AI hardware on shop floor at AMRC’s Factory 2050

3 min read

IBM AI (artificial intelligence) hardware on the shop floor of Factory 2050, at the University of Sheffield’s Advanced Manufacturing Research Centre (AMRC), is giving UK manufacturers access to AI and machine learning technologies.

“This is the first industry-focused AI system of its kind in the UK,” says Sean Wilson, technical lead for AI at the AMRC. “It’s the result of a very close relationship with IBM that we’ve developed over recent years, enabling the AMRC to open up another dimension of Industry 4.0 for our partners, and the wider supply chain.”

The computational muscle of the IBM Power9 AC922 server and FS9150 flash storage enables AMRC research engineers to shred the time it takes to develop an algorithmic model from weeks down to a few hours.

Wilson says: “A perfect example is a recent five-day assist we did with a regional SME, in which we delivered a proof of concept to understand the structure of written documents/forms using computer vision. Without the power box, it would have taken a couple of weeks to generate, train and test the deep-learning model. With PowerAI Vision software, we were able to upload and train the model in just over an hour, with 86% accuracy. This provided the client with the proof of concept he needed to approach investors about commercialising his product. In this case, the power of AI enabled the AMRC to test ideas in hours and understand whether it’s worth pursuing them further. Failing fast, and learning fast from those failures, is now very feasible.”

Unleashing the power of AI is also having an impact on the AMRC’s bigger partners. One of the UK’s leading industrial exporters, which has been collaborating with the AMRC, IBM Watson, and a Sheffield-based digital company, Razor, are using AI to make a step-change in the performance and quality of their screening and inspection processes for critical aerospace components. This strategy not only ensures passenger safety, by providing a continuously improving inspection process built on big data analytics, but delivers inspection rates 4,000 times faster than the original inspection.

The AI solution, which brings local and global expertise together, will soon be on the factory floor of the OEM, which states: “we’ve learnt more in the past 10 weeks at the AMRC than we have in 10 years on our own.”

Kieran Edge, AMRC technical lead for machine vision, who oversees the project, adds: “We used PowerAI Vision software for its classification and detection capabilities, and found we had tremendous success straight off the bat.”

For IBM, the deployment of the AI demonstrator at the AMRC is a way of lowering the barriers to the adoption of Industry 4.0 technologies in high value-added manufacturing sectors such as aerospace.

“It’s a way of demonstrating the art and science of the possible in a way that de-risks future investment in a technology which is often poorly understood by even some of the bigger players in the market place,” says IBM’s Ian Gardner.

PowerAI Vision software provides the AMRC and its partners with an open-source, easy-to-use framework and tools for building and managing computer vision models, including functions for installation and configuration, data labelling, model training, inference, and deployment.

Located at Factory 2050, the Power Systems AC922 server offers a fully optimised platform to support the massive throughput that a project’s workload requires. The server pairs POWER9 CPUs and NVIDIA Tesla V100 with NVLink GPUs, which delivers up to 5.6 times the I/O performance of x86-based servers.

The Factory 2050 team has also implemented a FlashSystem 9100 solution to boost throughput speed and storage efficiency.

“FlashSystem 9100 will really increase our ability to do training and iterations on the models that we’re generating,” says Wilson.

The enterprise-class storage solution combines flash performance and NVMe (non-volatile memory express) with the reliability of IBM FlashCore technology. Designed for security, the solution features built-in encryption and FIPS 140-2 certification. It integrates well with IBM Spectrum Storage capabilities, enabling scalable, tiered storage solutions in the cloud – another plus for the AMRC and its customers, who want to modernise their infrastructures using new private and hybrid cloud models.

“Cloud solutions suit many of the manufacturers who come to us for R&D support with AI, eliminating the hardware investment required to adopt intelligent automation,” says Edge. “However, for others, a hybrid cloud approach might provide a better balance. The containerised architecture implemented at the AMRC allows us to offer both to our partners.”

Head of digital at the AMRC, Professor Rab Scott, says the IBM solution is not only fast and powerful, it is also safe and secure: “What we’re offering brings the power of cloud into the heart of the factory without dragging cloud misconceptions with it. This on-premise cloud keeps data safe, handled in the local IBM and AMRC infrastructure so that our partners can de-risk innovation while knowing that their data is secure.”

Wilson adds: “What we have here is a massively powerful tool. Knowing how to use that tool safely while pushing it to its limits, is what the AMRC brings to the collaborative processes. Using Tensorflow, which can run on run on multiple CPUs and GPUs, we’re able to build custom research-led neural network models with access to performance levels that would cost tens of thousands of pounds a day in cloud land.

“This means the AMRC and IBM can sweat the box without having to worry unduly about the price; achieving big innovations at cost levels that are affordable to SMEs,” he continues. “The future of AI manufacturing is here in Factory 2050: now.”