Abstract: To solve the failure prediction problem in small sample machincal equipment,the issue of degradation-based failure time estimation for mechanical device was proposed.Since the degradation process was greatly affected by working conditions,the concept of local degradation rule was proposed and represented by the exponential function,and all the potential degradation patterns could be obtained consequently.The fitted exponential curves were used to produce samples of local failure time,and the local failure time observations were obtained at each time.During the lifetime of mechanical device,the extreme degradation patterns were rare while the moderate ones were frequent,thus the local failure times were assumed to be independently from a normal prior distribution.Parametric empirical Bayes technique was adopted to estimate both the prior and posterior distribution.The posterior distribution of last observation was treated as the global failure time estimation.The proposed approach was validated in a simulation case and a precision case of numerical control machine tool to show the better effectiveness.

Key words: mechanical device, performance degradation, failure time estimation, exponential function, parametric empirical Bayes

摘要： 为解决小样本机械设备的故障预测难题,提出一种基于性能退化的机械设备寿命预测方法。考虑机械设备退化过程受工况影响,提出局部退化规律的概念,并利用指数回归进行拟合表征,从而获得所有潜在的退化规律。通过表征局部退化规律的指数函数,计算局部寿命样本,进而获得各个时刻的局部寿命观测。鉴于机械设备在寿命周期的极限工况对应的剧烈退化少、中等工况对应的中等退化多,假设局部寿命来自某正态分布,并利用参数经验贝叶斯方法实现先验分布和后验分布的估计,全局寿命预测为最新观测对应的后验分布。通过仿真案例和数控机床的精度退化案例,验证了该方法具有良好的准确性和有效性。

关键词: 机械设备, 性能退化, 寿命预测, 指数函数, 参数经验贝叶斯