As the chief engineer and head of the department for digital transformation of manufacturing technologies at the Laboratory for Machine Tools and Production Engineering (WZL) within RWTH Aachen University, I’ve seen a lot of technological advancements in the manufacturing industry over my tenure. I hope to help other manufacturers struggling with the complexities of AI in manufacturing by summarizing my findings and sharing some key themes.
The WZL has been synonymous with pioneering research and successful innovations in the field of production technology for more than a hundred years, and we publish over a hundred scientific and technical papers on our research activities every year. The WZL is focused on a holistic approach to production engineering, covering the specifics of manufacturing technologies, machine tools, production metrology and production management, helping manufacturers test and refine advanced technology solutions before putting them into production at the manufacturing edge. In my team, we have a mix of computer scientists, like me, working together with mathematicians and mechanical engineers to help manufacturers use advanced technologies to gain new insights from machine, product, and manufacturing data.
Closing the edge AI insight gap starts and ends with people
Manufacturers of all sizes are looking to develop AI models they can use at the edge to translate their data into something that’s helpful to engineers and adds value to the business. Most of our AI efforts are focused on creating a more transparent shop floor, with automated, AI-driven insights that can:
Enable faster and more accurate quality assessmentReduce the time it takes to find and address process problemsDeliver predictive maintenance capabilities that reduce downtime
However, AI at the manufacturing edge introduces some unique challenges. IT teams are used to deploying solutions that work for a lot of different general use cases, while operational technology (OT) teams usually need a specific solution for a unique problem. For example, the same architecture and technologies can enable AI at the manufacturing edge for various use cases, but more often than not, the way to extract data from edge OT devices and systems that move their data into the IT systems is unique for each case.
Unfortunately, when we start a project, there usually isn’t an existing interface for getting data out of OT devices and into the IT system that is going to process it. And each OT device manufacturer has its own systems and protocols. In order to take a general IT solution and transform into something that can answer specific OT needs, IT and OT teams must work together at the device level to extract meaningful data for the AI model. This will require IT to start speaking the language of OT, developing a deep understanding of the challenges OT faces daily, so the two teams can work together. In particular, this requires a clear communication of divided responsibilities between both domains and a commitment to common goals.
Simplifying data insights at the manufacturing edge
Once IT and OT can work together to successfully get data from OT systems to the IT systems that run the AI models, that’s just the beginning. A challenge I see a lot in the industry is when an organization still uses multiple use-case-specific architectures and pipelines to build their AI foundation. The IT systems themselves often need to be upgraded, because legacy systems can’t handle the transmission needs of these very large data sets.
Many of the companies we work with throughout our various research communities, industry consortia or conferences, such as WBA, ICNAP or AWK2023 — especially the small to medium manufacturers — ask us specifically for technologies that don’t require highly specialized data scientists to operate. That’s because manufacturers can have a hard time justifying the ROI if a project requires adding one or more data scientists to the payroll.
To answer these needs, we develop solutions that manufacturers can use to get results at the edge as simply as possible. As a mechanical engineering institute, we’d rather not spend a lot of time doing research about infrastructure and managing IT systems, so we often seek out partners like Dell Technologies, who have the solutions and expertise to help reduce some of the barriers to entry for AI at the edge.
For example, when we did a project that involved high- frequency sensors, there was no product available at the time that could deal with our volume and type of data. We were working with a variety of open-source technologies to get what we needed, but securing, scaling, and troubleshooting each component led to a lot of management overhead.
We presented our use case to Dell Technologies, and they suggested their Streaming Data Platform. This platform reminds me of the way the smartphone revolutionized usability in 2007. When the smartphone came out, it had a very simple and intuitive user interface so anyone could just turn it on and use it without having to read a manual.
The Streaming Data Platform is like that. It reduces friction to make it easier for people who are not computer scientists to capture the data flow from an edge device without having technical expertise in these systems. The platform also makes it easy to visualize the data at a glance, so engineers can quickly achieve insights.
When we applied it to our use case, we found that it deals with these data streams very naturally and efficiently, and it reduced the amount of time required to manage the solution. Now, developers can focus on developing the code, not dealing with infrastructure complexities. By reducing the management overhead, we can use the time saved to work with data and get better insights.
The future of AI at the manufacturing edge
With all of this said, one of the biggest challenges I see overall with AI for edge manufacturing is the recognition that AI insights are an augmentation to people and knowledge — not a replacement. And that it is much more important for people to work together in managing and analyzing that data to ensure that the end goal of getting business insights to serve a particular problem are being met.
When manufacturers use many different solutions pieced together to find insights, it might work, but it’s unnecessarily difficult. There are technologies out there today that can remedy these challenges, it’s just a matter of finding them and checking them out. We’ve found that the Dell Streaming Data Platform can capture data from edge devices, analyze the data using AI models in near real time, and feed insights back to the business to add value that benefits both IT and OT teams.
Learn more
If you are interested in current challenges, trends and solutions to empower sustainable production, find out more at the AWK2023 where more than a thousand participants from production companies all around the world come together to discuss solutions for green production.
Find out more about AI at the manufacturing edge solutions from Dell Technologies and Intel.
***
Intel® Technologies Move Analytics Forward
Data analytics is the key to unlocking the most value you can extract from data across your organization. To create a productive, cost-effective analytics strategy that gets results, you need high performance hardware that’s optimized to work with the software you use.
Modern data analytics spans a range of technologies, from dedicated analytics platforms and databases to deep learning and artificial intelligence (AI). Just starting out with analytics? Ready to evolve your analytics strategy or improve your data quality? There’s always room to grow, and Intel is ready to help. With a deep ecosystem of analytics technologies and partners, Intel accelerates the efforts of data scientists, analysts, and developers in every industry. Find out more about Intel advanced analytics.
IT Leadership