Generative AI in mechanical and plant engineering

“The topic of generative AI has always been present, but this year it will be particularly in the spotlight at the Mechanical Engineering Summit,” predicts Guido Reimann, deputy managing director of software and digitalization at the industry association VDMA. The background is the dynamic development following the triumph of ChatGPT, Google Gemini, Microsoft Copilot, and other AI tools with underlying large language models, which are now turning many traditional things upside down in many places. Reimann: “Companies are enabled by the new technical possibilities to do things differently today than in the past, to automate them, and thus to completely rebuild entire processes and workflows.”

Why is generative AI the number one summit topic?

The organizers want to use this focus in Berlin to show what is already possible in practice today, what is technologically coming to companies, and how technological change can be concretely addressed. There will be an interplay between technology companies, which will show the current state of the art, and mechanical engineering companies, which come from domain knowledge and take care of the development of new products. Above all, practical lectures are intended to wake up the audience and provide food for thought and new impulses for their own company. Reimann: “Our member companies always like to learn from concrete examples of how others do it, in order to derive what could be changed in their own company.”

Generative AI vs. narrow AI

Basically, everything is already contained in the term. In contrast to pure machine learning (narrow AI), which collects data for a defined area, analyzes it, and derives actions from it, generative AI is inherently designed to create new things based on the enormous amounts of data from various sources, whether texts, speech, images, or films. Reimann: "The development of large language models, in particular, makes it possible to interact with AI in normal language, making the whole thing interesting." Today, anyone can basically operate an AI chatbot - it no longer has to be the IT specialist who makes complex database queries.

How GenAI is changing service processes in mechanical engineering

On this general basis, the technology is increasingly being applied in practice in mechanical engineering and other industries. A first step is to 'just' try out AI, improve research results, and the like. Further applications based on this are coming into play in mechanical engineering, for example, in the area of service. Service technicians are usually under great time pressure: users want their machine to function again as quickly as possible. Service employees are therefore faced with the challenge of having to review vast amounts of documentation for a machine or system in a very short time.

They may not know this particular machine very well due to a lack of experiential knowledge; furthermore, much information is only accessible through CRM or ERP systems and the like. Quick help in this and similar cases can be provided by a generative AI solution that works in the background, can access all relevant information, and prepares it accordingly well and quickly. This way, service employees receive detailed instructions in no time on how the maintenance or repair process must be carried out; more complex problems can also be enriched with 3-D models or augmented reality.

Automated 8-D reports

The technology group Wilo SE, with a focus on the pump industry, began to engage with the topic of artificial intelligence very early on. Even before the hype around ChatGPT, the AI pioneer based in Dortmund set up its own AI competence center five years ago. "Today, we offer all our employees access to Wilo GPT - our internal chatbot solution with corresponding security and data protection standards," reports David Graurock, head of the competence center. The AI system has been rolled out company-wide for a year and a half.

No specific use cases are prescribed for use, but everyone can use it at their own discretion. From the creativity of the workforce, some very exciting projects have already emerged, such as in quality assurance for suppliers. With the help of generative AI, the so-called 8-dimension reports requested from suppliers are now completely automated and sent to the relevant supplier. Graurock: "The system evaluates what information is available in the company, understands it, and generates a report based on it. However, before it is sent, an expert looks over it again to ensure everything fits."

Wilo started with a secure instance of Open AI's GPT model. Graurock: "Since then, we have been gradually enriching the system, from application case to application case, with more and more of our own company knowledge and data. The platform is continuously evolving."

How to achieve safe AI use in an industrial environment?

Can you blindly trust artificial intelligence? A question of significance that surely many users ask themselves. "If you get a wrong recommendation from the AI, then humans can only react incorrectly," warns the deputy VDMA managing director of software and digitization against a too careless handling of the new technology. AI does not mean putting aside one's own intelligence. "In practice, there can also be deviations and misinterpretations. These can be so-called hallucinations of the AI, but also own errors in operation due to rather clumsy questions."

To minimize these dangers in the corporate context as much as possible in advance, companies, according to the experiences of the industry association, adapt their AI chatbots to their own conditions. This means they do not use the publicly accessible offerings from Microsoft, Open AI, Google & Co., but instead 'build' their own environment where they precisely determine where and from which data sources the language model should draw the solution. As a result, the possible outcomes are already very limited. Reimann: "The language model ultimately only serves to interact in natural language within these possibilities."

Company information as the most important basis for chatbots

"The universal knowledge of a chatbot is not the top priority in industrial practice," confirms the head of the Wilo AI competence center. "All company information must be entered manually - also to prevent the chatbot from making something up entirely." In the long run, it will result in language models functioning only as a human-machine interface and always referring to reliable data sources in the background.

Interim conclusion: People must think in advance about how the use of AI should function sensibly and bring the information to be used by AI into a good quality state. Then the results will also be significantly better, and AI can become increasingly user-friendly. Ultimately, there is still a lot of mathematics and statistics behind this technology.

Introduction of AI in the company

How complex such AI introduction projects will be in one's own company is difficult to generalize and, according to the experts surveyed, primarily depends on the planned scope of services. "If the appropriate data is well-prepared, it can start very quickly, within just a few days," says Reimann. It is important that as many necessary employees as possible are closely involved in the company. An AI introduction project requires both IT expertise and the employees who will ultimately operate and use the chatbots; they must also assess whether the results achieved with generative AI are usable.

Strategically anchoring AI in the company

Projects that are more forward-looking, however, which may be closely related to one's own product development, are a completely different challenge and accordingly require more time. Generally, you start somewhat smaller, gradually advance further, and continuously develop your AI. In this way, as the user example Wilo shows well, more and more substance is added over time. Reimann: "In the long term, there is no way around a strategic plan. It is important to strategically anchor artificial intelligence in the company, to focus on use cases that are also value-adding or relieve the people involved in value creation of work, and to regularly put everything to the test. Searching for information is not inherently value-adding."

The role of humans in the AI-driven future

It is to be expected that with the rise of artificial intelligence, work tasks, the scope, and the environment of professions will change or evolve. The products themselves will also become more extensive and complex. All areas will change, but so will the opportunities in mechanical engineering.

Generative AI is initially considered a new tool, much like the PC was when it entered the workplace several decades ago. Reimann: “It will be similar with AI. Humans will always try to push things forward.” The really exciting discussion, according to the deputy managing director of software and digitization, usually revolves around whether jobs will be lost due to the introduction of AI. Here, the association clearly gives the all-clear: “At the moment, we still have the problem that we lack people in various areas of mechanical engineering who can work there,” Reimann positions himself clearly and distinctly. Using AI as an assistant is rather a good tool to reduce the time spent on some perhaps somewhat tedious tasks and to focus on the more varied and, above all, value-adding things.

Heterogeneous environment in practice remains

A final scenario from practice: In companies, young people usually encounter a very heterogeneous environment after their training: state-of-the-art machines alternate with 'dinosaurs' that have been reliably performing their duties for ten or twenty years

 In training, the focus is on newer things, but the old is still present. In this context, young people must be able to handle both sides. Here, too, generative AI can be very helpful in continuing to use machines that are no longer brand new in terms of sustainable use and thus conserving resources.

Edited by Dietmar Poll