The Future of Acid Corrosion Inhibitor Testing: A Machine Learning Solution
腐蚀
缓蚀剂
计算机科学
材料科学
冶金
作者
Sulaiman T. Ureiga,Nasser Hajri,Todd Green,Muhammad Umar Javed
标识
DOI:10.2523/iptc-24111-ea
摘要
Abstract The paper aims to present a machine learning virtual laboratory as an alternative to autoclave testing, which is typically utilized in determining corrosion inhibitor performance. The virtual lab aids in the optimization of acid corrosion inhibitor loading for stimulation jobs. The lowest acceptable/safe chemical dosage of the inhibitor is the optimal acid corrosion inhibitor loading. Hundreds of physical autoclave test results from a variety of settings were used to develop and test the virtual laboratory's underlying machine learning models. Metal coupons are tested with acid recipes in a laboratory autoclave under simulated wellbore conditions. The weight loss per unit area is measured in the autoclave lab test. The weight loss per unit area is measured in the autoclave lab test. Expected acid exposure period, acid strength or concentration percentage, downhole temperature, corrosion inhibitor type, corrosion inhibitor loading, and steel type are among the parameters that affects the testing. The results of the laboratory tests are categorized into two groups. One set includes 80% of the total autoclave tests which is used to train the machine learning model. The remaining 20% is used in a completely different set designated as the testing set. The testing set is used to validate that machine learning model predictions are accurate. For training and testing, several machine learning models are applied to the same data sets. Field engineers can utilize the machine learning model with the best confidence to design acid stimulation recipes with acceptable corrosion metal loss and low cost. The model can be dynamic, with updated tests being uploaded to the model on a regular basis to train it and improve prediction confidence. Eight machine learning models were trained and then tested through their tests. Each machine learning model tested the sensitivity of testing parameters using four distinct input cases to determine the best machine learning model. The outputs of two machine learning models have a 97.90 percent confidence level. The virtual laboratory will assist oilfield operators in lowering the costs of acidizing treatments while maintaining safe corrosion inhibition limits. In addition, the virtual lab would reduce the requirement for laboratory tests to determine acid formulas for each field job.