Quick, Create a Diversion!
Ultrasonic sensors and machine learning protect large stretches of oil and natural gas pipelines from illegal tapping.
Oil and natural gas pipelines play a vital role in the energy industry, providing a safe and efficient way to transport these resources to power homes and businesses across the world. These pipelines transport oil and gas from production sites to refineries, storage facilities, and distribution centers. They are often the most cost-effective and environmentally friendly method of transportation, reducing the need for truck and rail transport and the associated emissions.
Pipelines can span thousands of miles, and it is challenging to keep an eye on all of that pipe. Unfortunately, some people take advantage of this, and illegally tap into a pipeline to steal its contents. These illegal diversions violate pipeline quality standards, and can lead to potential leaks and possibly even explosions. This threatens the environment and the communities near the pipelines, and can also have significant economic consequences. Valuable resources are lost for companies and that will also mean increased costs for consumers. In addition, the repair and cleanup costs associated with pipeline damage can be significant, further impacting the economy.
A team of researchers at the University of British Columbia Okanagan has developed a technique for the inspection of high-density polyethylene (HDPE) pipes to scan for illegal diversions. Large stretches of pipelines can be monitored using this method that involves transmitting ultrasound waves through a pipe. These ultrasound signals are captured by a receiver, then processed by a deep neural network that can determine if a diversion is present in the pipeline with a high degree of accuracy.
Piezoelectric transducers were used to transmit and receive ultrasonic waves with a 40 KHz frequency. These transducers are capable of both generating an electrical signal in response to a sensed mechanical signal (vibrations in the pipe), and generating a mechanical signal from a driving electrical signal. Pairs of transducers (transmitter and receiver) are attached to the pipeline at regular intervals via clamps, which serves to enhance signal quality.
Using this design, an experimental setup was constructed, consisting of a ¾ inch HDPE pipe with clamped piezoelectric transducers. A dataset containing 2,292 samples (1,006 from control pipes and 1,286 from pipes with diversions) was collected using this setup. These samples were used to train a deep neural network to recognize the difference between an unaltered pipeline, and one with a diversion.
During validation, it was found that the model was 90.3% accurate when a single transducer pair was used, and 99.6% accurate with two transducers. This excellent classification accuracy shows that this new technique may prove to be useful in real-world applications to keep oil and gas pipelines safe. The team believes that these results can also be achieved for pipes that are different diameters, or are made from other materials.
In the future, the researchers are planning to explore if it is possible to enhance the system further by adding another step that estimates the location of the diversion using another deep neural network.