|
|
Through the stone, bronze, and iron ages the discovery of new materials has chronicled human history. The coming of each age was sparked by the chance discovery of a new metal. Despite the central importance of materials in enabling breakthrough technologies, even today the only way to develop new materials is through experimental driven trial and improvement. We have now implemented a computational tool that can rapidly, reliably, and robustly design a material with specified physical properties, and moreover have proven its accuracy by predicting four new alloys that were subsequently experimentally verified.
With the proven efficiency, veracity, and reliability at designing new high impact materials that have been experimentally verified, and with patent protection of the algorithm submitted, we are well positioned to dramatically broaden the range of materials that the tool can be applied to. With the turnover for UK industries relying on advanced materials reaching £170bn in 2012, there is a healthy market for advanced materials design. We are now working with industrial partners at Rolls-Royce, Samsung Electronics, and BP to further the materials design paradigm. The tool can help not only materials scientists, but moreover design engineers. At present engineers must design objects and products around the shortcomings of pre-existing but non-ideal materials. The tool could allow engineers to instantly optimize bespoke materials for their application, bringing materials into the heart of the design process
We have developed a tool [Patent applications EP14153898, US 2014/177578, & GB1302743] that can automatically computationally design a material with specified physical properties. The tool relies on databases of material properties with composition and heat treatment. These are constructed either from physically based models, first principles computational predictions, or experimental databases. The tool takes these databases and constructs artificial neural network models that predict each physical property. Each prediction includes a local uncertainty due to the sparsity, inconsistency, or uncertainty in the data. We are then well positioned to fully characterize each proposed material, and by taking into account the uncertainty in the underlying models calculate the likelihood that the proposed composition exceeds the specification. The separate probabilities for each property are then merged into a single probability that the material exceeds the entire specification. In the second stage, the program optimizes the material composition to maximize the probability of exceeding the targets. The optimization of probability is crucial: with multidimensional targets the probability of each property fulfilling its target are multiplied together, meaning that the overall probability can be much less than the individual probability. This approach allowed us to discover two new nickel and to molybdenum-based alloys whose properties have since been experimentally verified [Materials & Design].
In the future we aim to imbue the materials design approach with intelligence: allow it to understand and exploit the trends that material scientists do today. We have therefore developed the capability for the neural network to learn and exploits correlations between properties to fill in gaps in databases, for example hardness is proportional to yield stress. The tool interrogates the neural network to predict material properties along with uncertainty, and then searches for the composition and processing variables most likely to exceed a target specification, to provide the best possible guidance to experimentalists. One powerful possibility to exploit this is to combine ab initio calculations with experimental data. After learning the correlations between numerical simulations and experimental data the algorithm will allow us to take advantage of the ability of ab initio calculations to be performed at any composition to guide the extrapolation of the more accurate experimental results to any composition.
We initially focused on developing alloys that could have a significant impact on the gas turbine industry. Gas turbines drive ships, power stations, and aircraft all over the globe. Due to their widespread usage, gas turbines contribute up to 20% of the total UK CO2 emissions, so new technologies that improve efficiency will have a significant impact on preserving oil reserves, cutting greenhouse gas emissions, and saving money. We discovered two new nickel-based alloys for gas turbine engines, and two molybdenum-based forging alloys. The physical properties of the alloys have been experimentally demonstrated to exceed commercially available alternatives so all five alloys are now being further developed by Rolls-Royce plc. The nickel-based alloys fulfill the Advisory Council for Aeronautics Research in Europe 2020 targets to reduce the net CO2 emission from gas turbine engines by 1.5 billion tons/year worldwide, and save the airlines fuel costs.
We have exploited the materials design tool to predict four new alloys [Patent applications EP14161255, US 2014/223465, GB1307533, EP14161529, US 2014/224885, GB1307535, EP14157622, amendment to US 2013/0052077 A2, GB1403486, GB1408536, Acta Materialia, 61, 3378 (2013), and Materials & Design] that were subsequently experimentally verified. This tool can help not only materials scientists, but moreover design engineers. At present engineers must design new objects and products around the shortcomings of pre-existing but non-ideal materials. The capability to develop materials computationally would allow engineers to instantly optimize bespoke materials for their application, bringing materials into the heart of the design process.
The alloys developed include a new nickel-based alloy for turbine discs in jet engines, a novel alloy for lining the combustion chamber of a jet engine, and a molybdenum based forging alloy. Each alloy has eight individual physical properties that are predicted to match or exceed commercially available alternatives including fracture toughness, oxidation resistance, yield strength, creep resistance and processibility. These properties are calculated using a variety of techniques including ab initio calculations, physically based models, and interpolation of existing experimental data. Several properties for each alloy have subsequently been experimentally verified.
