AI in maintenance

woman in hard hat controlling machine via holographic remote

AI in maintenance: “You have to start slowly, like jogging”

The better and more efficiently machines and systems function and are operated, the more economically manufacturing companies can work. But this in turn depends entirely on plant availability, which can also only be ensured with functioning maintenance. AI and machine learning processes can help here and create concrete added value. Predictive analytics and predictive maintenance can be the first concrete steps in this direction.

Due to the increasing digitalization of business processes of all kinds, companies are now facing numerous new challenges. Some of the most important requirements in this context relate to the availability, safety and efficiency of machines and systems. At the same time, however, companies want or need to reduce their maintenance costs. Many experts see the use of AI in the maintenance sector as a way out of this dilemma. André Panné, Managing Director of Rodias GmbH, can confirm this: “Availability, safety and efficiency are the result of well-functioning maintenance. So the answer is actually yes. In reality, however, we see that maintenance in particular – beyond fancy small prototype projects – is not included in the budget planning for AI projects.” Maintenance IT traditionally lives on an island – with little or no connection to production.

Avoiding production downtime with AI solutions

The first step for specialists like Rodias is to use a change management approach to explain the current situation to those affected and find ways out of this silo mentality. The technical solutions to improve the maintenance IT and OT world with AI are available, but “unfortunately, the necessary mindset is not yet” (Panné). However, unplanned downtimes and the associated unplanned production downtime can be avoided with AI solutions.

“Depending on the industry, such incidents often result in costs of several thousand euros per hour,” says Guido Reimann, VDMA, Software and Digitalization. “With a data-based approach and permanent monitoring of critical components, maintenance can be carried out in a more targeted manner and unforeseen failures can be significantly reduced.” This is complemented by simplified operator guidance or assistance functions when setting up a machine or system based on AI solutions, “in order to avoid set-up times and unnecessary rejects” (Reimann). However, the path to a predictive maintenance solution can also be paved with stumbling blocks. Rarely does the first attempt lead to a finished solution, Reimann continues.

“Artificial intelligence contributes to improved competitiveness”

According to Stephan Bloehdorn, Associate Partner Industrial Sector & Practice Lead AI, IBM Consulting, conventional maintenance concepts – such as plant maintenance that only takes place when a particular plant is already defective – are no longer up-to-date or competitive: “The manufacturing industry is particularly characterized by global competition. This results in increasing requirements such as improved throughput times and greater flexibility in production processes – while ensuring outstanding production quality, of course. Artificial intelligence can make a significant contribution to improved competitiveness in all of these areas.”

“Significantly improve predictive maintenance with AI”

AI concepts that are increasingly seen as the ideal solution in the field of modern maintenance include predictive maintenance and predictive analytics. M.Sc. Nils Thielen, Chair of Manufacturing Automation and Production Systems, Friedrich-Alexander-Universität Erlangen-Nürnberg, can confirm this: “Especially if a predictive maintenance strategy is being pursued, it is worth considering AI. A good predictive model makes it possible – at least in theory – to take into account not only the current condition of the equipment, but also the work that is still planned for it. This means that the plant’s wear reserve can be used up as much as possible without leading to a breakdown.”

Rodias CEO Panné points out: “First of all, predictive analytics and predictive maintenance are not two different things. Predictive analytics is the generic term or process that can also be applied to many other areas of application. Predictive maintenance is based on predictive analytics. This approach is also not a major new invention, even if it can be significantly improved by using AI.”

“Prepare human decisions better and make them more reliably”

AI methods, especially machine learning, are already making it possible to map the behavior and interaction of the links in this chain much more precisely than before. “This means that human decisions can be better prepared and made more reliably because the AI system can now suggest the very best actions with priority from the variety of possible actions,” says Markus Ahorner, Managing Director and Technical Director of Ahorner & Innovators GmbH. “In some cases, these decisions are even automated.” Contrary to the fears of AI sceptics, humans will not be rendered redundant, quite the opposite. The machine enables them to work better and in a more targeted manner than before. And of course, in this context, AI can also become a means of combating the ever-increasing threat of a shortage of skilled workers.

Guido Reimann also clearly sees the benefits of AI in maintenance: “Continuous analysis of sensor values on the machines and systems gives the machine operator a much more accurate picture of the condition of their system.” In this case, if there is a sufficient amount of data and machine failures, the maintenance technician can also use machine learning or AI processes to predictively determine the maintenance intervals as required. A positive side effect is that cooperation between the operator and the machine manufacturer means that experience from other machines used by the same manufacturer can also be taken into account when creating the AI algorithms.

“The application scenarios are diverse”

Stephan Bloehdorn from IBM knows what AI can already achieve in maintenance practice today: “The application scenarios are diverse. The British consumer goods manufacturer Reckitt, for example, has set up a ‘Factory on the Future’ with IBM Consulting and technology partner Microsoft Azure, in which data analysis and AI form the basis for a 10% reduction in maintenance costs, among other things.” Niels Thielen from Friedrich-Alexander-Universität Erlangen-Nürnberg adds: “In practice, existing solutions are already based on assistance systems. For example, recommendations can be made for when to carry out repairs or maintenance. Different data can be used for this, such as the structure-borne noise of tools.” However, the decision as to whether the measure is actually implemented usually still lies with the person. Another option in this context could be for an AI application to search historical data for similar events in the past in the event of a fault and indicate the appropriate solution.

On the AI journey with condition monitoring

 “AI or machine learning is already a reality in various areas of application in mechanical engineering,” says VDMA man Reimann. According to experts, AI and algorithmic approaches are already making very successful contributions to system availability and preventive maintenance in machines and systems that continuously supply data in continuous operation. “Sometimes companies are not yet ready to use a neural network right away,” says André Panné. Sometimes companies still use the natural neural network of the plant or machine operator. And André Panné adds: “This is often the most charming approach to embarking on the AI journey.” This helps to reduce fears and fears about the future. This reduces fears and reservations and creates access to the data that can subsequently be used with AI.
Rodias has implemented this at Hans G. Hauri KG Mineralstoffwerke, for example, which operates a lime shaft kiln. If critical components fail at Hauri, it becomes very expensive. It has therefore paid off for those responsible to monitor the plant using a condition monitoring system. “The next step now,” says Panné, “would be to run an AI over the collected data in order to be able to evaluate even more data details and anomalies. You have to start slowly, like jogging.”

Would you like to discuss this topic directly with our experts?

Contact Us