ARTIFICIAL INTELLIGENCE

The refinery industry knows that when profit margins are razor-thin, even the smallest efficiencies from every aspect of business operations add up quickly. The sector is squeezed by a host of challenges that range from labor shortages to outdated infrastructure. Companies also race to meet rigorous global sustainability standards. To tackle these issues, the industry leans on automation and other advanced technologies that can mine granular efficiencies in practically every aspect of everyday operations.

Against this backdrop, at the Impala Platinum Base Metal Refinery in South Africa, computer vision and edge AI technology ensure efficient operations. A process optimization project with Britehouse Mobility, a subsidiary of NTT, shows how machine vision software can help improve the complex production of ammonium sulfate—an inorganic salt with a number of commercial uses.

#MachineVision #software can help improve the complex production of ammonium sulfate—an inorganic salt with a number of commercial uses. @GlobalNTT via @insightdottech

The Challenges of Producing Ammonium Sulfate

Used as an agricultural fertilizer and in the manufacturing of vanadium, ammonium sulfate is a byproduct produced in the refinery. The product produced at the facility must meet specific standards mandated by the Department of Agriculture, Forestry, and Fisheries. A key criterion affecting the ammonium sulphate produced at the plant is its nickel content, which must not exceed 200 parts per million (ppm). But nickel particles attach to the smaller ammonium sulfate crystals. An effective method to prevent the nickel to report to the final product is to screen the product using a vibrating screen.

Vibrating screens can cause a lot of dust, especially when very fine particles are present. The vibrating screens are therefore completely enclosed, which makes it difficult to see when the screening panels are blinded, and typically can be detected only by opening the screen while it is offline. Blinded screen panels allow the fine particles containing nickel to report to the final product and, if detected too late, will impact the quality of the ammonium sulfate.

Opening the screen periodically can be time-consuming. Consequently, an online monitoring system enables the production team to quickly react should there be any pegging or blinding in the screen. An additional advantage is that the cameras can easily detect other issues with upstream processes as the vibrating screen is the final step in the process.

Machine Vision and 3D Cameras in Action

The Britehouse Mobility solution shows the use of volumetric cameras and can prevent such problematic pegging and blinding of vibrating mesh screens, which can deteriorate the quality of the final product output. In conjunction with the cameras, a machine vision algorithm studies the output and recognizes when the screen is getting blocked.

Anyone who has taken a picture with a smartphone will appreciate the challenges involved in getting a non-blurry picture when there’s so much shaking involved. The challenge in developing a solution was to find a camera that could perform under such conditions.

“It’s not just a camera…it’s something that is robust enough to survive the brutal forces that come with that environment,” says Donovan Bell, Senior Solutions Architect at Britehouse Mobility. In addition, the camera had to move with the screen, to carry the same resonance, so the pictures are not blurry.

With the camera placed under adequate lighting conditions, the team trained a machine learning model to recognize what functioning and blocked screens looked like. The final results receive a rating so operators can judge the severity of the problem before intervening. Volumetric measurements deliver information not just about the extent of pegging and blinding but about the weight and dimensions of the undesirable nickel. Immediate data processes at the edge and a cloud-based extension, the Britehouse Mobility Atajo OnEdge platform, delivers visualizations and alerts operators as necessary.

“The solution improves safety and efficiency because it reduces the number of times people have to interrupt operations, take lids off, and have a look under the vibrating screen,” says Marco Capazario, head of the IoT division at Britehouse Mobility.

In addition, workers can now be reactive instead of proactive. It’s about more than just monitoring blockage, Capazario says. “It’s giving them insights into failures upstream and the ability to dig deeper in terms of root cause analysis,” he adds. Among the many questions the Britehouse Mobility/NTT solution can answer: Why are we experiencing efficiency losses at certain times and what’s happening upstream of the plant that’s causing and creating this? Can we identify this through data insights?

Collaboration Demonstrates Refinery Innovation

Through Intel’s guidance, Britehouse Mobility/NTT worked with Framos, which installed Intel^® RealSense^™ cameras in industrial-grade enclosures. The collaboration helped Britehouse Mobility zero in on a camera faster. The RealSense camera fits the bill because of its volumetric capabilities and depth perception.

Because the object-detection and volumetric-sizing application is not too complex, the solution does not need massive amounts of compute. “We are not trying to analyze real live video at 60 frames per second. But if we do, we can send that information up to the cloud and utilize Intel toolsets that can render those results for us,” Bell says.

The cloud-based Atajo OnEdge platform ingests and stores data from the gateways for historical analysis. Through dashboards and reports, the platform enables users to track long- and short-term trends.

Endless Applications for Computer Vision

While the Britehouse Mobility/NTT solution is specific to the Impala use case, its beating heart related to machine vision can apply more broadly. For Impala, the team is working on another application that relates to monitoring the safety and site compliance of mobile cranes.

“We have industrialized the hardware and have a software layer that’s highly configurable so we can bolt on modules and deploy specific applications rapidly,” Capazario says. “There are big opportunities for us to improve a whole host of processes.” And implementations do not have to be restricted to the refining industry alone. Manufacturing and other sectors are also ripe for the picking. “It’s almost endless the applications in which it can be used,” Bell says.
 

Edited by Georganne Benesch, Editorial Director for insight.tech.