Acoustic signal processing for well integrity monitoring systems

• Main Author: Ang, Yi Yang
• Other Authors: Gan Woon Seng
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/143728
• Institution: Nanyang Technological University

This thesis presents a systematic study of the processing of acoustic radiation to address the complex practical challenges involved in acoustic signal processing and its application to a passive well integrity monitoring system. The conventional approach to passive well integrity evaluation is to deploy single or multiple independent sensors to detect leaks within the wellbore. Even though these approaches have been proven to be successful in detecting leaks, they do not determine other crucial parameters such as the directions, the radial locations, and the types of leak in the wellbore. These parameters are important because they can enable the development of enhanced diagnostics of the condition of a wellbore. Accordingly, an array processing approach was recently developed to provide an additional dimension of information to localize the source of a leak. At the beginning of this thesis, the conventional beamforming used in array signal processing algorithms is studied to investigate the advantages of using sensor array for passive well integrity evaluation. Even though an array can provide localization information, an array’s spatial aliasing frequency limits the type of localizable passive source that can be used. The array spatial aliasing frequency is usually below the typical time-domain aliasing limit, which is one-half the sampling rate. This limitation prevents the array from localizing passive sources with frequency contents above the array spatial aliasing frequency. Accordingly, the far-field source model is investigated. If the wideband source satisfies a specific condition, an aliasing-free direction-of-arrival estimation can be achieved via sparse beamforming. However, this condition is not applicable to the recently proposed grid-free compressing beamforming technique because it was derived for sources with a single frequency. To alleviate this deficiency, a multiband grid-free technique is developed, and a new cost-function for a multiband source is derived from a theoretical investigation of a multiband grid-free model, where this function can be solved through semidefinite programming (SDP). The method generalizes the original single-band grid-free technique to a multiband representation. This is validated through simulations and experimental test results. Next, a study is performed on the nearfield source model, which reveals that a signal received from a point source in the nearfield is not bandlimited in the spatial frequency domain. Although many solutions are available in the literature, they all require modification of the array’s aperture, and this modification may not be practical in the passive well integrity evaluation problem. Thus, via investigation of the theoretical array’s steering vector, a condition is derived that guarantees an aliased-free localization. This condition is validated by extensive numerical simulations and experimental analyses. Effectively, the condition extends the operating frequency range of the array without requiring modification of the aperture. In most situations, it is desirable to use few sensing devices because the use of numerous sensing devices will drive up the cost and complexity of circuitry. To optimize this trade-off, the third study focuses on performing localization with a single sensing device. Here, it is found that a recording from a single moving sensor of a known trajectory can be used to form a one-sensor array (OSA) that achieves localization performance close to that of the Cramer-Rao bound (CRB) of a stationary multi-element array. However, this performance is restricted to harmonic sources. For non-harmonic sources, an innovative approach called the differential received-signal strength one-sensor array (dROSA) is developed. Notably, the dROSA approach does not require phase compensation, and its use is not restricted to harmonic sources. Finally, the classification of sources is also essential to a passive well integrity evaluation system. To that end, a support vector machine (SVM) learning approach is developed to classify the types of leaks into four categories, as follows: (i) liquid-to-liquid (L2L); (ii) liquid-to-gas (L2G); (iii) gas-to-liquid (G2L); and (iv) gas-to-gas (G2G). It is found that this approach correctly categorizes 93% of leaks from 1800 acoustic recordings collected from ten different oilfields.