Preventive maintenance planning based on Weibull distribution models: The impact of the random scatter behaviour of the threshold parameter

Abstract

The development of complex technical products also includes the planning of maintenance strategies and maintenance activities regarding the product use phase in field. The planning of maintenance packages within a preventive maintenance strategy is based on prototype testing data, expert knowledge and the statistical analyses of failure data of the previous product generation. The statistical analyses of a failure behaviour contains a statistical distribution model: The use of a three-parameter Weibull distribution model is common in reliability analytics. The Weibull parameters are threshold, scale and shape (hint regarding early failure period, random failure or operation time related failure). The parameter calculation can be done by state-of-the-art estimators (e.g. MLE). The problem is the random scatter range of the estimated parameters, which depends on the underlying data base. Therefore, calculation of parameters confidence interval respectively random dispersion range is obligatory. Especially the threshold parameter gives a hint regarding the upraising risk within a product fleet based on a certain failure behaviour. Consequently, the threshold parameter of a Weibull model, which describes an expected failure behaviour, is essential for the scheduling of maintenance packages regarding the subsequently following product generation to avoid an upcoming risk. This paper shows an empirical study regarding the random scatter behaviour and the characteristics of the threshold parameter and the impact on maintenance package planning within certain maintenance strategies. Furthermore the interdependences to the shape and characteristic life span parameters is worked out. The results of the empirical study are shown within an automotive case study.