How to ... speed up the development of high-performance materials
Explains how artificial intelligence can help bring new, high-performance materials to market for additive manufacturing applications, and other areas of design engineering.
Artificial intelligence (AI) and machine learning are now a part of everyday life. Whatever you are doing they are there, from the Google algorithm that predicts what you are looking for as you type, to social media feeds, and the songs and programmes platforms like Spotify and Netflix think you may enjoy. Machine learning is also driving significant advances in materials science, helping to design high-performance materials more quickly and cheaper than ever before.
In the last 10 years, improvements in computing power, combined with the growing availability of information, have enabled the development of deep learning algorithms that can analyse vast data sets. Capable of finding correlations between materials with different compositions and characteristics, these clever algorithms can generate new insights into material relationships that humans would otherwise find impossible to detect.
Interpreting the underlying parallels that exist between data from different sources, machine learning tools can identify the target properties needed for specific tasks and propose ideas for new materials – accurately predicting how their physical properties will respond to specific treatment processes or manufacturing techniques.
Until relatively recently, the adoption of AI in materials design was slow. Scepticism about using a ‘black-box’ approach was one factor, alongside concerns about poor quality data, inconsistent formatting and extrapolating information beyond what is known. That has started to change. Forward-thinking companies have started to realise that AI can help them extract more information and value from their data, and deliver huge time and cost savings in the material design process.
Companies in the field of additive manufacturing are leading the way in this regard, propelled by the need to create new materials for this vibrant, fast-paced industry. Recently, AI was used to successfully design a new material for a high-performance direct laser deposition application. Part of a research collaboration between several commercial partners and the Stone Group at the University of Cambridge, UK, this project served as an exemplar for future rapid innovation in materials discovery.
Direct laser deposition
Direct laser deposition, where a laser is used to locally melt powder into the final object, is used in a wide range of industries to produce and repair bespoke and high-value parts including aerospace engine components, turbine blades, and oil drilling tools. Quicker than conventional production techniques, direct laser deposition has the potential to save component manufacturers vast amounts of time and money. However, for this manufacturing method to reach its full potential, a new generation of materials is required that can accommodate the high temperatures and stress gradients generated during the production process.
For the project in question, the partners’ aim was to create a nickel-based alloy that could be used to create a combustor for a gas turbine engine. With the combustor exposed to operating temperatures in excess of 1,000°C, the alloy needed to possess good oxidation resistance, excellent thermal resistance, and have good yield-stress and fatigue life. Keen to avoid expensive, speculative experiments and with 13 physical properties to optimise simultaneously, the research team decided to use AI to speed up the material selection process.
The key challenge was to find an alloy that would minimise the risk of defects, pores and cracks occurring during the additive manufacturing process. At the time of the project, direct laser deposition had been tried on just 10 alloys, meaning there were just 10 data points to assess. The team knew this would not provide enough information to characterise a 3D-printed alloy comprised of 20 possible elements, which could be affected by at least 10 variables during additive manufacturing and heat treatment.
To solve this, a large database of welding information was fed into the AI platform to help guide its understanding of the defects that could be formed during direct laser deposition on new potential compositions. Welding was chosen because of its similarities to direct laser deposition, where direct laser deposition melts locally with a laser beam, welding melts locally with electricity.
Utilising the extra property-to-property information that the welding database provided, the AI platform was able to unearth new insights about material properties, suggest a new alloy and propose accompanying processing conditions that would be most suited to the direct laser deposition application in question.
Satisfied with the suggestion put forward by the AI platform, the research team tested the proposed alloy by conducting a series of experiments to confirm its physical properties.
In the process of identifying the alloy, the same approach to machine learning was used to exploit other property-to-property correlations, relating heat capacity and thermal expansivity to ease of direct laser deposition, and hardness to ultimate tensile strength. The platform also used thermodynamic properties from computer coupling of phase diagrams and thermochemistry calculations to suggest changes likely to occur in the new alloy.
In total, 10 additional physical properties were measured, tested and verified to exceed required targets.
Machine learning in materials science
In total, the team estimates that it saved around 15 years of research and in the region of US$10mln in R&D expenditure, proving the potential for AI to drive innovation and sit at the forefront of materials design.
There are millions of materials available worldwide that are characterised by hundreds of different properties. Using traditional techniques to explore the information known about these materials, to come up with new substances, substrates and systems, would be a painstaking process that would take years.
It is possible to achieve the same results quickly and efficiently, with modern machine learning techniques. AI is uniquely positioned to extract deep insights from complex materials design databases, and to leverage this understanding to design new materials that satisfy multiple requirements simultaneously.
The potential for this technology in the field of direct laser deposition and across the wider materials sector is huge – particularly in fields such as 3D printing, where there is a need to design bespoke materials that can sit at the heart of the rapid printing process. As takeup of machine learning increases across the industry, the number of AI success stories about materials fulfilling previously unattainable requirements promises to grow and grow.
Gareth Conduit is Intellegens Chief Technology Officer and Co-Founder