Contact: NINA DYBDAL RASMUSSEN nina.rasmussen@noricangroup.com TECHNICAL TOOLBOX ISSUE 21 operator so they can target and persuade non-compliant staff. In general, work with the whole team to identify obstacles for a positive change and build consensus for the chosen path ahead. Q: How do we train AI to gather more reliable data? A: Understand your challenges and then target specific actions. By fully understanding the problem AI is intended to solve, you can target data collection and other actions effectively. If scrap reduction is the goal, is the scrap level fluctuating or steady? Is it confined to certain patterns and lines, or does it affect all castings? Can we collect data on what is influencing quality problems or is it actually something outside the scope of AI? Analyzing data can’t fix an undocumented change in product design or report that staff are dropping, damaging, or losing castings. An AI is only as good as the data used to train it—and then to drive it operationally. It’s essential to put the processes in place required to gather reliable and consistent data. For example, noisy human data might be due to different production shifts recording different results for the quality of the same casting surface finish. Bias might appear by consistently capturing quality data in the wrong category. Algorithmic judgements (like analyzing digital camera data to estimate surface quality) tend to have much less noise and, if the model’s training data is unbiased, much less bias too. Q: How do you foresee the role of AI evolving in the coming years? A: By staying the course, you will start to realize real value in automation. Progress on changing systems and behavior takes time and can be frustrating. Persevere. By addressing challenges small and large, your foundry will progress on its digital journey, gain vital experience, and see the AI really start to control your process. With this foundation in place, true optimization becomes possible along with gradual, but continual improvement. As AI becomes business as usual, advanced users can start to move towards full automation. After successfully implementing and adapting the AI system, the next step for an advanced customer involves defaulting its PLCs to AI-suggested machine settings rather than substituting their own settings. That’s the first step towards letting the AI directly control the PLCs. Q: How is AI currently being utilized in small- to mid-size foundries today? A: So far, AI has been implemented by larger foundries but can be equally effective for small- and mid-size foundries. AI can help improve OEE in many ways by optimizing the control plan across all the different process steps, stabilizing the production line, and rapidly identifying — and correcting — any parameter deviations that need urgent attention at any time. Both scrap rate reductions upwards of 40% and uplifts in ranges from 66-86% are being seen in global foundries of all sizes, and improvement continues. Q: What about cybersecurity issues? A: Cloud-hosted IIoT and AI services for foundries are usually far more secure than traditional in-househosted systems. But like all IT systems, maximum cybersecurity requires everyone to play their part: the customer and its users, the service provider’s software developers and its operational team, along with hosting and other suppliers. You will want to understand the web hosting services to ensure global access along with where the data is being processed and stored. Understanding gateway connectivity and how the data it transmitted is also critical. Customers have complete centralized control of identity and access permissions for their users. In the cloud, multiple techniques like tenant isolation, one-way keys, data replication, and redundancy ensure user privacy and prevent data loss. Q: How are AI systems maintained? A: An AI is typically retrained regularly on the latest data to maximize its accuracy and effectiveness. Ideally, AI should be retrained quarterly, this will ensure AI optimization to be applied to new patterns as soon as sufficient historical data becomes available. This and any other administration, like implementing the latest software updates or security patches, is typically carried out “behind the scenes”. There should be no regular customer maintenance required, hence there’s no scheduled downtime. The advent of robots created better robotic cell operators. Likewise, we can say AI will create better foundry operators because they too will see the impact they will have on the bottom-line results.
RkJQdWJsaXNoZXIy NDI4Njg=