Numerous reliability models and testing practices are based on operational profiles. Typically, a single operational profile is used to represent the usage of a system with the assumption that a homogeneous customer base executes the system. However, if the customer base is heterogeneous, estimates computed on a single operational profile may be inaccurate. A single operational profile does not reflect the diverse customer patterns and it only "averages" the usage of the system, obscuring the real information about the operations probabilities. Decisions made on these estimates are likely to be biased and of limited usefulness. This paper presents a refinement methodology for the generation of more accurate operational profiles that truly represent the diverse customer usage patterns. Clustering analysis supports the refinement methodology for identifying groups of customers with similar characteristics. Empirical stopping rules and validation procedures complete the refining methodology. A complete example of the methodology is presented on a large application. The example evidences the different perspective and accuracy that can be obtained through this refining methodology.
展开▼
