DETECTING TRENDS AND PATTERNS IN RELIABILITY DATA OVER TIME USING EXPONENTIALLY WEIGHTED MOVING-AVERAGES

A simple, easy-to-use graphical method is presented for use in determi ning if there is any statistically significant trend or pattern over t ime in an underlying Poisson event rate of occurrence or binomial fail ure on demand probability. The method is based on the combined use of both an exponentially weighted moving-average (EWMA) and a Shewhart ch art. Two nuclear power plant examples are introduced and used to illus trate the method. The false alarm probability and power when using the combined procedure are also determined for both cases using Monte Car lo simulation. The results indicate that the combined procedure is qui te effective in rapidly detecting either a small or large step increas e in the Poisson rate or binomial probability over time.