Predicting Sand Rates in Oil and Gas Wells Using Acoustic Sand Detectors

Abstract Acoustic Sand Detectors (ASDs) are widely used in the oil and gas industry to detect and quantify sand production in oil and gas wells and pipelines. These detectors capture both the background noise generated by multiphase flow and the noise produced by sand particles impacting the pipe walls. Prediction of sand rates is essential for optimizing sand management strategies and preventing pipeline erosion. A novel approach is presented for predicting sand rates using Machine Learning/Artificial Intelligence (AI/ML) techniques. The method utilizes a model for estimating representative particle impact speed, tailored for each flow regime based on a large dataset of experimental observations. This innovative method enables sand rate prediction, using a single parameter the background noise generated by flow in the absence of sand, recorded by ASDs. Given the industry's need for sand rate prediction to enhance pipeline integrity and operational efficiency, this approach offers significant advancement in sand monitoring technologies. A novel technique has been developed to predict sand rates using an extensive dataset collected over the years from various large-scale experimental facilities, with pipe sizes ranging from 50.8 mm to 101.6 mm and sand sizes between 25 microns and 600 microns. The experiments cover a wide range of multiphase flow regimes, including annular, slug, churn, dispersed bubble, bubble flow, gas-sand, and liquid-sand conditions, with superficial gas velocities from 0 m/s to 49 m/s and superficial liquid velocities from 0 m/s to 6.6 m/s. A previously developed erosion prediction model for pipe elbows has been utilized to determine the characteristic particle impact speed, which is essential for calculating a calibration factor (C-factor) for given flow conditions. Additionally, Machine Learning (ML) algorithm has been trained and tested using 211 data points, to predict a new C-factor for any given flow conditions. The predicted C factors are then used to calculate the sand rate based on the data recorded by the ASDs. The Machine Learning (ML) models were utilized to predict sand rates for various velocity combinations across different flow patterns containing sand. The sensitivity of the predicted sand rate to changes in input parameters was also analyzed. Furthermore, the ML-based sand rate prediction method, trained on experimental data, was validated against actual field conditions and experiments conducted in other laboratories reported in the literature. The results demonstrate that the ML approach achieves strong training performance and reliable predictions for a wide range of flow conditions and pipe sizes that were not previously tested. The findings of this work enable operators to predict sand rates in pipelines using ASDs for a given production condition, relying solely on the background noise recorded by the detector. When calibration with only one sand injection is performed the approach is quite accurate even when flow conditions and particle rates and characteristics change. This approach allows for timely decision making in sand control, enhancing pipeline integrity and operational efficiency.