Factory 2050


AI theme, investigating natural language processing as well. It’s using IBM’s AI platform, and also assessing AI soſtware from Microsoſt, Cognex and MVTec. Edge said that aerospace firms, like many


companies, are interested in exploring the capabilities of AI, but that validating the technology would be a big task. ‘Validating machine vision alone is difficult,’ he said, ‘and you know what the algorithms are looking for, and how it will fail.’ An edge detection algorithm, for example, will fail to identify a defect if it doesn’t have a strong enough edge. However, with AI, Edge continued, the neural network has been trained on the data, but the user doesn’t quite know what features the network is looking at, and whether or not it will fail. Tis makes gaining confidence in AI a


difficult task, especially in aerospace, where the sensitivity of inspection is paramount – missing a defect on an aerospace component could be very costly. ‘It’s not just about getting the data but getting


the right data,’ Edge said. If a defect appears once in every 1,000 components, it’ll take years to reach a sufficient level of data to have confidence in the technology. ‘It’s a business case on whether you bite the bullet early and start doing this data collection,’ he continued. ‘It is an area where companies like Rolls-Royce are starting to look forward in their data


gathering and annotation, and having these stockpiles of data.’ Te AMRC is starting to investigate


how best the data from the Rolls-Royce automated inspection project can be used, including whether it can be used for AI. Te cell generates a large amount of data per component, according to Edge. ‘It’s a lot of data, especially when they’re making 100,000 of these components a year,’ he said. ‘It’s what you do with that data, how you label it, how you annotate it, and how you store it.’ AMRC is working to


@imveurope


www.imveurope.com


Digital data is going to become more prevalent moving into the next generation of aircraft


assess the risk of a lot of these technologies. It will put them on the shop floor and simulate a production environment, using real production parts. It also tries to shorten the period of adoption for some of this technology, and limit the time between a concept identified by a university and technology that can be put on the shop floor. ‘Machine vision can be considered a


risky technology,’ commented Edge. Tere are challenges involved in acquiring data, especially data about defects. Understanding how good the system is, relative to a human inspector, is also very difficult, because it’s hard to assess the capabilities of a human inspector


without bias. Te main risk, though, with using machine vision, is that if a machine misses a defect, then it’s reasonable to think the machine would always miss the defect and that it’s missed those defects in the past, because it’s the same algorithm. If a human misses a defect that’s human error, which is more of a one-off event, and the operator can be re-trained. But if a machine misses a defect, that can be a repeatable, mass-recall error. ‘Tat is a big hindrance for machine vision systems,’ Edge said. ‘Even if the system is better than a human, there’s trust in the human operator.’ Te AMRC has been using AI for object detection and scene


classification so far. Artificial intelligence gives better classification results than traditional machine vision techniques, according to Edge. Detecting scratches, for example, is where AI excels, but validating the technology will be difficult. Edge suggested two possible ways of validation: leaving an AI system running blind for a year or so, and comparing the results to reports from a human operator. Or, use two AI systems, one looking for specific defect types, and another running in the background looking for anomalies and things it’s never seen before, which would act as a failsafe. It’s then up to the customer whether it can


The AMRC’s demonstration lathe on which it has installed numerous sensors to show how well the machine is running 8 Imaging and Machine Vision Europe • Yearbook 2019/2020


University of Sheffield Advanced Manufacturing Research Centre


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52