失效型式识别和失效时间预测的方法与模型研究

Various complex engineering systems have been widely used in various industrial sectors. To make these systems operate safely, reliably and economically, there is a need to understand the failure pattern of a failure process (e.g., monotonically increasing, decreasing and bathtub curve patterns), monitor the degradation processes of key components and predict their failure times so as to appropriately select the operation and maintenance strategy and optimize preventive maintenance decisions. Therefore, two key issues to be studied are identification of failure pattern and accurate prediction of failure time. For the pattern identification problem, we will develop two types of flexible models for fitting the mean value function with complex shape and infer the corresponding failure pattern from the fitted model. We will study the prediction problem from two aspects. On one hand, we will introduce a weighted model fitting method. This method stresses the influence of the recent observations on the prediction accuracy through assigning large weights to these observations. On the other hand, we will explicitly consider the influence of future operational condition and environment factors on the failure or degradation process through development and use of reliability models with covariates so as to further improve the prediction accuracy. The research outcomes will provide new modeling methods and mathematical models for improving the safety, operational reliability and economy of industrial systems.

各种复杂工程系统已被广泛地应用于国民经济各部门。为确保这些系统安全、可靠、经济地运行，需要理解失效过程的失效型式（如单调地增的、减的和浴盆曲线型式），监测关键零部件的退化状况，预测其失效时间，从而选择恰当的运维策略并优化其预防维修决策。为此，需要研究失效型式识别和失效时间精准预测这两个关键问题。对于可修系统的失效型式识别问题，我们将开发两类富于弹性的模型供拟合具有复杂形状的均值函数，并由此导出对应的失效型式。我们将从两个方面研究退化过程失效时间的精准预测问题。一方面，我们将引入一类基于加权的模型拟合方法。该方法通过给新近的观察值赋以更大的权重来强调它们对预测精度的影响。另一方面，通过开发和使用含有协变量的模型，我们将明确地考虑未来的运行条件和环境因素对失效或退化过程的影响，从而进一步改进预测精度。项目研究结果将为显著地改善工程系统的安全性、运行可靠性和经济性提供新的建模方法和数学模型。

本项目经过四年的研究，已经圆满完成了预期的研究计划。研究主要围绕研究复杂工程系统的失效型式识别和失效时间精准预测这两个关键问题。我们从两个方面研究了退化过程失效时间的精准预测问题，一方面，我们引入了一类基于加权的模型拟合方法，该方法通过给新近的观察值赋以更大的权重来强调它们对预测精度的影响。另一方面，通过开发和使用含有协变量的模型，明确地考虑到未来的运行条件和环境因素对失效或退化过程的影响，从而进一步改进了预测精度。截至结题之日，共发表论文43篇，其中国际杂志15篇，国内杂志6篇，国际会议论文20篇，国内会议论文2篇，论文被SCI收录11篇，EI收录6篇，多篇论文受到国际同行的注意。