As a previously agricultural country, Luxembourg has a strong foundation in its manufacturing sector. Industrial complexes laid the groundwork for a thriving economy in the Grand Duchy, and even ranked within the world’s six largest producers of steel just prior to the First World War. While it may no longer be Luxembourg’s primary employment source, manufacturing still represents 8.3% of national employment with over 30,000 in workforce. And while the country began its first industrial revolution back in the mid-1800s, a new revolution is now taking its place that will see our manufacturing sector become smarter and more capable than ever before – Industry 4.0.
As a pioneer in the development of hard materials for machining and wear protection, CERATIZIT is now a global market leader, and has been located in Mamer, Luxembourg for over fifty years. Now based all around the world, their product solutions cover a plethora of industries – from steel and metal processing, to food, aerospace, medical, energy and even watches and jewellery. Earlier this year, they partnered with the University of Luxembourg’s Interdisciplinary Centre for Security, Reliability and Trust (SnT) to investigate how artificial intelligence and data science can be efficiently applied to their production lines, with the aim to improve performance, reliability and sustainability.
With manufacturing machinery equipment, there’s a fine line of precision when it comes to performing maintenance at the right time. If performed too early, as a business you’re not benefitting from the full capacity of the equipment and lose production time due to more frequent downtime, too late and you risk losing production time as a result of more extensive and lengthy repairs. Performing maintenance during an ideal window allows the production line to operate smoothly, as well as ensure that no unnecessary time is lost. Estimating this window of opportunity is known as predictive maintenance.
CERATIZIT is working in collaboration with SnT’s Security, Design and Validation (SerVal) research group to explore how to effectively determine this window of time. To achieve this, the researchers will need to acquire information about the machines, how they operate and whether there are key indicators of when a machine may fail. “Predicting with a good degree of accuracy when a machine will fail means relying on sensors, historical data from the production line and from the maintenance system,” explains Dr. Sylvain Kubler, a research scientist working on the project alongside Prof. Yves Le , Traon, and Dr. Maxime Cordy, a doctoral researcher. “We can then take that data and start to recognise a pattern of when a failure may occur, which can then be developed into a data model,” he continues. Kaouther Benguessoum
However, what comes next is the validation of the model to ensure that it doesn’t change over time. “This stage of development is very challenging, because optimising the system is just one part – assigning the most appropriate technician to work on the fault adds a layer of complexity when you’re dealing with many maintenance tickets and many technicians with specific strengths in their skillsets,” Dr. Kubler continued. Making optimisations where possible in these areas will not just enable CERATIZIT’s production line to become more efficient, it will also have a huge knock-on effect economically. Predicting maintenance will allow them to reduce the costs of calling external maintenance providers, extend the life of their machines by having them fixed at the right time, and ensure the reliability of production is maintained. Machines overdue for maintenance will decrease their overall lifespan and may incur errors during the production process that result in poor quality products, which will then result in the product being discarded as faulty.
This project will allow the applied data science team of CERATIZIT to work with the latest machine learning technology in the field of predictive maintenance including technology directly pioneered by SnT showing a great example of robust industrial and academic collaboration in the Luxembourgish landscape. During this first year, the team will work closely with CERATIZIT to identify that the machines are capturing the right data from sensors and cameras, as well as where new ones can be added to enhance this observation. They will also analyse the existing data to prepare the models that will form the basis of their prediction system in the years to come, as well as ensure that this model can be scaled according to the future growth of the company.
Speaking about the partnership, Dr. Gabriele Pozzetti, Data Science Manager at CERATIZIT, said, “It is essential for companies to have a robust dialogue with academic experts. This is especially true in a field as fast-moving as machine learning. We are seeing these technologies becoming a more and more important part of how we deliver value to our clients, and we believe this collaboration with SnT will strengthen this value creation even further.”