Imagine a technician on board his vessel in the middle of the ocean, facing a problem with the engine - an alarm flag was raised in the digital system. He takes a photo of the control panel with his mobile phone, uploads the image to the CEON TechBot, and gets an answer within a few seconds.
It may say “Based on the displayed alarm code, you’re experiencing an issue with the engine start attempt. Recommended actions: Ensure turning gear is fully disengaged.” Or otherwise: “Verify control air pressure is above minimum threshold.”
“That’s what the future looks like,” says Bernd Eberwein, Head of Digital Services 4-Stroke at Everllence in Augsburg, Germany. “AI is a really cool tool that is developing at the moment, and we’re only starting to scratch the surface about the values it can provide.”
In practical terms, “scratching the surface” refers to the recent soft release of the CEON TechBot which already makes it much easier for customers to get answers to technical questions. It is one of many AI capabilities Everllence is currently integrating into its digital service solutions portfolio.
The company prioritizes safety while maintaining reliable performance and high operational speed. In industries like power supply and shipping, safety is paramount because a wrong technical decision can result in costly downtime or even potential hazards. At the same time, the team is keeping a close eye on the rise of AI - while its potential is clear, the challenge is to separate real value from the inflated expectations.
A technician captures an alarm message on a control panel to receive immediate guidance via the CEON TechBot. © Stefan Hobmaier
Where do I look for quick answers?
Jorinna Gunkel from the Group Digital team, responsible for the release of the CEON TechBot, explains what advantage it can bring: “We have customers that are sailing around the globe operating these huge engines. And if something goes wrong, you are faced with a shelf full of documentation where you have to search for the right explanation in order to find out what to do.”
Retirement waves of seasoned experts, the pressure on younger staff to master complex systems quickly, and the relentless pace of product innovation all combined, leave customers increasingly unsure about the right course of action.
“This is where the CEON TechBot steps into that journey and helps customers help themselves to receive a response on the spot,” says Gunkel. Everllence consolidates millions of pages of manuals and documentation, as well as sensor data and service reports, so customers can simply ask the CEON TechBot questions, interacting with an AI-assistant, instead of manually searching through files. "If it becomes necessary to consult an expert from Everllence, our customers can always conduct a more informed call to receive support by a human."
“The CEON TechBot helps customers help themselves to receive a response on the spot.”
Jorinna Gunkel, Everllence Senior Digital Program Manager
No AI hallucination allowed
As demonstrated over the centuries-long legacy of Everllence, safety always comes first. “Our CEON TechBot is an assistant: A customer can choose to use this tool, but it doesn't take decisions,” says Gunkel. “And it knows only what to do to the extent that we have given it information.”
The most important distinction is precision. "By only allowing access to certain documents for that specific engine, we can assure there are no hallucinations and avoid misinformation from the open internet - to provide advice our customers can rely upon,” explains Bernd Eberwein, reminding us that ChatGPT sometimes is giving absurd or “statistically true” answers when it lacks real information. “If the CEON TechBot doesn’t have the information, it will answer ‘I don’t know’,” says Eberwein.
“So far, this approach resonates well with customers who seek faster technical guidance while staying cautious about the accuracy of general chatbots. We don't want to bring something to the market, which doesn't have the same level of carefulness that we also apply to our engines.” And ultimately, AI will never be the sole solution in service, but always one of several tools supported by Everllence experts and their knowledge.
Putting cybersecurity first
All Everllence products are equipped with several hundred sensors constantly measuring temperatures, pressures or vibrations. When they are connected to the Everllence CEON cloud, PrimeServ Assist experts can give real-time advice to customers based on this data.
Eberwein emphasizes the significance of cybersecurity. Everllence already collects global asset‑performance data via its CEON platform, yet the topic becomes more prominent as digitization pushes deeper into every aspect of business operations: “Customers raise concerns: Who can see what I'm doing? Can you change anything on the engine from outside?”
Everllence’s answer is multilayer protection: encryption of data in transit and at rest, strong access controls, and continuous security monitoring — including vulnerability management and intrusion detection — all backed by ISO 27001 certification.
From data patterns to early warnings
Two other AI capabilities which Everllence has already implemented into their digital service tools, are anomaly detection and computer vision. "Some features are already part of our PrimeServ Assist offering, but of course we always work to extend our capabilities", says Eberwein.
The team is developing virtual sensors based on machine learning algorithms. "An AI model can learn what certain indicative quantities look like if the engine works in good conditions, and then you predict that quantitiy based on previously seen data,” says Aman Steinberg, who is responsible for anomaly detection as well as computer vision at Everllence.
“If the currently measured value of the indicative quantity in the running engine deviates from the expected number, you know that this is anomalous behavior and it indicates that your engine is probably either deteriorating or at least operated in an anomalous way,” he says.
“An AI model can learn what certain indicative quantities look like if the engine works in good conditions, and then you predict that quantity based on previously seen data.”
Dr. Aman Steinberg, Senior Data Scientist at Everllence
Keyword: Predictive Maintenance
Steinberg and Group Digital’s various teams are working to broadly apply anomaly detection to individual parts such as injectors. “The buzzword is predictive maintenance”, explains Bernd Eberwein. “AI will help you understand in advance that a certain part might break.” That will mean a huge step forward: Utilizing the insights of predictive maintenance, the operator of the engine can order spare parts in advance. This is especially important if you consider vessels that are often in the open sea for weeks.
In the domain of computer vision, Everllence has successfully trained models for several components which are now deployed for an internal pilot testing phase: main bearing shells, pre-chamber gas valves and scavenge port inspection images. Steinberg explains the complexity of training a model for a specialized use: “The task is always to train the model by examples. So you provide examples of images where you know that the bearing shell exhibits no wear, or it shows certain levels of damage, and you label these images accordingly.”
Computer vision models are trained using labeled images to detect wear patterns on critical engine components. © Stefan Hobmaier
But you need a lot of images to enable a computer vision model to train on the numerical representations of your labeled images. And when the user uploads a new image, the model compares what it has already seen to what it currently sees. “The goal is that the AI is able to determine the wear class of the component so crew or service staff retrieve immediate feedback about the condition of their equipment,” says Steinberg.
From hype to tool
Eberwein is convinced that AI will add value to engine operation and customers. According to the “Gartner hype cycle”, there is usually a technological trigger at the beginning of every new solution that raises expectations sky-high. Afterwards comes the trough of disillusionment, when people are disappointed because things turned out to be more complicated.
“What follows is the plateau of productivity when people understand: AI is not a quick game changer, but a tool that can significantly add productivity in certain areas,” says Eberwein. “And if you keep that in mind, you can avoid falling in the trough of disillusionment, because expectations are realistic from the start.” Eberwein’s message: ”We do not blindly follow the AI hype, but rather pursue a targeted approach, focusing on where productivity can be increased and where it can add value to our customers.”
About the author
Berlin-based journalist Moritz Gathmann has been reporting as a correspondent for German publications from various regions of Europe since 2004. His work has been published in Der Spiegel magazine, Frankfurter Allgemeine Zeitung, and other media. Moritz Gathmann is a member of Primafila Correspondents, specializing in energy and healthcare reporting.