Predictive Analytics for Data-Driven Decision Making

Newsletter Subscription

Keep up with new content on the site, receive exclusive content and commentary, and learn about activities within the Straight Talk community.

Employing predictive analytics to improve oil and gas production uptime and curb maintenance costs presents a complex IT challenge – and a human one, as well.

By Hugh Banister, CIO, Santos

Santos is an upstream oil and gas exploration company which originated in South Australia. We’ve been producing oil and gas from the Cooper Basin in Queensland for 44 years to supply the domestic market. And we have some very old equipment assets in the field, run by professionals who have been doing this all their working lives and have made decisions, based on experience and intuition, about how to manage that critical equipment.

Take a compressor, which boosts the pressure of gas as it is extracted from the earth so that it can be transported efficiently through a pipeline. One of our men in the field might listen to the sound it is making and decide to increase or reduce the feed, based on his experience and gut feel. If his intuition is wrong, there is a direct impact on production.

For years, that human assessment was the best way to do things. Today, there are other options, which allow for more fact-based decision-making. To take advantage of these new technology options, however, the company is having to change.  And the IT function is playing a critical role in that change, transforming Santos from an intuition-based to a data-driven organization, resulting in actionable intelligence and a direct impact on corporate profitability.

Taking Advantage of Greener Fields

A testing ground for this new decision-making approach is an $18.5 billion initiative Santos is leading called the Gladstone Liquid Natural Gas project, or GLNG. In partnership with Malaysia’s PETRONAS, France’s Total, and South Korea’s KOGAS, we have begun this year producing natural gas from coal seams in Queensland’s Surat and Bowen Basins, shipping it via a 420-kilometer gas transmission pipeline to a new plant on Curtis Island, and converting it there into liquefied natural gas for sale to the global market.

Because the process of extracting gas from coal seams is different from our conventional method of extracting gas from sandstone reservoirs, the project is a true greenfield operation. This has allowed Santos to take a very different approach to building and managing assets in the Queensland fields and to improve safety and decrease costs. We’re replacing a lot of what was traditionally done by people onsite with automated solutions. These are integrated into a back-end collaboration and control environment in a centralized location that is staffed by our experts around the clock.

The GLNG project has helped us understand how we can use data and information to make better decisions throughout our entire operations. The concept of smarter production has been around for at least eight years. But the sensor technology and advanced analytics required to achieve this have only recently matured to a point where automating production is achievable and effective. The key problem continues to be not a shortage of information but rather the overwhelming abundance of data produced by these systems. It can be very difficult to identify the wood from the trees, so the secret is to identify exceptions and filter out the information where nothing exceptional is occurring.

Putting Predictive Analytics to Work

Our primary goals are to keep production flowing and to control maintenance costs. And that requires both reliable data-gathering equipment in the field and the analytics and data-delivery systems that will empower professionals to make better decision about production and equipment in the field.

We started with predictive analytics that focused on equipment reliability, which would give us some advanced warning when a piece of equipment was about to fail. We work in very remote areas, with vast distances between wells and compressor stations. Doing regular maintenance on the equipment is labor intensive and costly. And if a piece of equipment does fail, production stops. We stand to lose tens of thousands of dollars for every hour a well is offline. And it costs more to repair a failed piece of equipment than to repair something that’s about to fail.

To build the complex models that would enable this sort of predictive maintenance, we used data collected over the last 20 years in one of our more mature fields, called Big Lake. This helped us discover the common signatures for failure and then compare those signatures to our current operations. Today, using that predictive model, we typically get 48 hours notice that a piece of equipment is about to fail.

We recently constructed our Brisbane Operations Centre, or BOC, for remote monitoring and control of our GLNG asset. There, a team of experts remotely controls the upstream operations of our wells, compressors, and pipeline. The BOC looks a bit like Cape Canaveral, with large walls of screens delivering information from the field in a way that’s easily digestible to those who need to make decisions.

A big challenge in developing these models has been avoiding false-positive results. Some of our field staff in the Cooper Basin have worked there for 30 years. Sure, sometimes their gut decisions aren’t right. But if we alert them to an issue where there is none, they will immediately lose confidence in these models. We can’t afford that. So we have to make sure the predictions of equipment failure are correct as close to 100% of the time as possible.

Winning over the Skeptics

The team in GLNG lapped up the new working environment in Queensland. It was a completely new area of the business, and they were able to work that way from day one. That’s the easiest way to introduce this kind of change. Everyone in the BOC believed passionately in the benefits of a data-driven environment and were pushing the boundaries to make it work.

The brownfield sites are a different story, particularly for our unionized field force that worries about job security. There, we have to take a much more consensus-based approach. We’re not trying to control upstream operations from Adelaide but rather offer collaborative support from the head office.

We do what we can to physically break down any barriers between corporate headquarters and our field operations. Here at our corporate center, we’ve created a very industrial looking environment. There are no ceiling tiles in the building so the pipes are exposed. The professionals who work in our collaboration center wear the same “Santos Blues” safety kit as those in the field. We want to minimize visible distinctions, so there doesn’t appear to be much difference between sitting in the office or working in a remote site in the Cooper Basin.

Still, change is difficult. The always-on video allows everyone to communicate as if they are in the same location. It is just like saying something to the person sitting next to you, even though he may be 600 kilometers away. This also helps with situational awareness and improves safety. We recently found one of those cameras in a remote location with a wig over it. I think that the initial fear was that the cameras were being used to spy on individuals. Clearly more change management is required to explain the purpose of the always-on video – and prove the value of our new approach to everyone.

 

The Takeaways 
Today’s predictive analytics technologies allow for more fact-based decision making, rather than decisions relying on human assessment skills. But companies need to change how they operate if they hope to take advantage of these new options—and IT will play a critical role.
Because of the overwhelming abundance of data these new systems produce, the challenge is finding ways to filter out superfluous information in order to identify major problems.