Christian Casazza, customer service director of the Durst Group, explains how its new AI-based Premise project is expected to advance predictive printer maintenance in the near future.
“Hello. In the next few days your printer needs to be serviced, otherwise your machine risks a breakdown. Let’s discuss now to find a resolution.” Does that sound unrealistic? It shouldn’t. You can expect predicting printer misbehaviours to become commonplace - and Durst is developing a technical infrastructure with machine learning technologies that means predictive maintenance measures in production environments will soon become a reality.
Our aim is to have no unplanned service requirements by 2025. A tall order perhaps. However, it was with this in mind that in February this year we announced the start of the ‘Premise’ project with the Free University of Bozen (unibz). It’s funded by the European Union and also involves TechnoAlpin, the manufacturer of snowmaking systems.
Sensors have been used in our machines for years to collect the data necessary to run them. Now they have an additional important role for collecting information for predictive maintenance.
Historical data from customers across the world has been collected (with their agreement, of course) via our Durst Analytics software for 18 months. This is stored on a database at our headquarters in Brixen, Italy. And now as part of the Premise project with unibz we will be looking to ‘train’ the algorithms to work out a predictive model. We will then be able to say with some certainty what will happen in real-time in terms of the potential for breakdowns.
The analysis is based on a framework for AI [Artificial Intelligence] or machine learning prediction analysis. This provides the basis for the Premise project - headed by Johann Gamper, professor and vice rector for research at the Faculty of Computer Science - that involves students and experts from the university. It means working closely with the university to get these prediction models to foresee breakdowns, misbehaviour, when necessary maintenance is required, and so on…
The customer always has the right to say they don’t want to share the information - even when it’s anonymised, as it always is - and, of course, we respect that decision.
As already indicated, our core strategy statement is to have no unplanned service by 2025. By constantly tracking and analysing sensor data and parameters, we will be able to determine whether a part will fail before it does so. This gives us the time to proactively organise with our customers a service intervention according to their production planning.
For example, by measuring the power consumption of a motor or measuring the differential pressure, conclusions can be drawn about the condition of a spare or consumable part.
It’s a service that has been largely reactive - a customer opens a job ticket or calls to let us know about a problem and wants us to fix it.
The accuracy of predictive maintenance is something we are going to have to figure out during the project. In most cases, we believe the predictions will be extremely accurate. For a low-cost component, it is better to change it rather than risk a breakdown. But for an expensive component, a customer probably doesn’t want to replace it until the last moment.
There is a trade-off between substituting something too early against being too late.
With this project, we believe we can test technologies that we have been researching with our industrial partners on the basis of specific case studies and adapt them to specific requirements. This Premise project will enable us to go one step further and use artificial intelligence methods to make these predictions and interventions even more efficient and to be able to apply them to complex, causal relationships.
We have done our homework, the foundations are in place and now we’re ready for the next step.
The Premise project is part of the much bigger plan and a further component in our vision of a smart factory, where networked infrastructures, intelligent production systems and innovative software enable an automated business process. It’s been accelerated by the pandemic because a lot of the traditional service intervention is based on the experience of technicians and knowledge gained by regular servicing of the machines. However, that close contact between the technician and the machine is being lost when international traffic is limited so we have to substitute, to a certain point, the competence of a service technician by a clever, intelligent algorithm.
The predictive maintenance developed in the project framework and the machine learning techniques used will in future trigger the maintenance of the printing systems independently in order to guarantee predictable and trouble-free operation. It’s difficult to gauge when there will be a full roll-out. However, I expect we will have our first user cases by the end of next year. And who knows, perhaps the first one might be a success story from the UK!