The study authors focused on MPEA, special alloys composed of more than three elements (in roughly equal proportions) that exhibit remarkable mechanical properties. These materials are much stronger, tougher, and more resistant to heat and wear than conventional alloys.
They are already used in aircraft parts, surgical implants, and clean energy systems. However, the traditional method of creating MPEA involved testing numerous combinations of elements. This process is costly and time-consuming.
To overcome this challenge, Deshmukh and this team developed a smart, data-driven system that combines the power of machine learning and algorithms. They started by gathering a large dataset of existing MPEA.
Basically, information like which metals were used, how those metals were arranged at the atomic level, and how each alloy performed in terms of strength, flexibility, and other mechanical properties was fed to the system.
Two machine learning models were trained using the information. The first was a Stacked Ensemble Machine Learning (SEML) model, which used the composition (the elements and their amounts) to predict how strong or flexible an alloy would be.
The second model was a Convolutional Neural Network (CNN), a type of AI usually used for analyzing images, but here, it was used to examine how atoms of different elements were arranged next to each other inside the alloy, as this structure has a big impact on how the metal behaves.
Once these models could predict properties accurately, the team used algorithms to improve the results. These algorithms generated many possible metal combinations, kept the best ones, and mixed them to create even better versions. Over time, the system was able to zoom in on the most promising alloy designs.
“Our design workflow, combining advanced machine learning and evolutionary algorithms, provides interpretable insights into materials’ structure-property relationships, offering a robust approach for the discovery of diverse advanced materials,” Fangxi Wang, first author and a postdoc researcher at Virginia Tech, said.
The researchers are confident that their work will lead to the design of embedded engines that not only sound quieter but are also quiet when measured.
“By linking turbulent flow ingestion patterns to how people perceive noise, we are giving engineers the tools to design future aircraft that truly sound as quiet as they look,” added Ahmed in the press release.
The efforts are crucial in light of the European Union’s mandate to reduce aircraft noise by 65 percent by 2050. The researchers believe that their work will aid in design of quieter aircraft to be used for large scale air transport as well as for electric vertical take-off and landing (eVTOL) aircraft in the urban air mobility (UAM) sector.
When the engines operate at high thrust, during take-off, the fan suction is much stronger and disrupts the airframe boundary layer flow. This produces an extremely unsteady turbulent flow alongside fan-induced flow distortion. The interaction between fan blades and the distorted air flow results in fan haystacking, where the large span of the rotating blades further slices the unsteady flow.
“These two hidden sound signatures make future embedded aircraft engines feel perceptually irritating, not just loud,” explained Feroz Ahmed, who was involved with the work while at the University of Bristol in a press release.
Novel approach to robot training
This approach is different from traditional methods that often rely on extensive trial-and-error learning or costly imitation of large groups of human experts.
HUMAC leverages the human capacity to understand and predict the intentions of others. It enables robots to learn to anticipate their teammates’ actions, adapt strategies dynamically, and solve challenges demanding coordinated, collective intelligence akin to human teams.
The HUMAC framework involves brief instances where a human operator takes control of individual robots within a team during training, offering guidance at crucial strategic junctures – much like a coach providing targeted advice during a game.
These short demonstrations teach the robots intricate collaborative tactics, such as setting up ambushes and encircling targets.
The study authors focused on MPEA, special alloys composed of more than three elements (in roughly equal proportions) that exhibit remarkable mechanical properties. These materials are much stronger, tougher, and more resistant to heat and wear than conventional alloys.
They are already used in aircraft parts, surgical implants, and clean energy systems. However, the traditional method of creating MPEA involved testing numerous combinations of elements. This process is costly and time-consuming.
To overcome this challenge, Deshmukh and this team developed a smart, data-driven system that combines the power of machine learning and algorithms. They started by gathering a large dataset of existing MPEA.
Basically, information like which metals were used, how those metals were arranged at the atomic level, and how each alloy performed in terms of strength, flexibility, and other mechanical properties was fed to the system.
Two machine learning models were trained using the information. The first was a Stacked Ensemble Machine Learning (SEML) model, which used the composition (the elements and their amounts) to predict how strong or flexible an alloy would be.
The second model was a Convolutional Neural Network (CNN), a type of AI usually used for analyzing images, but here, it was used to examine how atoms of different elements were arranged next to each other inside the alloy, as this structure has a big impact on how the metal behaves.
Once these models could predict properties accurately, the team used algorithms to improve the results. These algorithms generated many possible metal combinations, kept the best ones, and mixed them to create even better versions. Over time, the system was able to zoom in on the most promising alloy designs.
“Our design workflow, combining advanced machine learning and evolutionary algorithms, provides interpretable insights into materials’ structure-property relationships, offering a robust approach for the discovery of diverse advanced materials,” Fangxi Wang, first author and a postdoc researcher at Virginia Tech, said.
The researchers are confident that their work will lead to the design of embedded engines that not only sound quieter but are also quiet when measured.
“By linking turbulent flow ingestion patterns to how people perceive noise, we are giving engineers the tools to design future aircraft that truly sound as quiet as they look,” added Ahmed in the press release.
The efforts are crucial in light of the European Union’s mandate to reduce aircraft noise by 65 percent by 2050. The researchers believe that their work will aid in design of quieter aircraft to be used for large scale air transport as well as for electric vertical take-off and landing (eVTOL) aircraft in the urban air mobility (UAM) sector.
When the engines operate at high thrust, during take-off, the fan suction is much stronger and disrupts the airframe boundary layer flow. This produces an extremely unsteady turbulent flow alongside fan-induced flow distortion. The interaction between fan blades and the distorted air flow results in fan haystacking, where the large span of the rotating blades further slices the unsteady flow.
“These two hidden sound signatures make future embedded aircraft engines feel perceptually irritating, not just loud,” explained Feroz Ahmed, who was involved with the work while at the University of Bristol in a press release.
Novel approach to robot training
This approach is different from traditional methods that often rely on extensive trial-and-error learning or costly imitation of large groups of human experts.
HUMAC leverages the human capacity to understand and predict the intentions of others. It enables robots to learn to anticipate their teammates’ actions, adapt strategies dynamically, and solve challenges demanding coordinated, collective intelligence akin to human teams.
The HUMAC framework involves brief instances where a human operator takes control of individual robots within a team during training, offering guidance at crucial strategic junctures – much like a coach providing targeted advice during a game.
These short demonstrations teach the robots intricate collaborative tactics, such as setting up ambushes and encircling targets.