Mr. Klingstedt, at the digital networking days 2021, you presented a project from the semiconductor industry that you had implemented. Can you briefly outline what the task was?

Hans Klingstedt, Senior Project Manager at Smart Systems. © Smart Systems

This project involved the early failure detection of industrial valves in an ultra-pure water circuit. Globalfoundries was the challenge sponsor in this innovation project, for which we put together an individual team of experts and developers to implement an extension of the predictive maintenance measures using the latest technologies, tailored to this use case. Our goal is to accelerate the implementation of cutting-edge technologies in innovative products. In this project, we have selected the Ganymed multi-sensor platform as the hardware to equip this "smart sensor" with AI software that can predict a failure based on anomalies. This type of data pre-processing, analysis and classification on the edge device, in this case the Ganymed, is classified as edge computing. This hardware-related processing has enormous potential in areas such as data transmission, latency reduction and general integration issues. More and more use cases are dealing with latency issues and an alternative to the common approach of sending large volumes of data exclusively to the cloud and analyzing them there.

Where was the greater challenge: the hardware used or the programming of the AI?

Let me put it this way: The challenge was to embed the AI model on the hardware. This challenge was not yet solved with the step of quickly building an AI model using suitable libraries and making very good predictions. The model we customized had to be implemented on the smart sensor with its corresponding restrictions. To meet this challenge, we integrated the start-up Coderitter into the project team, which successfully tackled this important milestone.

Were the requirements for the solution at the semiconductor manufacturer higher than in other industries?

Compared to other industries, semiconductor manufacturers have a very high level of digitalization maturity. Nevertheless, there are critical processes and infrastructure components in production that can be made more resilient with flexible and easily integrated solutions. At the same time, however, care must be taken to ensure that - as in our case - such a predictive maintenance solution fits in well with Globalfoundries' overall facility management and data management concept. We were able to demonstrate this very well with our Edge AI solution in the scenario described.

Where do you see the greatest growth potential for predictive maintenance in the coming years? What are the fields of application?

The rapid technological development of powerful processors will enable the simplified implementation of AI models on an edge device in the future. In addition, more and more standard use cases are being developed and corresponding libraries are being made available to facilitate access for companies. Industrial IoT applications will now include more and more components, machines or general locations that can integrate smart services and enable high benefits through optimized maintenance cycles and early failure detection.

I see predictive maintenance primarily at the edge, but can there also be applications on a larger scale? How does the topic fit into the overall picture of the networked industrial plant?

We believe that the majority of data can be processed on the edge. However, cloud services are useful, if not essential, for device management and the optimization of AI models. Here, user companies will continue to decide on the basis of their strategy to what extent they select data-driven services from different suppliers or build up internal expertise for the integration and operation of predictive maintenance solutions. This will complement machine data that goes beyond standard PLC control and delivery specification compliance data.

Do you already have a new predictive maintenance project to talk about?

Predictive maintenance continues to be a central component of our innovation projects. In our current Digital Product Factory, our 3-month co-innovation format, we are building a robot pay-per-use solution in which we are evaluating robot data with SAP's Predictive Assets Insights module as part of the project. In addition to insights into the health of our robot colleagues, this also provides us with insights into what activities they perform and how exactly they perform them, which in turn provides valuable information about the product, the environment or interaction with people. Here too, the future will be about making robots accessible to more companies. Predictive maintenance solutions will help us to make this human helper even better and more reliable. We are already looking forward to the next Digital Product Factory in the second half of the year, where we will once again set standards in the field of predictive maintenance through co-innovation.