Unsteady pressure patterns discovery from high-frequency sensing in water distribution systems

Area of Science:

• Hydraulic Engineering
• Data Mining
• Signal Processing

Background:

• Pressure transients are a significant cause of pipe failures in water distribution systems (WDSs).
• The complex behavior of these transients is not fully understood and difficult to model accurately.
• High-frequency pressure data from sensors offer a potential avenue for deeper analysis.

Purpose of the Study:

• To investigate the behavior of pressure transients in WDSs using high-frequency sensor data.
• To develop and validate a data mining approach for detecting and clustering pressure transients.
• To identify recurrent and consistent patterns indicative of system behavior or anomalies.

Main Methods:

• A Time Series Data Mining (TSDM) approach was developed.
• A modified two-sided cumulative sum (CUSUM) algorithm was employed for transient detection.
• Dynamic Time Warping (DTW) was used for similarity measurement between transients.
• K-means clustering was applied to discover characteristic patterns.
• Performance was evaluated using SSE, Silhouette, and Calinski-Harabaz indices.

Main Results:

• The TSDM approach successfully detected and clustered pressure transients in WDSs.
• Consistent and unique patterns of pressure transients were identified across different sensor locations.
• The method demonstrated efficiency in extracting hidden information from high-frequency pressure signals.

Conclusions:

• The proposed data mining technique offers a robust method for analyzing pressure transients in WDSs.
• This approach enhances the understanding of transient behavior and its relation to pipe integrity.
• It provides a valuable tool for proactive management and maintenance of water distribution infrastructure.