Sandia National Laboratories’ Machine Learning algorithm could provide a quicker and more economical way for testing bulk materials in the automotive, aerospace, and other industries.
When production is halted, it leads to heavy expenses. As a result, manufacturers test materials such as sheet metal for moldability before they are utilized. This is for ensuring that they will not be ruptured when stamped, stretched, and strained as they are formed into different parts.
According to Sandia scientist David Montes de Oca Zapiain, the paper’s lead author, Organizations are frequently utilizing commercial simulation software fine-tuned to the results of different mechanical tests. These tests, however, can take months to complete.
Montes de Oca Zapiain also stated that while certain high-fidelity computer simulations have the ability for evaluating moldability in a matter of weeks, organizations require access to a supercomputer as well as specialized competence for running them.
Sandia has demonstrated that machine learning can significantly reduce the time and resources required for calibration of commercial software because the algorithm does not require information from mechanical tests, according to Montes de Oca Zapiain. The method also does not necessitate the use of a supercomputer. Furthermore, it paves the way for faster research and development.
Montes de Oca Zapiain stated that this algorithm can be used efficiently for finding lighter materials with fewer resources while maintaining safety and accuracy.
Replacement of mechanical tests with algorithms
MAD3, which stands for Material Data-Driven Design and is pronounced as “mad cubed.” It is a machine-learning algorithm that works due to metal alloys that are made of microscopic, “crystallographic” grains. These grains combine to form a texture making the metal stronger in certain directions than others, a phenomenon known as mechanical anisotropy.
Montes de Oca Zapiain explained that the models are trained so that they can comprehend the association between crystallographic texture and anisotropic mechanical response. An electron microscope is required for obtaining the texture of the metal. After obtaining that information, it can be fed into the algorithm which will forecast the data required for simulation software without the performance of mechanical tests.
Sandia, in collaboration with Ohio State University, trained the algorithm based on the results of 54,000 simulated material tests utilizing a technique known as a feed-forward neural network. The Sandia team then fed the algorithm 20,000 new microstructures for testing its accuracy, comparing the algorithm’s calculations to data collected from experiments and also supercomputer-based simulations.
Hojun Lim, a scientist from Sandia who also contributed to the research, stated that in comparison with high-fidelity simulations, the developed algorithm is approximately 1,000 times quicker. Active work is in progress for improving the model by integrating the advanced features for capturing the evolution of anisotropy, which is required for accurately predicting the material’s fracture limits.
Sandia, a national security laboratory, is conducting additional research to investigate if the algorithm can curtail quality assurance procedures for the United States nuclear stockpile, where materials must meet stringent standards before they are accepted to be used in production. The National Nuclear Security Administration financed the ML experimentation via the Advanced Simulation and Computing program.
Sandia formed a cross-disciplinary team for developing user-friendly, graphics-based Material Data-Driven Design software, allowing other institutions to benefit from the technology. It was created with the help of over 75 interviews with prospective users assisted by the Department of Energy’s Energy I-Corps program.
Source link
