In the US, 64% of the overall consumed energy commodities are transported using a network of more than 2.6 million miles of pipelines. The Protecting our Infrastructure of Pipelines and Enhancing Safety (PIPES) Act of 2016 mandates the safe and efficient movement of energy products in the country. Oil and gas pipelines are laid over long distances and sometimes traverse across national borders, making it difficult to physically inspect the pipeline and identify the defects. Defects that are not detected in time can cause sudden explosions, endangering public life and safety. However, the reality is most energy infrastructure systems are still operated manually, requiring a large number of engineers and analysts to control the systems and identify defects.
Simpler and efficient defect detection
Currently, defect detection is a two-step process involving identification and verification by analysts. A sensor-based tool called pipeline inspection gauge (PIG) is inserted into the pipeline and generates huge volumes of data as successive inspection points are just 0.02 mm apart on the pipeline. Typically, this data is processed manually and checked for defects. This requires intensive human effort, making the detection process tedious, time consuming and cost prohibitive.
By using a robust machine learning approach in oil and gas industry, to analyze sensor data, you can make defect detection simpler, more effective, and efficient. The process involves aggregating the sensor data based on the nature of defects. For example, a high positive value data point indicates a dent in the pipeline, whereas a high negative value indicates an offtake.
However, even when the actual defects are few in number, a very large imbalanced data set might be produced once the aggregation process is completed. This could happen because of the bias and inaccuracy in the predictive model used on the data. Machine learning algorithms also tend to produce inaccurate results when faced with imbalanced data sets. In such cases, the chances of misclassification of defect class are high. The good news is there are ways to address these concerns.
Improve reliability of defect prediction models
You can use the over-sampling and under-sampling (OVUN) approach to create a balanced training data set that can be submitted to the machine learning algorithm to train the model. Typically, in the model building stage, approximately 10 machine learning algorithms are applied along with a tenfold cross validation method. This helps prevent model overfitting and underfitting which can lead to poor predictions on new data sets, and ensures a considerably large reduction in the number of false positives. At the same time, it ensures that none of the actual defects are missed, indicating a true positive rate of 100%. Support Vector Machine and Gradient Boosting Machine algorithms provide the best results in detecting the defects with minimal overfitting. For instance, Support Vector Machine algorithm enables a 90% reduction in false alarms.
Processing large volumes of sensor data using Hadoop clusters and deploying machine learning algorithms across the Big Data management system helps you accurately predict defects, while minimizing the processing time and costs. You can also run the task across nodes to further reduce the processing time, while the model is retained in a separate node and used to drive predictions on the newly processed data and detect defects.
Future proof oil and gas monitoring
The oil and gas industry recognizes that defect detection in the pipelines is a complex and tedious task, fraught with a high level of security and safety concerns. Identifying actual defects using machine learning in oil and gas industry can help companies enhance the safety, productivity, and life span of the oil and gas assets. In addition, it considerably lowers the effort required to complete the task, saving oil and gas companies significant manpower costs. As digitization sweeps across industries, now is the time for oil and gas companies to invest in machine learning for better outcomes, compliance and competitive advantage.