Alloys are mixtures of elements that together create a material that is stronger, more durable, and sometimes lighter than the elements individually. Traditional alloys like steel or bronze have been around for centuries and usually combine two or three metals to form an alloy. However, the metals we need for advanced technology, like jet engines or smartphones, are becoming more complex. Enter Multi-principal Element Alloys or MPEAs, new mixtures that contain five or more metals in nearly equal amounts.
The challenge with making these advanced alloys is that the best recipes often contain rare and expensive elements known as "critical raw materials" (CRMs), like tantalum, tungsten, niobium, and hafnium. These materials are crucial for high-tech equipment but are difficult to source sustainably. They can cause environmental damage when mined and create risks for global supply chains when countries can't easily access them. To reduce our reliance on these CRMs, scientists from the Indian Institute of Technology (IIT) Guwahati, the University of Leeds, UK and The University of Manchester, UK, are working to develop MPEAs but without using any critical materials. Using machine learning algorithms, they have created a framework for creating new sustainable alloys.
The researchers generated a dataset of 3,608 new materials using Thermo-Calc 2024, software for predicting materials and studying their properties, and TCHEA7, a database of complex alloys. Both CRM (e.g., Hafnium, Tantalum, Niobium, Tungsten) and non-CRM elements (e.g., Aluminium, Chromium, Copper, Iron, Nickel, Titanium, Molybdenum, Manganese, Tin, Zinc, Zirconium) were considered.
Machine learning algorithms were then used to predict the hardness of these new materials. The algorithms used included Decision Tree Regressor (DTR), Random Forest Regressor (RFR), AdaBoost Regressor (ABR), Gradient Boost Regressor (GBR), XGBoost Regressor (XGBR), and Extra Trees Regressor (ETR). Of these, the Extra Trees Regressor (ETR) algorithm was found to perform the best at predicting the hardness of these materials.
Next, using the best-performing ETR model, scientists applied metaheuristic optimisation, an advanced problem-solving algorithm, to design compositions with high hardness without the use of CRMs. The algorithms explored included Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Ant Colony Optimization (ACO), Cuckoo Search Optimization (CSO), and Whale Optimization Algorithm (WOA). Among the models tested, Cuckoo Search Optimization consistently identified MPEA materials with hardness values near Thermo-Calc predictions, with an acceptable margin of error. Using this method, the researchers were able to generate novel materials without any CRMs, with hardness values matching or even higher than those with CRMs.
To validate machine learning predictions, an alloy Al6.25Cu18.75Fe25Co25Ni25 was synthesised. Its experimentally measured hardness showed close agreement with both Thermo-Calc and ML evaluations, confirming the accuracy of the prediction.
This research could significantly lower our dependence on CRMs, which would help protect the environment by reducing the need to mine these scarce materials. It could also make certain technologies less expensive and more accessible by easing the global supply chain pressures caused by their scarcity.
The framework does have limitations, however. While the study successfully identified new alloy compositions, it primarily focused on hardness, one of many properties needed for practical applications. Future research needs to investigate other properties, such as how these alloys react to temperature changes or resist corrosion. Moreover, the dataset used for machine learning predictions consisted solely of synthetic data. More real-world testing is necessary to verify that these computer-designed alloys perform just as well or better under various conditions. By conducting more experiments, scientists can fine-tune their predictions and expand their methods to discover alloys for a wider range of uses.
The clever use of machine learning and optimisation tools in designing new, eco-friendly alloys presents an exciting frontier in materials science. This research not only pushes the boundaries of what we can achieve technologically but also promises to do so more sustainably.
This research news was partly generated using artificial intelligence and edited by an editor at Research Matters