Drives & Controls Magazine November/December 2022

30 n CONTROLLERS AND SOFTWARE November/December 2022 www.drivesncontrols.com AI: hope, hype and fomo T en years ago when we received enquiries about AI-based projects, they were mainly driven by hype or “fomo”– fear of missing out. When asked why they wanted to get into AI, the answer, more often than not, was“because my boss askedme to”,“because we have a lot of data”or“because it seems like an interesting area”. None of these were particularly solid reasons and certainly didn’t justify the hefty investment that AI applications would have required at the time. That is changing, and hope – rather than hype or fomo – is now the key driver. Users don’t ask whether we can help themwith AI anymore. They ask us whether we can help with predictive maintenance, quality control or process optimisation. They come to us with a problem that they want us to help solve. The adoption of AI is no longer a motive or an aim in itself. Instead, AI has become an enabler. Whatever the industry, the goal is the same – producing high-quality, defect-free products at a lower cost, using less energy and less labour. And AI can be one of a suite of techniques for achieving those objectives. Just as attitudes have evolved, so has technology. Although AI has existed as a concept since 1957, early applications were unfeasibly expensive and slow. It took a month to obtain the results of a simple calculation due to the limited processing power available at the time. Thanks to advances inmobile technology, computer storage and processing speeds, calculations can today be carried out in milliseconds and costs have plummeted. Although tech giants such as Amazon and Google have been using AI for some time, the technology is still in its infancy in industrial contexts. I would liken its lifecycle stage to that of robotics 15 years ago, when you needed a maths degree to control a six-axis robot. To implement AI-based systems, you still need experts; you need to understand what you are doing and it only makes sense in niche applications where the cost of entry can be justified by the benefit. It is also important to remember that AI is not a panacea. As machine-builders, data scientists and engineers, we can be guilty of defaulting automatically to technology for answers, when the more straightforward answer is far simpler and less sophisticated. Take a broken conveyor, for example. This is an engineering problem that can be identified and resolved using a traditional mechanical approach. It is the less obvious, intermittent issues – for instance, manifesting inmicro- stoppages – where AI can add value. AI problem-solving in practice Take this real-life example. We were called in to help an automotive customer that was having problems withmicro-stoppages. After performing a data scan, we carried out a“sanity check”. This involved connecting probes to the machine to create pictures of the signals that were being generated to establish what was actually happening. We then developed an experiment to pinpoint the root causes. This enabled us to identify and resolve about ten issues. The one that sticks inmy mind involved a sensor malfunction: one of the sensors we were monitoring seemed not to be working. When we asked the customer to check, they discovered a broken connector. We also identified some programming issues, including a logic mistake that was replicated in many machines on site, which they were then able to fix. They saved tens of thousands of euros in scrapped products as well as halving downtime, which translated to an extra four hours of production time every month. In another application, we are working with a food industry customer to improve seal integrity. By applying an AI approach to the sealing operation, we expect to increase shelf lives by several days and to minimise the possibility of faulty seals, thus eliminating the risk of retailers rejecting complete batches. Most of these projects have used Omron's AI Controller – the world’s first AI system that operates“at the edge”. The controller recognises patterns based on process data collected on the production line. It is integrated into our factory control platform, whichmeans it can be used directly in a machine, cutting losses. With examples like these and with AI being such a hot topic, it would be easy to assume that every manufacturer is on board with AI. But this isn’t the case. Examples of AI being used in factories are few and far between and projects rely heavily on the expertise of technology providers. In ten years’ time, it will be a different story. New tools will make AI far more accessible and user-friendly, enablingmanufacturers to take ownership of AI and run with it. n Artificial intelligence (AI) is everywhere these days. Or at least it seems to be judging by its media coverage. Tim Foreman, manager of Omron’s r&d operation in the Netherlands, examines the reality of applying AI in industrial applications. Omron’s AI automation controller integrates AI functions with machine control, leveraging information at the machine level in real time. It can detect momentary irregularities quickly and accurately, and feed the data back to the controller. As well as allowing machine-level trend monitoring, this can also reduce quality defects on high- speed production lines.