Utilizing Artificial Intelligence for Determining Threshold Sand Rates from Acoustic Monitors

Abstract Acoustic sand monitors are non-intrusive passive devices and they can be used to provide a warning when sand is being produced in pipelines and/or determine sand rates when calibrated. These devices are attractive to be installed on individual wells or manifolds to detect sand production in offshore production units. One drawback of such devices is the noise created by flow impacting the pipe walls. Thus, if the impact speed or the rate of sand is low, the monitor will not be able to recognize/differentiate the noise that is generated by solid particles impacting pipe walls. The minimum sand rate that can be recognized by the monitor is called a threshold sand rate (TSR). A significant amount of data has been compiled over the years for TSR for various flow conditions, pipe sizes, and sand sizes with stainless steel pipe material. However, it has become challenging to use this data to develop a theoretical model for different flow conditions, pipe, and sand sizes. The goal of this paper is to present an artificial intelligence (AI) approach to determine TSR. The acoustic sand monitor data utilized in this work has been generated mostly in the lab for sand sizes ranging from 25 to 300 μm varying sand concentrations and pipe diameters 2, 3, and 4 inches in vertical and horizontal directions downstream of stainless-steel elbows. The TSR data, pipe sizes, gas, and liquid rates, and calculated erosion obtained by a mechanistic model (characterizing other parameters that are not directly included) have been used to train and create an AI model that can predict TSR for different flow conditions. The parameters of the AI models, including elastic net, random forest (RF), and support vector machine (SVM), are optimized using nested cross-validation and the model performance is evaluated by the R-squared score. The AI model was used to predict TSR for several new test conditions collected on a 4-in flow loop. Random samples of past data were also compared to predicted values for further validation. Additionally, this relationship revealed by the AI model was used to predict field data for TSR data available in the literature. The model agreement with published data is encouraging and suggests that this model can be extended for a variety of flow conditions and pipe sizes not tested before. This work provides significant TSR data, a framework describing a novel methodology to utilize Artificial Intelligence to correlate the TSR with pipe size, erosion model indicating the severity of impact speed of particles, sand size, and superficial liquid and gas velocities.