It will be hard to look at the humble cylinder in quite the same light again.
Artificial intelligence is reshaping how industrial gas companies manage their cylinders, with new technologies enabling smarter registration, improved traceability, and predictive maintenance regimes.
Tracking software specialist TrackAbout and distribution software company Datacor partnered on a recent gasworld webinar and are applying AI to longstanding pain points in cylinder lifecycle management. It is helping reduce errors, improve data integrity, and drive greater operational efficiency.
Elizabeth Wallace, Vice-President of Product at TrackAbout, outlined how AI is being embedded into the registration process from the outset to eliminate manual inputs and ensure accurate data from day one.
“The most important part of asset management is knowing where your assets are, and using them as efficiently as possible,” said Wallace. “So we are trying to remove human error at the point of registration by using AI and visual recognition. It all starts with a picture.”
“We want to focus on reusing our assets, and how can we go faster and track patterns – and that begins with a visual reference,” she added.
Instead of relying on manual entry of product codes or attributes, the company is developing a process in which an image of the cylinder serves as the baseline record. This enables better validation, makes re-identification easier if a tag is lost, and reduces inconsistencies in master data that can cascade through a customer’s system.
Each cylinder is stamped with a range of data points, such as owner, manufacturer, barcode, valve information, serial number, weight, capacity, pressure and standard.
“We don’t want someone putting in the wrong product code or selecting attributes that don’t apply. That creates risk, and this is data that is used across the business. So our goal is to get it right the first time, accurately identifying the cylinder from a visual from day one.”
Wallace also highlighted the issue of serial number duplication, explaining that cylinder serial numbers are not always unique across manufacturers.
“If I have a visual reference, we will be able to tell if it’s a duplicate,” she said. “AI will be able to use historical data, and use the information that we know on cylinders, and tell us exactly what the next steps are.”
Lou Zhang, Senior Lead of Data Science at Datacor, explained how the AI model uses deep learning and computer vision to identify key information on cylinders, including stamped markings, serial numbers, and retest dates, directly from photographs.
“Gas cylinders are particularly challenging because the text is curved, embossed, and often worn,” said Zhang. “But we trained our model on tens of thousands of images, feeding it data so it learns to read cylinder markings with high accuracy, even in poor lighting or in cluttered environments.”
The system uses segmentation – a software process that identifies and separates objects or regions within images or data – to isolate the cylinder, strip out the background, and then create heat maps to identify likely areas of text. These snippets are then analysed and parsed into the correct data fields within the TrackAbout application.
Zhang said the aim is to reduce manual workload and error rates by up to 90%, particularly for large distributors managing thousands or millions of assets. “Instead of manually walking around and inputting 10 cylinders, one photo will be able to log an entire crate,” he added.
Wallace added that the benefits go beyond registration. AI tools are being used to analyse cylinder images for signs of degradation – such as oil, arc burns, or surface damage – enabling maintenance decisions and reducing safety risks.
“We’re building these capabilities into ruggedised devices and smartphones, so operators in plants or drivers in the field can use them in real time,” she said. “It’s about using AI to make faster, better decisions, whether that’s maintenance, relabelling, or compliance.”
AI is also being applied to track patterns and flag inconsistencies in cylinder movements. Wallace highlighted how TrackAbout’s system can now identify when steps in the cylinder lifecycle have been skipped, such as a missed pickup or an unrecorded delivery, and automatically generate corrections or push notifications.
TrackAbout uses cloud-based software to digitise manual, paper-based processes. ©TrackAbout
“We know the real world is messy. People forget to scan assets or skip steps. Our software can identify where that happens and apply rules to fix it. But more importantly, we can show our customers why it’s happening, who’s doing it, and how to stop it.”
Wallace, who has worked in the cylinder business since 2002, said she hasn’t seen as much change as in the last two years.
“Datacor is making huge investments in technology and AI tools and we’re going to harness those as much as we can to make a better product for our customers,” she added.
Looking ahead, both Wallace and Zhang see further potential in integrating AI into customer-facing portals, giving users the ability to generate reports, forecast demand, and interact with their data through natural language interfaces.
“It’s about giving customers more control and making the tools intuitive,” said Wallace. “We want to empower users to get the answers they need quickly. This is where AI is heading.”
To watch the webinar on demand, click here
