From personnel-based operations to AI-driven automated maintenance

How can AI and other technologies be used to execute complex projects in an effective manner in the energy sector? During the industrial digitalization conference Ignite, I discussed the potential of AI and machine learning to automate and improve maintenance together with experts from Aize.

The energy industry, and other asset-based industries, share a mutual issue: the time-consuming and costly activity of inspection and detection of rust, degradation, corrosion, defects and other forms of anomalies, all related to asset health and performance in upside as well as subsea locations. Is there a way to plan and execute these activities in a more automated, effective, safe, remote and digitized manner? 

Based on my many years of experience in AI projects with a focus on predictive maintenance, digital twins and operational intelligence projects, I know there are smarter ways. With the help of data, AI and IoT capabilities, complex projects can be executed in a more time- and cost-effective way.

Automated rust detection

One of the companies that Cognizant works with to accomplish this is Aize. They offer a comprehensive digital twin for brownfield assets enabling E2E (end-to-end) functionality and solutions to improve operations, maintenance and integrity management cycles.

“Today, a majority of inspections is performed by people in the field looking at equipment and taking photos, so-called visual inspections, but also by inspectors looking at a video feed or an aerial drone,” says Ronny Øye, Portfolio Manager, Aize.

This is a cumbersome task; the amount of material collected from the field is huge and inspectors spend a significant time reviewing the media, assessing any defect and producing a report. A typical one-day drone inspection could take three, four days to report.

Using AI/ML analytics to identify what photos and video frames to focus on, and further assist in the assessment and quantification of defects, will significantly improve reporting efficiency and data quality. Aize has already developed the first steps towards automated rust detection using a machine learning approach to analyze drone images collected during storage tank surface inspections. 

AI for image analysis

Besides the mundane job of inspecting thousands of images or hours of video recordings, you might also need significant domain expertise to correctly classify anomalies. Fortunately, these tasks are perfectly suited for automation made possible through recent developments within AI and image recognition.
 
“In principle, any object that can be identified by us looking at a picture can also be detected by an AI-based image recognition system. With performance comparable to – or outrivaling – human experts, this means we are starting to see very interesting applications of this technology,” says Vegard Flovik, VP AI & Data Science, Aize. 

One example is within visual inspection and quality assurance. Here, the technology is put to use to scan images and video recordings where it automatically looks for certain defined characteristics such as corrosion, damages, cracks, bad welds, etc. 

One of the challenges is that to train the models to perform well, you also need lots of example images to learn from. Essentially, it does not matter if the AI performance is comparable to human experts if the time and resources required to develop the solution are greater than the potential cost savings of putting it to use. 

95% rust detection accuracy with AI

This problem was approached in a recent project I was involved in. It started with a business value discovery phase, where our experts worked with the client to determine business opportunities, demonstrated the value AI and ML can have on the operations and mapped them with exact AI/ML requirements, and mapped manual steps that can be eliminated/enhanced with AI. 

At a high level, the solution involves capturing frames from drone videos and detecting/measuring surface corrosion on offshore assets using machine learning models. The operator uploads the drone video in a UI interface, and then the Rust Detection API is invoked to perform an automated analysis to estimate degradation. The AI model offers very high accuracy (about 95%) in rust detection and localization down to pixel level, and all information is shared with the field engineers to take appropriate maintenance actions. 

This is just one example of the application of AI. In the end, the user does not care whether you are using some fancy AI algorithm as part of the product or not. They just want a solution that can solve their main pain points and make visual inspection faster and better. This is also one of the key focus areas for us going forward: to successfully integrate all of this as a complete end-to-end solution for visual inspection that is fully integrated with the larger software ecosystem. 

If you’d like to learn more check out our AI offerings.