You are here

Using data to advance from reactive to predictive repair

05th December 2017

Interview with Kishore Sundararajan, chief technology officer of GE Oil and Gas

Inspection is changing in the oil and gas sector. Kishore Sundararajan, chief technology officer of GE Oil and Gas explains how predictive data analytics, robotics, and artificial intelligence will soon be used to deliver advanced inspection services.

Kishore Sundararajan
Kishore Sundararajan

Routine inspections can be slow and costly, and often include humans performing high-risk tasks. Data is manually collected and processed, and can take weeks to analyse. By reducing high-risk tasks through robotics, Avitas Systems can make inspection processes safer and more efficient through data automation, decreasing costs by up to 25 per cent. By performing inspections based on anticipated risk, instead of regular time intervals, it can also help to increase asset longevity.

“The inspection services industry requires cutting-edge technologies to help avoid unplanned asset downtime and to deliver new, valuable insights,” Kishore Sundararajan, CTO of GE Oil and Gas says.

 “We will use state-of-the-art robotics, automated defect recognition, and cloud-based technology to give customers the customised service and data they need to advance from reactive to predictive repair.”

Avitas Systems partners with GE Global Research and market leaders in drone and robotics technology to develop ground-based and aerial autonomous and semi-autonomous robots with a wide variety of sensors used to inspect assets. The solution, built on GE’s Predix platform for the Industrial Internet, analyses inspection data, integrates regulatory and external data sources, such as weather, identifies defects automatically and recommends optimal inspection and maintenance schedules. The system can fuse data from diverse sources and independently analyse the relationships between them for deeper insights and incorporates user feedback to make defect detection smarter and more accurate.

Customers will be able to access inspection data in real time through an inspection platform that includes customised dashboards and reports. The platform also includes 24/7 live alerts that enable customers to monitor their data from any location.

 

Why is the inspection process so important?

Unplanned asset downtime is a top issue for the oil and gas industry, and can cost operators millions of dollars. Avitas Systems will help enhance the efficiency of inspections, and can help our customers and others avoid significant costs by reducing downtime and increasing safety.

When you look back at inspections and the way they’re being carried out in the industry and I don't mean just oil and gas, I mean a wider side of industry, they have not evolved, changed, been disrupted in decades. This gives us an opportunity to look at doing things differently and what enables that are a couple of technology enablers, cloud services, computer power and memory, communication infrastructures that have become cheap and ubiquitous and fast, and then the evolution of artificial intelligence and robotics coming together in a different way with a core structure that becomes almost consumer oriented.

 

Is there a demand from industry?

Although it is the convergence of these three technology trends that is allowing this to happen there is also a driver in the industry; the price of oil. What it’s doing is demanding asset productivity from our customers. We lower the cost of inspections. How do we lower the costs of inspections? We’re getting rid of scaffolding. For example, to climb up a flare stack, it takes three days to put the scaffolding up, three days to take it down and then interpreting the images that somebody has seen. That’s just an example of taking cost out.

The other part is that you can do inspections on demand. there is no need to schedule a building for scaffolding. I can fly a drone up there to do the inspection on demand, thereby we create flexibility. We make inspections more effective because we can send combinations of thermal imagers, IR imagers to light our images to high definition cameras and we start seeing things that are usually not detected by the human eye. We are also looking at combining different inspections in one survey.

We don't want to be a drone company; we want a partner with drone providers. We don't want to make things that swim, we’ll partner with companies like Saab. The second piece is the differentiation of the cloud infrastructure.

 

How do you handle high velocity image data?

With real-time streaming of high velocity images into the cloud you get petabytes of data. You don't want to end up with a cost of just storing data, so you need to think cleverly about your cloud infrastructure. That’s the differentiator in our cloud offering.

We are using deep learning to reinterpret images that are coming through. Automatic defect recognition is one, at the very bottom, then you start doing classification defects for how harsh or how tough these are. The third is, can I now take that data and combine with weather and chemistry to start doing crack propagation analysis or corrosion propagation analysis?

Now we can also do change of state. For example, we do quarterly inspections. Can I compare the images that came in a quarter ago and say what has changed? That’s also part of the intelligence.

 

Drones are obviously core to this process. How do you manage them?

This is one of the biggest things we do. If you have a million drones, you can’t have a million operators; it is just not a not a sustainable model. We work together with GE Aviation because commercial flight management and flight planning is something that GE is very good at. Their technology is applicable to our drones. Think of drones inspecting transmission lines, pipelines or railway lines, the drones have to fly out of sight, so a drone operator is not the answer. I need this drone to fly out of sight miles at a time and come back home safely. That’s where the intelligence and the flight management system will make a difference.

 

Is there any problem with running drones out of line of sight?

We are working with the FAA right as we speak. At the moment, they are permitted by exception, but we’re working with FAA to draft the framework. There are three kinds: there are hobby drones, delivery drones such as Amazon and then there are people like us who want to use drones for work. We’re working through with the FAA about how to craft rules so that you don't interfere with civil aviation and cause accidents.

 

Has this system been used in a real-life situation yet?

We only started this journey in October last year and we’ve been validating with clients. We’ve done about a half a dozen proof of concepts with the various customers all the way from the Middle East through Europe to the United States.  We have proven to them we can do this. We are going through the compliance checklist that customers have. We are validating that we are compliant with the inspection plans and checklist and as of this quarter, we are in a place that we’re coding and taking orders. We haven't announced any but we’re very close to signing on the dotted line on a few.

 

Is the system utilising edge computing?

Let's talk about what happens on the edge because we have analytical tools on the edge and they are used for routing and planning and execution. They decide how should we do this inspection, how should we fly, what path should we take around flare stacks and then how do I complete that inspection? That’s one segment of tools that are at the edge. They are managed from the cloud but they’re really executed on the edge.

The other part is intelligence in the cloud stacks to determine how much data should we have in one storage versus cold storage versus hard storage because I need certain amount of data in hard storage but I want to keep that minimal because it’s expensive.

Then you have intelligence at the third layer which is around automated defect recognition classification and there you have two types of learning going on, machine learning and deep industry learning. One is learning from the same inspection being done over and over.

 

How does the system use predictive analytics?

Let's say that we have really achieved certain amount of minimal status on augmentative defect recognition, automotive classification of defects and the machine learning is all set out. Now I have the platform, the basis, to combine weather data and weather predictions to say, ‘You know what, that particular corrosion or that particular risk or that particular buckling that I see at the top of your flare stack is at a higher risk this week than any other week because there’s a hurricane around the corner’.

 

What future innovations are planned?

If you stay within the inspection space, let's take the example of tank farm inspections. Typically, at tank farms the scheduling happens for somebody to take tank level readings. Then there’s another scheduling that takes place to have a crew who can check for spray, peel and corrosion. Can I combine all these things and do only one inspection because now I can have a high-definition camera, thermal imaging camera and LiDAR on a flight so I can do these at the same time?

That’s where I see it changing and evolving. The other one is big transmission towers and transmission lines. They’re a very hard space to inspect but if I fly and I use a LiDAR-based inspection, I can start measuring the how far these transmission towers are leaning. They lean with life, because they’re carrying the weight of these transmission lines. Now I can give better data to the operators saying that this transmission tower is at a higher risk opposed to another one because it's leaning over more.

These are the spaces I see evolving. Now the third space that is not inspection, but people are calling us and asking if we can run a flight and produce a 3-D model of a plant or oil field and we’re starting to experiment with that.

Related topics: