Hybrid Statistical Process Control Method for Water Distribution Pipe Burst Detection

Abstract

Statistical process control (SPC) identifies any nonrandom patterns in the system output variables of a water distribution system (WDS) by comparing them to their normal historic mean and variance. While each SPC method has different performance characteristics, there has been little effort expended to develop a hybrid method that combines the different characteristics. This paper proposes a hybrid SPC method that combines a modified Western Electric Company (WECO) method and the cumulative sum (CUSUM) method. First, the original WECO method is modified to incorporate a user-defined parameter c that manipulates the tolerance for warning and control limits to fit the specific network of interest. Then, the best parameter set is identified for each of the two individual methods so that coupling them should not increase false alarms. The detection effectiveness and efficiency of the WECO, CUSUM, and hybrid methods were compared by using common data sets obtained from a hydraulic model of the Austin network. The results showed that a simple coupling of individual SPC methods with different detection characteristics can significantly improve pipe burst detection probability while reducing false alarm rates and average detection time.