Greedy Pursuit: Algorithms Show Promise in Measuring Multiphase Flow

Much study has been done on the application of ultrasound tomography to multiphase flowmeters, where a meter is configured to generate cross-sectional images of the flow inside a pipe in a noninvasive manner while still maintaining direct contact with the fluid. Reconstructing these images requires a high level of computational complexity and processing time, and the limited amount of data available can make the process more difficult.

Recent work by Saudi Aramco looked at compressed sensing—a technique that has been used in medical imaging and wireless communications—as an option that could optimize the conventional structure for data acquisition and compression in multiphase flow measurement.

Compressed sensing is a signal-processing technique where the signal being acquired is compressed at the time of sensing, capturing the useful information and condensing it into a small amount of data. The data can then be used to reconstruct the original signal back from a limited amount of measurements.

Speaking at the 2018 SPE Annual Technical Conference and Exhibition, Shamael Al-Shuhail, a technology solutions architect at Saudi Aramco, discussed her paper (SPE 191562) outlining a comparative analysis of compressed sensing algorithms against a conventional approach for image recovery performed at Saudi Aramco’s Upstream Research Center. She said the potential for improved image reconstruction can help companies more accurately quantify phase fractions in production wells.

“These are the images we want to recover. From these images, we’re able to see how much a percentage of water we have in our gas and we can look at it in a proactive manner. We can replace our dependency on separators in the gas stream,” she said.

Al-Shuhail said the use of compressed sensing in aiding image recovery builds upon previous work done on ultrasonic tomographic meters. She wrote of such meters that had been proposed over the past decade, including a multiphase flowmeter where ultrasonic sensors are installed through the pipe to form a ring-shaped array around the pipe’s circumference. Such a flowmeter allows for direct contact with the multiphase fluid without causing any disturbance to the flow passage. However, Al-Shuhail said that reconstructing an image in these systems is challenging because of the limited amount of data collected, especially in field applications.

Generally, images of natural objects have a scant representation of data when transformed into the wavelet domain, and images with few sharp edges and large smooth regions have a sparse structure. Al-Shuhail said that this makes compressed sensing a convenient reconstruction tool. Compressed sensing, she said, helps simplify the framework used for sampling and compressing data in part because it can recover the original data using a small number of measurements.

Al-Shuhail said that there are several ways to reconstruct an original image from a small number of measurements, so consequently the means to structure compressed sensing algorithms are diverse. In her paper, she outlined Saudi Aramco’s efforts to survey three greedy pursuit algorithms and run a comparative analysis in terms of image quality and quantity performance.

Greedy pursuit algorithms are a category of compressed sensing algorithms designed to select the data that seem to be the best at any given moment. They can be described as iterative data recovery techniques: In each iteration, a local optimal solution or a signal is chosen to build an approximation, and this process gets repeated enough times until a globally optimal solution can be made. Al-Shuhail said that greedy algorithms are known to be fast because they are simple to implement compared to conventional optimization algorithms.

“We’re getting both speed and simple implementation by adopting these algorithms,” she said.

In the study, greedy algorithms all showed better recovery when the researchers increased the number of measurements used for reconstructing the image. Another test showed that reducing the number of measurements beyond a certain limit caused the image recovery to suffer substantially. Al-Shuhail wrote that the company will extend its study into enhancing image recovery while maintaining low complexity and speed, as well as validating the results to create a more accurate image.


Don't miss out on the latest technology delivered to your email monthly.  Sign up for the Data Science and Digital Engineering newsletter.  If you are not logged in, you will receive a confirmation email that you will need to click on to confirm you want to receive the newsletter.