CarnaudMetalbox Engineering and T-DAB develop innovative process monitoring solution

Consumer demands for more sustainable, infinitely recyclable metal packaging in the wake of the single use plastic pollution crisis is driving a need in the industry that CMbE serves for faster, more productive, more efficient and more reliable metal can forming machines.

Simultaneously, these manufacturers are losing experienced operators who can identify problems with these machines to retirement.

CMbE identified the need to support its customers by introducing process monitoring  which would improve the machine uptime to cope with capacity demands and would help mitigate the skills shortages.

Michael Halstead, design verification engineer at CMbE said: “The beverage can Bodymaker is a powerful metal forming process and these machines operate around 350 cans per minute. The harsh low frequency vibrations produced by the machines combined with aggressive metal forming lubricant coolants, present a hostile environment for applying typical sensors or measurement equipment.

“Our project, Canvolution, aims to design, develop, manufacture and commission a network of intelligent Bodymakers capable of monitoring the real-time can-forming process in order to advise operation decision-making to increase productivity with minimal human intervention.”

Working with long-term partner Computer Controller Solutions, specialists in high-speed data acquisition systems, CMbE solved the challenge of capturing the data generated by its machines, and needed a partner who could help with the data analysis.

Halstead added: “When we trialled the data acquisition, we generated a lot of data - 1TB per week, per machine. We needed a partner who could provide an architecture to securely transmit, store and analyse that data, to do some machine learning on that live data in order to predict process change, and to visualise the data we had collected locally and remotely.”

T-DAB’s solution needed to overcome some key challenges including the complexity and expense of moving such an abundance of IoT data from individual machines to a centralised place for machine learning and AI processing, ensuring reliable internet connectivity, as well as data privacy and security.

The key to the solution was to turn the problem on its head. Rather than move all the data to where the ML or AI model is for training, T-DAB proposed moving the model to where the data is generated.

It managed this using an OctaiPipe distributed AI platform and a catalogue of ML pipelines. It means that the data collected from the machine is processed through automated ML models close to the source that can make the predictions. That insight is then communicated to the operator or triggers an automated function of the machine.

Meanwhile, this processed data is batched up and transmitted to the cloud which hosts a specialised time series database called influxdb. There it is able to perform model visualisation and execute model training before being deployed back to the edge.

Having the models at the edge means that the machine can continue providing data even if it loses connectivity to the cloud.

Eric Topham, CEO, said: “These are the critical elements that allow CMbE to have these intelligent Bodymakers where the machine learning can run independently on individual devices, but at the same time we can leverage the data across all of those devices for model training up in the cloud.”

Data analytics is giving CMbE advanced understanding of the can-forming process which will lead to the ability to offer recommendations for optimisiation of machine set-up, to ensure the Bodymaker functions more reliably and consistently.

Data dashboards give operators real-time visualisation of the machine health and process health, giving some indication of when some intervention is needed.

By introducing elements of predictive maintenance, operators know earlier when to replace machine tooling, therefore optimising its usage, when to service machine alignment which avoids unnecessary work, and alert them to imminent process failure, which reduces operator response time and triggers preventative action to reduce downtime.

Halstead added: “Our first foray into machine learning, process monitoring and predictive maintenance on the Bodymaker as part of the Canvolution project has been a step change in our understanding of the can-forming process and the potential insights offer an impressive range of benefits to our business and our customers.”