Control valves are common components of oil and gas production operations. Typically, thousands of control valves must be inspected throughout the turnaround period. However, not all of them are faulty, and selecting only the anomalous ones can save significant time and resources.
As a result, we built a machine learning-based solution that can evaluate valve states and suggest the type of failures using routine historian data already accessible in the company's database during the last five years. When valves are identified as warning or alarming, the three most typical causes, stiction, passing, and oversizing, are successfully detected.
The algorithm has now been migrated and is ready to run on the company's cloud platform.