The results show that under nonuniform monitoring conditions,the proposed approach obtains higher prediction accuracy and better fitting performance compared with other exponential degradation models.