Multi-Sensor Fusion for Transient-Based Pipeline Leak Localization in the Dempster-Shafer Evidence Framework

Detecting a small leak in a water-supply pipe with a wave scattering power of the order of noise or smaller is a challenging problem because its signature in a measured signal is weak. Experimental data show that multiple sensors enhance the evidence about a leak in a transient test and, thus, increase the possibility of successful leak detection in noisy environments. Therefore, a leakage localization scheme is proposed, which fuses multi-sensor measurements in the Dempster-Shafer evidence framework. The signature of a leak in each measurement is extracted and translated into a piece of evidence regarding its presence and location. Then, the pieces of evidence from different sensors are fused using the Dempster's rule of combination to form a unified leak location estimation. The proposed method is model-free and is thus insensitive to imprecise knowledge of pipe system. The gain of the multi-sensor fusion mechanism on the leakage localization accuracy is demonstrated via both numerical and experimental data.