[4F3] Confidence bands for cumulative event number evolution curves

D Razafindrakoto¹, J Lacaille², J Djaballah² and J Bertrand²
¹University of Paris, France
²Safran Aircraft Engines, France 

For a fleet of engines to remain in service, the cumulative number of failures must be kept under a threshold set by the authorities. If the cumulative number of said event reaches a critical level during a fleet’s service, a statistical study is conducted to predict the evolution of that number. This study integrates the history of the fleet, a scenario describing the occurrence of the event and an inspection strategy put in place to prevent the occurrence of said event. This results in a predictor capable of predicting possible curves describing the evolution of the cumulative number of events under these specifications. To interpret this predictor, only the average curve is estimated using a sample of predicted curves. Although the average curve is informative, it alone does not reflect the variability of possible curves. The aim of this work is to describe a generic method for computing bands that ‘contain’ the generated curves with a probability higher than a specified threshold. An important aspect of these bands is that they ensure these statistical guarantees using only a finite number of predicted curves.