Predictive maintenance is a key aspect of what is known as “Industry 4.0”, that is the digitalisation of industrial processes. The concept of predictive maintenance is when data from industrial plants and their environment are evaluated using AI, machine learning and other complex procedures, to determine the optimum time for maintenance. In other words and somewhat disparagingly formulated: the old analogue approach of saying “The machine sounds odd, let's check it out” is replaced by a comprehensive digital analysis of all the available data relevant to the operation of the machine.
Dr. Matthias Haun, Professor of Cognitive Cybernetics and the Philosophy of Cognitive Sciences, is an expert in predictive maintenance. He is Head of the “Predictive Maintenance” project in the Faculty of Electrical Engineering and Information Technology at Offenburg University of Applied Sciences. Funded by the Carl Zeiss Foundation, he and his team are working to make predictive maintenance a market-capable “solution” – especially for small and mid-sized companies/businesses. Together with his colleague, David Gelantia (vice project manager in the “Predictive Maintenance” project) Dr. Haun explains what he expects from predictive maintenance and what opportunities he sees in this approach. Profitability is often the first issue raised in these projects.
What are the benefits of a very complex and time-consuming approach compared to the maintenance process used to date that adheres to fixed intervals and only deviates from the maintenance rhythm in the event of “problems”?
“Fixed maintenance intervals do not always make sense, because they do not correspond to the real maintenance requirements of an industrial plant. Regular maintenance is either done too often or problems occur because the next maintenance date is too far into the future. Both are associated with significant costs,” explains Dr. Matthias Haun, continuing: “You cannot underestimate the cost of maintaining an industrial plant. The entire system needs to be shut down for days to maintain a large production plant. The entire place is shut down for three or four days. The cost is enormous. The same holds true for a fault.”
How can predictive maintenance prevent high maintenance costs?
“Predictive maintenance is based on a multimodal analysis of all the relevant machine, process and environmental data. The aim is to prevent malfunctions as well as unnecessary regular maintenance. Maintenance is only carried out when it is really necessary, and ideally before a fault can occur. That saves billions. We currently assume that in many cases up to 30 percent of maintenance costs can be saved with predictive maintenance. This is a massive cost reduction,” explains the expert Dr. Haun. Mr. David Gelantia adds: “If it is possible to act before a fault occurs and prevent machine downtimes, then a company can achieve a crucial competitive edge in the long term.”
Analysis of machine and environment – sounds complex, doesn't it?
Professor Matthias Haun explains: “Yes, it is extremely complex, which is why the data cannot be calculated using traditional methods. It requires new methods from artificial intelligence research, such as model learning, AI etc. to be able to make valid predictions. The complexity is enormous. And it requires a large amount of data from a wide variety of sources, such as status data, environmental data, weather data and management data.” David Gelantia adds: “We are not just analysing a number of factors that we collect from the industrial plant, but also factors from the spatial environment, such as room temperature and humidity.”
Is a project, such as this, actually financially viable for SMEs?
Cybernetic expert Matthias Haun is certain: “In all cases, predictive maintenance also makes sense for SMEs and especially small companies. The bigger companies have been active in this field for some time; the opportunities are clearly evident to them and the investment is not a problem. It is not so easy for small and medium-sized businesses. It requires a comprehensive preliminary analysis to determine the investment associated with predictive maintenance and from what stage it becomes worthwhile. But SMEs, in particular, cannot afford high maintenance costs nor many failures. But it is, of course, a major obstacle to buy and implement a project of this size. A lot of preparation and persuasion is needed because its success often only becomes apparent after three to four years – a very long time for companies.”
For whom does predictive maintenance work and when does it make sense?
“Actually, predictive maintenance works for every company. Of course, it's easier with new machines. I also work in Asia, but that's not comparable: Over there they often start on a greenfield site and build the industry fully digitally in accordance with the latest standards. In Germany, on the other hand, we often have to deal with decades-old plants, which do not initially deliver data. Measuring sensors and data transmission systems therefore need to be retrofitted,” continues the 56-year-old professor and entrepreneur.
Is it possible to retrofit old machines digitally?
“Yes, fundamentally always. There are now a number of sensors that can be attached to the machines – often, as it were, “with minimum invasion”. For example, measuring sensors for temperature, humidity, pressure etc. can be subsequently installed. Audio measurements are also taken. In some pipework systems, you can hear whether everything is OK by taking an audio measurement. We do not leave out a single machine. But determining the machine data is not always the simplest part of the project,” explains the computer scientist and philosopher.
However, the technical equipment per se fitted on the machines is only a small part. Predictive maintenance – like many aspects of digitalisation – goes hand in hand with a fundamental change process. You need to change the processes, experts qualified in the subject who understand what is being done need to be present on site, and the environmental data needs to be recorded etc. According to Professor Haun: “Setting up a project of this kind in an existing company is extremely complex and all-encompassing. Everyone involved needs to be on board from day one, and success will not come from today to tomorrow. The management therefore needs to be convinced that predictive maintenance is a worthwhile undertaking. That is often not the case today.”
