Velocity Prediction of a Pipeline Inspection Gauge (PIG) with Machine Learning.

Area of Science:

• Engineering
• Artificial Intelligence
• Petroleum Engineering

Background:

• Pipeline inspection gauges (PIGs) are crucial for oil and gas pipeline maintenance.
• Current PIG speed measurement relies on odometers, which are prone to errors due to loss of contact.
• Accurate PIG velocity is essential for efficient pipeline operations.

Purpose of the Study:

• To develop a novel method for estimating PIG speed using neural networks.
• To reduce measurement errors associated with traditional odometer systems.
• To enhance the reliability of speed measurements in pipeline inspection.

Main Methods:

• Utilized neural networks (static and recurrent, including Long Short-Term Memory) to model PIG speed.
• Employed differential pressure and acceleration as input features, replacing odometer data.
• Developed a prototype PIG with a Raspberry Pi 3 embedded system for data collection.
• Trained and evaluated models using data from a dedicated PIG testing facility.

Main Results:

• Neural network models successfully learned the relationship between differential pressure, acceleration, and PIG speed.
• The proposed approach demonstrated the potential to overcome odometer-related measurement inaccuracies.
• Trained models showed capability in estimating PIG velocity based on physical parameters.

Conclusions:

• Neural networks offer a viable alternative for accurate PIG speed estimation.
• This method can complement existing odometer-based systems, increasing overall measurement reliability.
• The findings contribute to more efficient and dependable oil and gas pipeline inspection and maintenance.