Seventy years ago WW2 code-breaker and founder of computer science Alan Turing proposed that a ‘learning machine’ could become artificially intelligent. Some advances are starting to show promise. However, as a sub discipline of informatics, there’s still a lot of deep learning and artificial neural networking to undertake before the latest generation of injection moulding machines can fully predict and react autonomously to a given production situation.
Several Sumitomo (SHI) Demag automation experts shared their insight and the AI advances that are emerging at an inaugural online workshop hosted last month by Polymer Technology Ireland and the Institute of Materials (IOM). With machine learning at the heart of future autonomous and intelligent systems, understanding the data independencies and interoperability will be critical, highlighted presenters Dr. Melanie Rohde-Tibitanzl and Dr. Thorsten Thümen.
AI has now leapt beyond academic curiosity and has begun edging towards reality. New research by the World Economic Forum predicts that by 2025 the division of tasks between humans and machines will be equally split, with reduced demand for people globally to perform operational, machinery repairs, assembly and stock keeping duties[i]. Additionally, the current COVID-19 pandemic has seen a rapid digitalisation of work processes.
Handling large data pools
To accelerate the development of autonomous machine learning in a moulding production environment, Sumitomo (SHI) Demag has been busily working behind the scenes with a consortium of scientific academics and technology pioneers. This has resulted in them uncovering what they describe as the first building block of the machine learning process – a digital shadow. It involves collating vast amounts of data from numerous sources, which in most instances would be beyond a human’s cognitive ability to analyse.
UK Managing Director Nigel Flowers puts it into context. “Currently, the quality of injection moulded components and process stability remain highly dependent on the knowledge and experience of a machine operator. However, there are complex interactions between part quality and machine process parameters, as well as other fluctuating and influential factors such as melt temperature, raw material batch fluctuations or inhomogeneous tool temperatures, which aren’t captured and analysed through a single data source.”
Although intelligent, smart machines clearly have a role to play, interoperability and connectivity between different technologies is a critical stepping stone. For a computer to truly imitate human behaviours and deduce why an incident has occurred or what will happen next, it needs the know-how and connectivity between all of the machine assets.
Pushing for standard interfaces
One of the biggest obstacles to machine learning and digital factories of the future is not having a standard interface. “For seamless communication to take place across machinery assets, all of the elements, including sensors and processing data, needs to talk to one another in the same language. That’s what OPC-UA seeks to accomplish – a universally compatible digital interface,” notes Nigel.
The basic OPC-UA protocols covered by Euromap 77 are now in place to facilitate standard data exchanges between injection moulding machines and Management Executive Systems. Also in development are Euromap 82.1, allowing for the moulding machine to talk to the temperature control assets, and Euromap 82.2 which communicates data between the hot runner controller and machine.
Leaping into the Industry 5:0 world
Nigel states that OPC-UA interoperability has already started to generate valuable pellet to pallet processing data. However, what’s coming down the line next really excites the team, where man and machine are reconciled to deliver intelligent automation.
A project initiated last year at K-2019 in collaboration with Institute for Plastics Processing (IKV) has already started to develop a neural network of data and calculate the optimal process settings on an IntElect manufacturing cell. This short pre-study project proved that with the automated processing parameters, machine set up time was 4.5 times faster and generated 78 percent less start-up waste.
Next, Sumitomo (SHI) Demag is seeking to create a totally open IoT platform that will allow the introduction of third-party solutions. Thorsten comments: “We are collaborating with a German start-up to build out the framework for this open platform, which will feature simulation results, processing data, operative feedback and eventually AI optimisation guidance. The base has to be carefully selected, as we see this being a constantly growing platform that will integrate data from multiple sources and enable customers to add their own topics, including KPIs.”
Of all the AI disciplines, machine learning for the Sumitomo (SHI) Demag scientific thinkers is the most exciting and beneficial aspect for moulders. “Until machines are able to understand what and why something happened to data being processing, and then have the introspective ability to rationalise and make a decision autonomously based on that data, no one can really claim that they have cracked AI.
“However, machine learning is clearly less to do with mathematical modelling and more to do with compatible interfaces across all machine assets. In that respect it becomes an all-purpose AI. Having all machinery assets talking together on the same platform is like singing from the same music sheet. Which ultimately delivers time saving and productivity benefits to moulders,” ends Nigel.
[i] Source: Future of Jobs Report 2020. World Economic Forum
