Abstract: Reliability data analysis often faces two types of heavily censored data. One type of data is randomly censored and the other type of data is generated by preventive maintenance actions. It is challenging to construct reliability models using these two types of censored data. This research introduces the mechanism of generating heavily censored data and recently proposed parameter estimation methods through numerical simulation analysis. Results of numerical simulation show that recently proposed parameter estimation approaches are more effective than the classical methods when processing heavily censored data. After that, research gaps of processing preventive maintenance censored data are identified. Simulation is used to analyze mechanisms of generating two typical preventive maintenance censored data. Three new parameter estimation methods based on the independent competing risk model, the proportional residual life model, and the mixed assumption model are then developed. Both simulation and real-world datasets are analyzed to validate the importance of selecting appropriate assumption of modelling. Analysis results show that parameter estimation methods developed for highly randomly censored data can deliver relatively accurate estimates; the parameter estimation methods for preventive maintenance censored data are efficient, and the estimation method selection depends on the data generation mechanism or data characteristics analysis.

Key words: life data analysis, highly censored data, randomly censored data, preventive maintenance censored data, parameter estimation