Article ID: | iaor19921700 |
Country: | United Kingdom |
Volume: | 39 |
Start Page Number: | 115 |
End Page Number: | 123 |
Publication Date: | Jan 1990 |
Journal: | Applied Statistics |
Authors: | Follmann Dean A. |
This paper presents methodology for analysing failures of machines that are repeatedly turned on and off and at some point fail. Because a machine is at risk to fail both when it is on and when it is off, different models for failure are used for each of these periods. During on periods, a Weibull regression model is assumed with time since switched on as time index. The Weibull shape parameter can describe ‘burn-in’ or a transient harmful effect following switching on. Failures following off periods are modelled with a logit regression model. For both the Weibull and the logit models, covariates are used to describe historical usage and other factors associated with failure. An important issue for such machines is how usage itself affects failure. The models can be used to predict the chance of failure under different usage scenarios. In this way, less harmful usages can be recommended. An example, using a radar, is provided and generalizations discussed.