User Stories

Thames Water Aims to Reduce Leaks by More Than 25% Using a MATLAB-Based Leak-Location System

Challenge

Pinpointing leaks is a highly specialized task. The Thames Water region has 32,000 km of mains and 18 million pipe joints, and many of the most serious leaks occur in complex networks of pipes buried 3 feet below the ground. Working entirely above-ground, leak-detection teams find the approximate location of the leak using flow meters to pick up any unusual change in water flow and electronic correlators, in conjunction with CRL's LeakFinder, to detect and analyze leak noise and pinpoint its source.

A correlator has two transducers (specialized microphones) and a signal-processing unit. The transducers are placed on the pipe at access points up to a few hundred feet on either side of the suspected leak. Their outputs are fed into the signal-processing unit, where they are digitized and cross-correlated. If the two outputs have a common signal (for example, the noise made by a leak) cross-correlation will produce a peak If the leak is offset from the midpoint, its sound will reach the transducers at different times, and the peak will be offset from the center by this time difference. This time delay is used to calculate the distance of the leak from the midpoint and determine where to start digging.

Although more cost-effective than exploratory excavation, correlators often fail to produce reliable results because their signals are prone to distortion. CRL needed to develop an easy-to-use correlator that would deliver a correlation signal undistorted by random noise or clutter.

"Using MATLAB to rapidly evaluate a number of different potential solutions was a great benefit—it saved us a lot of time and effort."

John Golby
CRL
DSP Group Sales Director

Solution

CRL engineers began by minimizing sources of random noise, a relatively simple and inexpensive procedure that involved sampling and processing the data at high resolution using a standard commercial 16-bit sound card and double precision (64-bit) floating point arithmetic on a PC. They digitized and sampled the two channels and then calculated the complex DFT (Discrete Fourier Transform) to transform the input time domain signal into the frequency domain.

To deal with clutter, they first applied a frequency filter to "whiten" the power spectrum. They then removed echoes by analyzing the auto correlation of each channel. (An echo will cause a peak in the auto correlation function of one channel only.) Finally, they developed a novel phase coherence analysis to determine which parts of the frequency spectrum contain useful signal. They used the output to construct a weighted frequency filter that significantly improved the detection of weak signals. Optimal performance was achieved by using this filter in combination with the whitening filter and the echo cancellation algorithm.

Software development proceeded in two phases: algorithm development followed by real-time algorithm implementation and GUI design. CRL relied on MATLAB throughout the algorithm development phase.

They began by implementing the 'industry standard' cross-correlation algorithms, which they redesigned to improve performance. The MATLAB environment enabled them to rapidly evaluate, develop, and test their innovative algorithms in a high-level language without needing to optimize code for real-time execution. MATLAB's comprehensive function library meant that a lot of code could be developed with a minimum of effort. "Knowing that these core functions worked correctly enabled us to concentrate on the problem at hand and details of our algorithms rather than their implementation," said CRL's John Golby.

The chart plotting and presentation features of MATLAB were also invaluable during this phase. Without this capability, real-time chart updating would have been impossible—and one of the most innovative features of the CRL correlator might never have come to light. As CRL focused on trying to find the variable frequency leak signal in the broadband noise, they used MATLAB to try several innovative approaches. "The versatility of the display features in MATLAB enabled the complicated output of the phase coherence analysis to be evaluated," said Golby. "Without it, the value of this technique might not even have been identified."

The echo cancellation and phase coherence analysis algorithms were written in C code and implemented on the LeakFinder research platform. Thames Water is currently conducting extensive field trials of LeakFinder and discussing implementing the algorithms in a commercial instrument.

Results

Products Used