The word “Random” might confuse us when we refer to Nowlan and Heap’s failure pattern B that describes “age independent failure”. When using the word random applied to an event, we tend to infer that the event is unpredictable. Randomness suggests a non-coherence, such that there is no intelligible pattern or combination that may be discerned. But maintenance and reliability engineers today believe such is not really the case with respect to asset failure events.
The field of Condition Based Maintenance (CBM) would appear to negate the idea of randomness applied to equipment reliability, since its premise is that of predictive maintenance. How exactly does CBM reduce or reconcile randomness with predictibility? Let us assume that Weibull’s equation for Reliability
where t is age, β and η are parameters, describes the relationship between a part’s (or failure mode’s) age and its reliability, its probability of surviving to that age. By years of experience maintenance engineers know that age alone is usually insufficient information upon which to judge an item’s ability to survive the next time interval, say one month or 10000 widgets, when asking the question at the current moment in time. All maintenance planners and engineers recognize that additional data, called “conditional data varaiables” are required, that would indicate the part’s current state of health. But they don’t know how, to incorporate that additional information into the reliability calculation. They don’t even have a systematic method by which to judge which condition data variables contain predictive content. Nor can they evaluate the degree to which condition data would mitigate the “randomness” that interferes with practical failure prediction.
This article will introduce a simple method for systematizing CBM prediction. Instinctively we know that while the age of an item while not a determinative factor certainly influences the propensity of a part to fail. We will explore the questions: What is the relative influence of age on failure? And, what is the relative influence of each of the other relevant condition variables on failure probability?