Exploring AI at the Edge in the Digital Industry
How AI at the Edge is changing the implementation of systems within the Digital Industry – particularly from the perspective of Smart Embedded Systems.
On the 9th of October 2020, KTN in partnership with EPoSS hosted the first webinar in the AI at the Edge series, looking at how AI at the Edge is changing the implementation of systems within different sectors – particularly from the perspective of Smart Embedded Systems. Watch the recording here.
KTN has recently refreshed its mission as “KTN exists to connect innovators with new partners and new opportunities beyond their existing thinking – accelerating ambitious dreams into real-world solutions.” Nilam Banks, KTM for Enabling Technologies highlighted the many areas where KTN has deep expertise and highlighted three particularly relevant areas: AI & Robotics, Manufacturing Made Smarter and Precision Medicines. Powerful connections are made through introductions, organising webinars and aim to help companies form collaborations for R&D projects.
EPoSS is an industry-driven policy initiative focussed on collecting R&D and innovation needs from industry, to define policy and funding requirements relating to Smart Systems and Integrated Micro- and Nano-systems. Elisabeth Steimetz, Office Director EPoSS, gave details of the members of EPoSS, which cover key industry organisations of all sizes and RTOs, with some of the examples of their technologies. One of the main EPoSS activities is to create a common European approach on Innovative Smart Systems – from Research to Industry – formulating agreed Roadmaps for European action on R&D (the EPoSS ID and ECS-SRIA) aiming to mobilise public and private resources, infrastructure and financial resources. EPoSS has some key activities in the progress towards Horizon Europe, integrating the EUREKA community into EPoSS and defining the new ECSEL KDT initiative. Membership of EPoSS allows organisations to collaborate with key players in the sector and to contribute to influential roadmaps and to take part in workshops, conferences and a range of joint activities.
Mark Lippett, CEO, XMOS, opened the main conference with a presentation “The Edge of Tomorrow”. In the context of AIoT (the combination of Artificial Intelligence with the Internet of Things) with a predicted 65 billion connected devices by 2025 delivering 180 Zettabytes of data and a $3trillion spend, XMOS had commissioned research on the impact of these changes. Based on the responses of 200 engineers, they found that 40% indicated that AIoT had the potential to radically change technology, and 44% said that AIoT was critical to improving the way we interact with products. Their analysis had Manufacturing and Smart home as the two largest opportunities. Mark gave examples of use cases for a selection of market sectors, and then went on to describe the barriers that must be overcome – leading to the need for a new kind of processor architecture, such as the XCORE AI. The report is available for download here.
Nicolas Lehment of NXP presented “Why AI is heading for the Edge”, starting with the three main applications for AI at the Edge within Industrial Applications of Vision, Voice/Sound and Anomaly Detection. Following a brief review of the innovations in AI and Computing, Nicolas covered the main topic of his presentation: Where should AI be implemented? – in the Data Centre, at the Network Edge or on the Device. He examined the need for a variety of platforms to address different challenges, but gave the promise of AI on the Edge as Fast & Dependable intelligence, Independence for Bandwidth Constraints and Data Privacy. Nicolas’s final slide gave an overview of Machine Learning Processing Requirements showing required performance for different applications.
Adrian Moran of Ikerlan presented “The Needs of AI at the Edge in Industrial Environments”. After a brief introduction to digital platforms, Adrian covered Classical Cloud-Edge Architectures looking at the differences between the training and inference operations being undertaken in the Cloud, as opposed to the required Cloud-based Training and implementing the Inference operation at the Edge. When the training and inference are split between the cloud and the edge, there is the need to examine Distributed Training or Federated Learning. Adrian gave a selection of Use Cases and ended with conclusions on Cloud AI as opposed to Edge AI.
Gash Bhullar of the Welsh Digital Manufacturing Innovation Hub (DMIH) presented “AI at the Edge of Manufacturing”. Gash started by showing a timeline of the various programmes that they have been involved within the run-up to Digital Manufacturing. He then presented some of the implications of AI to manufacturing systems. Gash outlined the main application areas: Improved Machine Operating Performance; Improved User Interface and Machine Monitoring with the Advantages of No Data Link, More Secure and More Responsive. Before providing a summary of his presentation Gash gave a live demonstration of machine vision in practice. The example he gave was the application of detecting face masks – being able to not only check that someone is wearing a face mask but also to check that the mask is appropriate for the task being performed – e.g. paint spraying.
Steve Cammish of Adlink presented “Edge AI, Vision & Logistics”. After a brief overview of Adlink, Steve gave two cases studies of using Vision in industry. The first was a high-volume warehouse operation requiring improved accuracy & efficiency of loading the correct packages onto pallets. The system monitored the pallets, reading barcodes and highlighting errors (e.g. wrong package or bar code not visible) leading to improved productivity and reduced theft/damage to packages. The second case study was monitoring a production line to check that the correct contents were included. The improved accuracy from the system was material in winning new contracts. Steve finished by discussing the roles for a successful operational roll-out, some of the new business models and how Vizi-AI devkit can provide a starting point for new implementations.
The AI at the Edge webinar series is aimed at both technology users, who want to understand the capabilities of technologies, as well as product development engineers, designers and product managers who are looking to improve the capabilities of the systems that they design.
9th October | Digital Industry (Webinar recording and slides are now available)