Design of a modified sequential probability ratio test (SPRT) for pipeline leak detection

A classical SPRT likelihood test for sequential independently distributed data is often used in pipeline mass balance leak detection to distinguish between true leaks and false alarms in the minimum time consistent with a user defined error tolerance. However such time series data would not be expected to be independent, especially as it is often moving averaged to remove noise and unwanted transients. In this paper a modified SPRT test is derived using a simple Gaussian Markov process to model a correlated time series. Application of the modified test to correlated time series data is shown to reduce false alarms below that of a classical SPRT.