Digital Technologies were among the Hot Topics at glasstec 2024. Hardly surprising because this topic has now become unavoidable also in the glass industry.
A look at how deploying AI works in practice provides exciting insights: in glass production highly advanced AI-assisted systems are already deployed today to complement or even replace conventional test methods. A company very much at the forefront here is Cerrion AG.
We spoke with Nikolay Kobyshev, Founder & Chief Technology Officer (CTO) at Cerrion AG and speaker at glass trends live 2024, about the specific challenges associated with AI-assisted defect detection. How does Cerrion deploy the new technology, how do AI-assisted systems differ from conventional test methods and which future developments does he expect for the use of AI in the glass industry?
Which are the specific challenges posed by AI-assisted defect detection in glass production at present?
The monitoring of glass forming processes is extremely complex. Due to the hot and dirty environment the installation of traditional sensors or standard quality-control cameras is not practical or even impossible. At Cerrion we use off-the-shelf CCTV cameras that monitor up to 5 sections of IS machines from a distance of several metres. For carousel machines one or two cameras are sufficient to cover the entire unit. Each camera observes the process from a different angle, with frequent visual distractions such as maintenance work or wobbling of sections. This is why the AI system needs to be extraordinarily robust and have exceptional powers of generalisation in order to understand what is actually happening in the forming area.
Since the majority of production cycles run normally, the AI has to maintain an extremely high level of precision. Even at a 99% precision rate, one in every hundred recognitions would trigger a false alarm that might disrupt operation. This is why systems in practice require almost perfect reliability (99.9999… %) to offer true industrial added value. This is very demanding because the system not only has to avoid false alarms but also be extremely sensitive to detect every individual process deviation.
Add to this that glass is very hard to monitor because it is very dynamic: from the gob to the finished bottle or wine glass the material shape and colour change in a matter of seconds. The adverse manufacturing environment makes it all the more difficult to keep the cameras clean at all times. Therefore, AI systems have to be rugged enough to also operate reliably with soiled camera lenses.
What are the experiences of companies already cooperating with Cerrion? What makes their technologies so special?
Cerrion supplies AI agents that automatically detect different production anomalies, be these production issues, quality problems or unsafe situations. Once a production issue has been identified the agents automatically carry out actions: activating blow-out devices for smoke extraction or stopping a machine or section or notifying the operator via loudspeaker, alarm displays, WhatsApp or MS Teams. They feature automatic escalation: if a section crash remains undetected despite an alarm, for example, the system automatically performs a safe stop to prevent an escalation. Needless to say, users are in complete control of the AI-agents’ behaviour and escalation paths via the Cerrion Dashboard.
Results are impressive: our customers have registered a 90% reduction in safety and fire risks as well as a reduction of losses at both the hot and cold end of more than 34%.
In addition, Cerrion delivers root cause footage of the event including a 30-second context before the incident for each production anomaly detected. This allows the factory teams to not only establish when problems occur but also to visually analyse and better understand the root causes. In a typical factory, staff would check up to 100 of these videos per day, which in turn substantially increases their analytical skills.
How do AI-assisted systems differ from conventional test methods?
Traditional test systems measure specific product properties such as bottle height and compare these with the pre-defined limit values for each product type. By contrast, AI-controlled systems learn independently what is an acceptable production quality while only requiring minimal user input. AI systems adapt automatically and instantly to new products so that no manual recalibration is required for product changes. Since AI systems do not depend on fixed measurements, they can detect a far wider variety of defects and production anomalies. With conventional test systems it is impossible to detect section crashes using simple CCTV cameras – with AI this is easily possible. Cerrion successfully detects section crashes with over 50 production lines worldwide!
Since AI systems have a deeper understanding of the underlying reality, they react less sensitively to input conditions. AI-based systems can use low-cost and generally available hardware such as simple CCTV cameras thereby rendering superfluous expensive and space-consuming devices for ensuring optimal test conditions. You just have to install a CCTV camera in the relevant area and the system is ready to operate. No need to worry about clean backgrounds, optimal lighting conditions or perfect angles of vision.
What future developments do you expect for AI in the glass industry?
In the near future, AI will cover more and more manufacturing areas. Since it understands manufacturing processes in an increasingly holistic way this will lead to the development of AI agents. Such agents, as already offered by Cerrion today, work like humans: they require simple but comprehensive input such as videos of CCTV cameras to enable them to autonomously draw conclusions about what is happening. They can comprehensively monitor the complete production area and in the event of anomalies either act automatically or call humans for support. This gives rise to more efficient HR management, enabling fewer operators to react faster to the problems detected by AI in various areas of production.
Cerrion now already sends real-time notifications with images by WhatsApp or MS Teams direct to operators’ mobile devices and automatically transmits signals to machines – for instance for a regular stop in the event of malfunction. This enables fewer employees to manage bigger areas more effectively.
The rising availability of detailed production data through AI-assisted monitoring will give rise to a holistic understanding of manufacturing processes, so that companies can identify and solve problems at the root.
In addition, AI systems will increasingly control manufacturing processes where each product is unique, for instance when pulling the stems of wine glasses. Currently controlled by manual inspection, these processes will in future benefit from the AI’s ability to generalise learnt patterns thereby making for effective, automated monitoring.
