Component fatigue life is a major concern in the design of gas turbine Auxiliary Power Units as it directly influences the reliability and life cycle cost of the end product. Accordingly, there is heavy emphasis placed upon designing components which safely maximize their fatigue life. Typical industry practice for managing fatigue has relied on what is commonly referred to as the “safe life approach” in which retirement lives are analytically determined for components and hardware is removed from service before fatigue related failures can occur.
The safe life approach is deterministic in nature since stress analysis results based on minimum geometry, material properties and maximum load are used to set a single life for the component. However, service experience shows that fatigue failures can occur before service life and that actual service lives are distributed over a large range of values as a result of variables not accounted for by deterministic methods.
In order to better achieve the goal of minimizing product life cycle costs while recognizing the variable nature of fatigue lives, Sundstrand Aerospace (SA) has developed a Life Management Plan (LMP) which includes probabilistic methods to augment the company’s standard safe life methodology.
Sundstrand’s LMP builds on the safe life methodology by using statistical distributions along with Monte Carlo simulations to predict initial component cracking rates. These initial predictions are used to guide an inspection program which provides actual cracking data. As this inspection data base grows, the initial simulation is modified to include the inspection data and the predicted failure rates are updated. This provides Sundstrand with a tool to manage failure risk in the field as well as to provide early warning of negative trends.
This paper will discuss how Sundstrand’s LMP was devised and implemented as well as what lessons have been learned and what changes are planned for future incorporation. A case history will be cited to illustrate how the LMP has worked, comparing predictions to actual experience.