Statistical Modeling of High Cycle Fatigue with Censored Data

Characterization and prognosis of materials subjected to very high cycle fatigue is an important aspect of life cycle design and management for structural components. This procedure is even more critical for components in devices that are implanted into humans because of the lack of opportunities for periodic inspection and the severe consequences of failure. The primary structural component of a commercially available heart valve consists of highly-resilient, cold-worked Elgiloy wire, a cobalt-chromium based alloy. In order to be assured of sufficiently high reliability for the component, an extensive high cycle fatigue testing program was conducted. According to one medical device regulatory guidance, the structural component is expected to have a fatigue life in excess of 600 million cycles under worst case physiologic conditions. Using basic principles for fatigue life evaluation, a constant lifeline for mean stress versus stress amplitude was statistically established. The fundamental model for the analysis was a generalized Weibull distribution for which the parameters were stress-dependent. The model was optimized using maximum likelihood methods and confirmed by chi-squared goodness-of-fit evaluation. The methodology used for this investigation, well established in fatigue engineering, was shown to be equally effective when applied to life science problems, with the potential for benefit to patients.