Two UK metal-forming plants are retrofitting their decades-old machinery with cameras and machine learning to turn manual processes into digitally controlled systems. The problem is that many British factories rely on long-life equipment that was never designed for Industry 4.0—sensors, imaging, or automated control. Without these, manufacturers struggle to achieve the precision needed to enter high-value markets or compete globally. This project tackles that gap by linking real-time camera footage of metal shaping with intelligent algorithms that adjust the process on the fly. If successful, one plant could break into the nickel superalloy market for aerospace and oil-and-gas sectors, aiming for £6.5 million annual turnover from a market that is currently static and low-margin. The other plant, making specialist pipes for offshore energy, would gain the shape-control precision and agility needed to win large international contracts. Beyond these two sites, the approach—digital imaging plus machine learning for process control—could transfer to forming plastics, ceramics, glass, or any manufacturing where shape evolution matters. Secondary gains include reduced waste, energy savings, and predictive maintenance.
View original technical description
The challenge for much of UK industry in adopting Industry 4.0 principles and smart technologies is how to retrofit these to improve performance of existing long life-time plant. The project addresses this using the example of two very different processes for forming of specialist high-value metal products for sectors vital to UK and export supply chains. Both industrial partners operate capital intensive technologies unique in the UK.The common theme between these processes is the challenge to link digital visualisation of the process using camera/imaging technologies during operations with intelligent processing and machine learning to develop on-line control systems with substantially reduced variability and increased accuracy in process control. Process imaging is not currently part of either process, yet accurate control of shape evolution and the process parameters needed to achieve this are vital to both. Work includes process characterisation, creation of digital twins and then intelligent interactive process models which will be used with additional plant sensors and instrumentation to demonstrate benefits in industrial trials.In one case the aim is to improve process control and predictability sufficiently to allow entry into a completely new high value market (nickel superalloys for specialist engineering, aerospace and oil and gas sectors) with an initial aim of £6.5M pa turnover (~1% of global market) and 6-7% predicted annual growth as opposed to static low margin markets currently served. The change is radical and with only current standards of process control the plant fear risk of serious damage to equipment if following traditional development paths.The second sector is specialist large pipe for on and off-shore applications. Business is highly competitive but has strong global growth and is typically won by bidding for large projects. Success involves ability to demonstrate high precision on shape control together with agility in adapting to new requirements. Digital process imaging linked to intelligent control will bring major competitive advantages.Secondary advantages will arise from improved control leading to yield and energy benefits, reduced waste and opportunities in predictive maintenance.Whilst the demonstrator plants are in the metals sector, the approach and technologies should be readily transferable to other processes where digital imaging can be linked to machine learning and intelligent process control, also presenting opportunities for technical delivery partners. Examples might include forming processes in other materials (plastics, ceramics, glass, other metals) or diverse other processes where evolution of shape and position are important.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know