A Simple Probabilistic Model With Extended Kalman Filter To Predict Multi-leak In Pipelines

TL;DR

This paper presents a probabilistic model combined with an Extended Kalman Filter to predict leakages and burst locations in pipelines, improving detection accuracy in noisy conditions.

Contribution

The paper introduces a novel application of EKF to a probabilistic pipeline leak prediction model, enhancing detection robustness and reducing parameter dependence.

Findings

EKF significantly improves leak detection accuracy.

The probabilistic model closely matches deterministic predictions.

Leakage and burst locations are accurately identified even with noisy data.

Abstract

Pipelines for water supply are susceptible to burst-leakage due to fluid pressures of various nature. High pressure heads resulting in circumferential and (or) axial stresses larger than the material yield stress could cause pipe failure. Of equal concern is the local boiling or cavitation effect in regions of fluid pressure dropping below its vapor pressure, which in turn develop air bubbles that get transported through the pipeline, bursting later at remote locations. We initially developed a simple probabilistic model based on Method of Characteristics (MOC) to simulate burst leakage in pipelines, and compared with a pure deterministic hydraulic model. We had not considered cavitation effects for simplicity. The results indicated that the simple probabilistic model was only marginally different in its prediction of the transients on comparison with the latter. In order to determine the position and amount of leakage in the distribution system, the detection method based on simulating hydraulic transients was further evaluated using Extended Kalman Filter (EKF). We found that this non-linear filtering approach on the fluid transient model considerably reduced the number of input parameters required, and it was able to predict leakage rate and burst positions even in a highly noisy environment.