Materials joining batteries and AI
Lien Ngo explores materials’ place in the intersection between batteries and AI, and delves into the Internet of Materials.
In July 2018, UKRI ran workshop on AI and the data economy for batteries. There were speakers from both the Faraday Battery and the AI and Data Economy ISCF Challenges, from the Hartree Centre and IBM Research, from Williams Advanced Engineering and the Centre for Environmental Data Analysis.
Why was an AI for Batteries event organised? Because materials will play a large and valuable part in the intersection of AI and batteries.
AI here is shorthand for the array of digital tools – artificial intelligence, cognitive computing, machine learning and big data management, among many others. The next truly enabling technology, the computers that we use now for calculating and storing will be parts of networks used for interpreting information and generating insight. In silico will move to in aethere. In five to 10 years, even more research and development will start at a computer – it won’t be just to look up journal articles.
The community could be a winner here by recommending materials as a testbed and pushing for an Internet of Materials. This is a distributed search over the scientific data, including literature, patents, private data and experimental data from, e.g., the national facilities.
It has challenges that AI is poised to address. Materials science is an old discipline and so the data set has been around for centuries, and it is an interdisciplinary science with exploitation routes in every market sector and so translation between knowledge owners is important, as is interpretation between knowledge generators and users. It also has a vast and bewildering body of data in many forms, both quantitative and qualitative – spectra, images, structures, reactions, mechanisms, symmetries, longrange and shortrange order. More importantly, materials exploitation is capital intensive, high risk and takes a long time from initial development to market. Any tool to accelerate this process carries a strong economic case.
UK making ground
The UK is well placed to move quickly on this as it has world-leading expertise in both AI and materials science, and, as the AI for Batteries event showed, there is also lots of interest in the subject. Battery materials may be an ideal place to start. Data from the Faraday Battery Challenge ISCF programme could be used as a valuable example dataset to play with. It can take two decades to design and bring a new material to market and this is reflected in the battery innovation pipeline.
Current breakthroughs are mostly based on papers published 20–30 years ago and are already factored into the forecasts for energy density and cost. To ensure we have new materials and battery concepts in 10–15 years’ time, we need to be working on it now and that’s part of what the Faraday Challenge aims to do. Battery cell and systems models currently in use all need a vast library of materials property data – the Internet of Materials is one way to deliver this.
At the workshop, the Internet of Materials concept was introduced, which has been worked on in collaboration with the Hartree Centre, IBM Research and the Sir Henry Royce Institute for Advanced Materials. UKRI is by no means the first to think of an AI-enabled platform to make sense of and generate insight from the vast amounts of materials data in the world, there are already companies specialising in data and AI who are playing in this space.
However, the UK is well-positioned for this and has the potential to be able to launch the concept into something tangible for the community to trial. The Royce Institute is a key centre of materials expertise, the STFC encompasses computing expertise as well as operation of the large research facilities which generate huge amounts of data, and there is a thriving community of companies on both the supply and demand side for materials that are used to working with the rest of the innovation landscape. For example, IBM Research have made inroads into one of the necessary digital tools, with its Cognitive Discovery PDF parsing tool, called the Corpus Conversion Service, which can ingest 100,000 PDF pages per day, from peer review science journals, and then apply machine learning to extract knowledge with 97% accuracy.
The Internet of Materials concept needs to address these requirements:
Be usable for technical experts who are not materials scientists or computer scientists. In other words, a user who is an expert in their field of, say, batteries, can query it without computer science training or a CS interpreter and get reasonable information, insight and value from at least the first level results without needing a translation from a materials expert
Be valuable and usable for industry. There are many materials databases around the world. They often cover a very small subset of materials and are primarily useful for academic researchers who are already experts in that subject
Be able to ingest data in all forms, such as text, images, spectra, tables and qualitative comparisons
Be able to securely ingest and process data from different sources. A user should be able to access this tool from their computer at work and it will answer queries using publicly available data as well as data stored by her company without ever needing that privately held data to be moved or copied from its secure home. And of course, no one else would have access to that data unless the owner wished to share
Make it easy and valuable to share data. Researchers should be encouraged to share all data, especially failed experiments, and data that were collected and not used for a journal article or other dissemination route and would therefore have been lost to the community. Digital tools should be developed that collect metadata without contributors and users needing to enter endless lists of information.
Putting technology to use
How would an Internet of Materials be used? For example, a new, innovative antireflective coating material is needed. To start, a search for good anti-reflective materials could be carried out, or, if a current coating is being replaced, a search for materials with a refractive index above a certain value. There are lots of options, and the software will sort through these – metals, oxides and those with good transparency in the optical range. A filter can then be applied to those for which there is information for the material as a coating rather than a bulk material, maybe below 200nm thickness, and coatings that allow for very good transparency in the visible range. This will still give lots of options, which, again, the software would sort in a way that makes sense, rather than supplying a huge list of simple keyword matches.
To go further, requiring a coating material that can be applied at low temperatures and is flexible at 200nm thickness and is durable, for example. At some point, the software can give classes of materials that match the requirements, along with some quality factors and judgments.
As an industrial user trying to find a material that matches its needs, a few strong contenders can now be identified and looked at more closely, and you’ve found those contenders in hours, rather than weeks of expensive literature search by someone who can interpret ellipsometry results.
For an academic researcher, this could be the point at which you can see where there are holes in our knowledge. The software can help do some intelligent extrapolation and give ideas for rich areas of research.
Going back to batteries, the Internet of Materials could be used to discover cathode materials that have comparable or better capacity than current materials, but reduce reliance on cobalt, for example. In addition, a new supercomputer, Michael, has recently been installed at University College London that will be used exclusively by over 100 researchers working on Faraday Battery Challenge projects focused on creating new models for battery systems and researching next-generation, solid-state batteries. Named after Michael Faraday, the £1.6m supercomputer will work on research challenges such as simulation to offer insight into how existing battery materials work. Improved computer simulations of battery performance will increase the rate at which improvements are made to commercial models, accelerating the rate of mass-market adoption. The insight generated using Michael could ultimately be of valuable input to an Internet of Materials, which in turn could provide additional targeted options for battery scientists.
There is opportunity here to enhance the offering to businesses. For example, one model might be for a consortia of companies that agree to share data with each other, such that searches and queries from one company in the consortium could encompass the others’ datasets. Or there could be a fair use programme – if you wish to access certain datasets, you must make yours available. Businesses might be built that work on visualisations of the query results.
One eventual goal is that the Internet of Materials is seen to generate enough value that it encourages digitalisation of previous work, therefore making that data available again. The tools for the platform should be able to ingest data in many forms, but even the best tools won’t be able to read the data on a shelf of lab books or in boxes of microstructure images. If even a fraction of that dark data were made available, it would be a huge boost to the body of knowledge.
The Internet of Materials is an ambitious project. It requires digital tools that are not developed yet. In fact, there are many aspects that still sit squarely in the research space. We will need to support academic work on some of the tools. It’s also an inherently interdisciplinary challenge – it cannot be delivered without materials science, computer science, physical sciences and engineering expertise, but also confidentiality, security and ethic considerations. It requires widespread engagement and trust – public sector labeling reduces the fear of technology lock-in. It is too large a challenge for any single institution to undertake and therefore requires open transparent collaboration from businesses big and small, and this in turn requires that we show we understand industry’s security and confidentiality concerns and the onus is on us to prove that the value is worth it.
Lien Ngo is Innovation Lead for Advanced Materials at UKRI- Innovate UK.