·Table of Contents
PC Multimedia Based Leak Detection System for Water Distribution NetworksOsama Hunaidi, Alex Wang
National Research Council Canada, Institute for Research in Construction, Ottawa, Canada, K1A 0R6
Tel.: +1 (613) 993-9720, Fax: +1 (613) 952-8102
Alternatively, leaks can be pinpointed by using modern leak noise correlators. These are more efficient, yield more accurate results and are less dependent on user experience than listening devices. Leak noise correlators consist of acoustic sensors such as accelerometers and hydrophones, wireless signal transmitters and receivers, and an electronic processing unit. As shown schematically in Figure 1, the sensors are attached at two contact points with the pipe (normally fire hydrants) that bracket a suspected leak. The signals are transmitted from the sensors to the processing unit wirelessly. The processing unit computes the cross-correlation function of the two leak signals to determine the time lag between them. It then calculates the location of the leak based on a simple algebraic relationship between the time lag, sensor-to-sensor spacing, and sound propagation velocity in the pipe. Several makes of leak noise correlators are available commercially.
|Fig 1: Schematic illustration of the cross-correlation method|
In this paper a new leak detection system is introduced. The new system, called LeakFinder, is personal computer (PC) multimedia-based and incorporates several new signal processing and analysis innovations. It locates leaks using the cross-correlation method. Advantages of the new system are low cost, flexibility, and improved accuracy. Features of the new LeakFinder system are described in this paper. Details are also presented for its three main modules: data acquisition, signal conditioning, and spectral analysis. The accuracy of the new system is demonstrated using actual leak signals for plastic and metallic pipes.
|Fig 2: Main graphical user interface||Fig 3: Main output screen|
LeakFinder introduces several improvements including the option of calculating the time lag using an "enhanced" cross-correlation function, finely tunable digital pre- or post-filters for suppression of interfering noise, a wide range of spectral estimates including auto spectra and the coherence function, optional display of leak signals, and permanent storage of leak signals for off-site analysis. LeakFinder is also inexpensive which will help many users, including those in developing countries, who cannot afford existing systems.
The soundcard should be set to stereo mode (i.e., dual-channel input) unless it is used for recording leak sounds at one point only as in listening surveys. For accuracy, leak signals should be recorded using 16-bit resolution. The sampling rate is set by default to the lowest rate of the soundcard (usually 11,025 Hz) which is high enough for leak signals. Leak signals are usually recorded for a duration between 30 to 60 seconds.
In record mode, the volume control should be adjusted to utilize as much as possible of the soundcard's voltage range, without overloading it, to achieve a high signal-to-noise ratio. The volume control of the record mode can be set to the appropriate level online while checking the
level of leak signals using LeakFinder's preview function. The preview function records and displays leak signals without saving them to disk. Once the recording volume control level is selected, the signals can be recorded and saved to disk using the record function. The complete time history of saved signals can be displayed and printed using Leak Finder's display function.
Instead of pre-filtering the leak signals, the user may opt to post-filter the cross-correlation function. Post-filtering is considerably more efficient especially if the leak signals are long or if the cross-correlation function has to be calculated several times for different filter settings in order to find a definite peak. The filters implemented in LeakFinder are of the recursive 4th order Butterworth type. Leak signals are filtered in both the forward and reverse directions to eliminate time delays caused by the response of filters.
A unique feature of LeakFinder is that the speed at which leak signals are played back can be altered. For example, the speed is increased or decreased arbitrarily by simply overriding the sampling frequency at which leak signals were digitally recorded with a higher or lower sampling frequency. This is very useful when playing back leak signals of plastic pipes (or other non-metallic pipes) which only have very low-frequency components, e.g., in the infrasound range, and thus cannot be heard by an unaided human ear. The speeding up of the playback of low-frequency signals shifts their frequency content to a higher range at which the sensitivity of human hearing is high enough to detect the leak sound.
where L1 and L2 are the position of the leak relative to sensors 1 and 2, respectively, and c is the propagation velocity of sound in the water pipe. By substituting L2 = D - L1 in the above equation, the position of the leak relative to point 1 is found to be
where D is the distance between the sensors, usually measured on site or read off system maps. The propagation velocity depends on the type and size of the pipe. Velocity values can be obtained from pipe manufacturers or are easily measured on-site using a known in-bracket or out-of-bracket simulated leak.
The time lag between leak signals can also be found using LeakFinder's enhanced cross-correlation function. For narrow-band leak signals, this function provides improved resolution, i.e., better definition of peaks, in comparison with the standard cross-correlation function, achieved by computational pre-whitening of leak signals. This is helpful in the case of plastic pipes and in situations where leak sensors are closely spaced. Another advantage of the enhanced cross-correlation function is that there is no need to filter leak signals. This eliminates the uncertainty involved in selecting filter cutoff frequencies. The user needs only to specify the frequency above which the amplitude of the auto-spectra of the leak signals becomes insignificant, normally 50 Hz for plastic pipes and 500 Hz for metallic ones.
Appropriate cutoff frequencies of low and high-pass filters depend on the type of pipe and sensor-to-sensor spacing and therefore no fixed rules can be specified. Normally, the cutoff frequencies are selected so that they correspond to the frequency range where the amplitude of the auto-spectra of leak signals is significant and the coherence function is high. However, this is not always evident, especially for plastic pipes for which the amplitude of leak signals is very small and their frequency content is narrow. Recommended high and low-pass cutoff frequencies for typical plastic water distribution pipes, e.g., 6" or 8" PVC pipes, are 15 and 100 Hz, respectively. The high-pass limit may need to be increased or decreased in small increments, e.g., 1 or 2 Hz, until a distinct peak emerges in the cross-correlation function. For metallic pipes, leak signals contain much higher frequency components than plastic pipes and consequently a high frequency range may be used for calculating the cross-correlation function, e.g., 200 to 800 Hz.
Leak signals rarely contain frequency components above 1000 Hz in the case of metallic pipes and above 200 Hz in the case of plastic ones. Therefore, in order to speed up digital filtering and spectral analysis of the leak signals, the sampling frequency of recorded signals can be reduced optionally to selected frequencies, e.g., 500 Hz for plastic pipes and 2000 Hz for metallic ones.
Fig 4: Comparison between results obtained using LeakFinder and third-party spectral analysis software|
TurboLab 4.3): (a) leak signals in 6" PVC pipe, (b) leak signals in 8" ductile iron pipe
|© AIPnD , created by NDT.net|||Home| |Top||