Healy, John D. (1986, ASQC) Bell Communications Research, Red Bank, NJ
CUSUM procedures are one of the most powerful tools for detecting a shift from the good quality distribution to the bad. CUSUM procedures have the advantage that they are essentially sequential tests and, as a result, the good properties of sequential tests are carried over to CUSUM procedures. Usual CUSUM procedures assume that the mean of the underlying distribution is constant in the good state and then shifts to the bad state at an unknown time. This document presents a set of "adaptive" CUSUM procedures which allow the underlying mean to slowly change over time in the good state. The original motivation for the problem was the detection of trouble conditions on an electronic switching system based on call counts. Expected call counts at different times of the day slowly change when no trouble condition exists.