Random forest regression prediction of solid particle Erosion in elbows

• Published in: Powder technology 2018-10, Vol.338, p.983-992
• Main Authors: Zahedi, Peyman, Parvandeh, Saeid, Asgharpour, Alireza, McLaury, Brenton S., Shirazi, Siamack A., McKinney, Brett A.
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial intelligence; Catastrophic events; Data points; Disasters; Elbow; Erosion; Erosion rates; Fittings; Gas production; Learning algorithms; Machine learning; Multiphase flow; Oil and gas industry; Oil and gas production; Parameters; Petroleum industry; Pipe fittings; Prediction models; Random Forest; Regression; Regression analysis; Regression models; Transportation; Transportation systems
• ISSN: 0032-5910; 1873-328X
• DOI: 10.1016/j.powtec.2018.07.055

Solid particle erosion is an inevitable problem in oil and gas production and transportation systems, and it can cause severe damage in pipe fittings in which the flow direction changes suddenly. Erosion occurs in a variety of fittings, and one common fitting which is one of the most vulnerable to erosion is the elbow. Repair and/or replacement of pipe fittings in the appropriate period can avoid catastrophic disasters such as hydrocarbon release events. Therefore, accurate prediction of erosion is a subject of interest for the oil and gas industry. Both experimental and modeling approaches have been conducted in the past under different flow conditions and materials. The goal of this study is to use relevant experimental parameters as predictor variables and utilize a machine learning algorithm to predict erosion rate and validate the prediction under different conditions with test data. It is assumed that important parameters in erosion magnitude are material characteristics, pipe diameter, particles properties and size and carrier fluid properties. 201 experimental data points with a broad range of conditions were selected for the analysis. Predicted erosion rates through different machine learning techniques were compared, and Random Forest regression model was selected as an effective and alternative approach for erosion prediction. This method was examined on data points and compared with experimental results. Good agreement was observed in this method with experiments which provides the framework for a new way to predict solid particle erosion which is simpler yet accurate in comparison with available prediction models. [Display omitted] •A statistical approach was developed for predicting solid particle erosion.•Random Forest aggression algorithm along with other machine learning algorithms were utilized in this work.•Predicted erosion magnitudes were compared with experimental data in variety of conditions.