Nonparametric Regression-Based Failure Rate Model for Electric Power Equipment Using Lifecycle Data

作者：Qiu, Jian; Wang, Huifang^*; Lin, Dongyang; He, Benteng; Zhao, Wanfang; Xu, Wei

来源：IEEE Transactions on Smart Grid, 2015, 6(2): 955-964.

DOI：10.1109/TSG.2015.2388784

• Asset management
• data mining
• failure rate
• health index (HI)
• lifecycle data
• nonparametric regression
• proportional hazards model (PHM)

In order to analyze the fault trends more accurately, a failure rate model appropriate for general electric power equipment is established based on a nonparametric regression method, improved from stratified proportional hazards model (PHM), which can make maximum use of equipment lifecycle data as the covariates, including manufacturer, service age, location, maintainer, health index, etc. All of covariates are represented in the hierarchy process of equipment health condition, which is beneficial for processing and classifying the lifecycle data into multitype recurrent events quantitatively. Meanwhile, based on new definitions of single health cycle and time between events, recurrent inspecting events distributed with martingale process can correspond with event-specific failure function during equipment lifecycle. On this occasion, more inspecting events can be utilized in a complete cycle to predict potential risk and assess equipment health condition. Then, stratified nonparametric PHM is employed to build the multitype recurrent events-specific failure model appropriate for competing risk problem toward interval censored. Lastly, the example in terms of transformers demonstrates the modeling procedure. Results show the well asymptotic property and goodness-of-fit tested by both of graphical and analytical methods. Compared with existing failure models, such as age-based or CBF model, this improved nonparametric regression model can mine lifecycle data acquisition from asset management system, depict the failure trend accurately considering both individual and group features, and lay the foundation for health prognosis, maintenance optimization, and asset management in power grid.

• 出版日期2015-3
• 单位浙江大学