Does predictive maintenance with good technicians not work on gut instinct as well?
Matthias Haun and David Gelantia really do not underestimate the gut instinct of experienced employees. In their experience, the human factor is an essential factor in predictive maintenance: “Good technicians really know a lot about their machines. They also have an insight into the environment: Which employees are working on the plant? How good is their qualification, and how careful are they with the technology? How clean is the production plant? How careful are bosses and managers? What quality of material is used? All of these factors play a key role and over the years lead to a wealth of experience lacking with a machine-based approach." It was therefore clear to the team in Offenburg that this knowledge had to form part of their approach to the project: “At the outset, we conduct workshops with the various teams and technical experts to integrate this knowledge into the project. This human knowledge is formalised and fed into our database. This is known as knowledge engineering. After all, the pure measured data alone is not enough to bring about optimisation on a large scale. Experience-based knowledge needs to be included as well.”
Overall, the approach taken at the Offenburg University of Applied Sciences is very comprehensive and therefore particularly complex. It does not make acceptance any easier, but is of particular importance for SMEs, for two reasons: “On the one hand, a project of this size is only worthwhile if it ultimately really involves long-term savings. Secondly, SMEs are losing more and more of their experienced experts to retirement. There is a lack of young talent due to the shortage of skilled workers.” This means that new ways need to be found to replace today’s “gut feel” approach.
Predictive maintenance – yet another datenkraken?
Listening to Dr. Matthias Haun and David Gelantia, it becomes clear how comprehensive the data analysis is on which predictive maintenance is based. Is there any willingness in small and mid-sized businesses to share this data and provide such a deep insight?
The willingness to do so varies, in the experience of the Offenburg researchers: “There is, of course, resistance and concerns, as with any drastic change. But many fears can be overcome by bringing those involved on board. Works Councils, for example, were often very successfully involved in the projects. As far as data is concerned, most companies want to keep it in-house and evaluate it themselves. This is also possible, but requires the necessary knowledge and expertise. However, outsourcing the data to an external service provider is also an acceptable option for many.”
The major challenge, according to Professor Haun, is not so much the amount of data that needs to be visualised, but the expertise required to work with this complex database: “The most difficult thing is reading the data and using it meaningfully. We need competent experts for this. It is not our goal to hand companies a smart dashboard and then leave them alone with it. We want to get predictive maintenance up and running in practice and show the company management the Return on Investment (ROI) based on clear results. This is the difficult and long-term part of the project.”
How successful has the Offenburg University research project been so far?
Dr. Matthias Haun is satisfied with the current status of the project: “We have set ourselves quite a task. We do not just want to create a product or service package that can be implemented for mid-sized companies. We would also like to get actual projects “up and running” and achieve our first successes. We are very much still in the early stages at the moment but we are very optimistic about the future. We have already secured our first partner companies. We believe that our approach, which is very complex and difficult to communicate, is nonetheless a good fit for companies. With every company we examine very closely whether predictive maintenance can be implemented and in what form. Only if we are convinced that it can be successful will we set up the project.”
Predictive maintenance is not yet very high on the agenda of many mid-sized companies. Many companies are still very cautious as this is a very comprehensive topic, which involves a lot of effort and where the ROI is only evident after some time. “There are no shortcuts with predictive maintenance. The data needs to be meaningfully systematised and collated in a structured manner to “feed” the system.” And, according to the two experts, this system would have to be running for a while before the first results appear. Nonetheless, a solution is to be developed by 2023 that is workable and specifically tailored to SMEs.
The goal is attractive – for insurance companies too
Despite all the obstacles and difficulties, the team at the Offenburg University of Applied Sciences believes that predictive maintenance will also prevail for SMEs: “The goal we are aiming towards with predictive maintenance is important and the right one. Older industrial plants, in particular, need to concern themselves with reducing downtimes and remaining competitive. It is not just the financial savings that are enormous – production reliability also increases. Both are essential for the financial success of industrial companies.”
When asked whether predictive maintenance is also an issue for insurance, Dr. Haun answers with a persuasive smile: “If interruptions to business are reduced, the cost of the corresponding insurance can also be reduced. That would be worthwhile for companies and insurers alike. I would be very excited to talk to ERGO about this.”
Industry 4.0: What does the future look like?
Dr. Haun and David Gelantia are certain of one thing: “Digitalisation issues pose major challenges for SMEs.” They are almost always associated with a change process, and almost always deal with complex and fundamental changes and interventions in existing processes. On the other hand, the need for action is growing every day: “It is not uncommon for us to get technical staff back from retirement to incorporate their knowledge in our project. However, it is not always easy to motivate someone to come back to the company to share their knowledge and expertise with us. And it is not a solution in the long term.” The pressure to pursue technically new paths is rising daily. “This is currently giving greater impetus to the issue of predictive maintenance, which is not so new after all.”