Artificial intelligence promotes the development of materials manufacturing

Research teams from the Massachusetts Institute of Technology, the University of Massachusetts at Amherst, and the University of California at Berkeley hope to use a new artificial intelligence system to automate the gaps in materials science. This system uses research papers to mine and use it to produce specific materials. Source: Chelsea Turner/MIT In recent years, research work such as the Materials Genome Project and the Materials Project has produced a large number of computational tools for designing new applications in a range of applications from energy, electronics to aviation, and civil engineering. material. But the process of developing these materials still relies on a combination of experience, intuition, handwork, and literature reviews. Research teams from the Massachusetts Institute of Technology, the University of Massachusetts at Amherst, and the University of California at Berkeley hope to use a new artificial intelligence system to automate the gaps in materials science. This system uses research papers to mine and use it to produce specific materials. Elsa Olivetti, Assistant Professor of Energy Research in the Department of Materials Science and Engineering (DMSE) at the Massachusetts Institute of Technology in Richfield, Atlantic, said: 'Computational materials scientists have made great progress in 'what we are going to do.' But it is precisely because of these achievements. , The problem has become 'However, what should I do now? 'The researchers envisioned a database containing material preparation methods extracted from millions of papers. Scientists and engineers only need to enter the name of the target material and any other criteria (precursor material, reaction conditions, manufacturing process), and the system will Give a suitable preparation plan In order to realize this idea, Olivetti and his colleagues developed a learning-only system that can analyze research papers. This system can infer that the part of the paper contains material preparation methods and classify keywords according to the preparation steps: Target material name, quantity, equipment name, operating conditions, descriptive adjectives, etc. A paper published in the latest issue of 'Material Chemistry' proves that the intelligent learning system can infer the general characteristics of the material category based on the extracted data , Such as the temperature required for material synthesis, or the specific characteristics of different materials prepared by different physical methods due to changes in preparation conditions. Olivetti is the main author of the paper, and the other authors are MIT graduate students Edward Kim and DMSE Postdoctoral fellow Kevin Huang, computer scientists Adam Saunders and Andrew McCallum of UMass Amherst, and Gerbrand Cede, Dean and Professor of the Department of Materials Science and Engineering at Berkeley. Filling the gaps. Researchers use a combination of supervised and unsupervised artificial intelligence learning techniques to build their System. 'Supervision' means that the test data supplied to the system is manually labeled; the system will try to find the correlation between the original data and the labeled data. 'Unsupervised' means that the test data is not labeled, and the system learns to The data categories come together. Because the extraction of material preparation methods is a completely new field, Olivetti and her colleagues do not have the luxurious annotated data sets that different research teams have accumulated over the years. So they can only get about 100 from themselves Many papers annotate their data. The artificial intelligence learning standard is a very small data set. In order to improve it, they used Google to develop the algorithm of Word2vec. Word2vec can view the context in which words appear, that is, the syntactic structure in sentences And other words around it, and combine words with similar contexts. So, for example, if one paper contains the sentence 'we heat titanium tetrachloride to 500°C'Heated to 500℃' sentence, Word2vec will combine the two keywords 'titanium tetrachloride' and 'sodium hydroxide' together. With the use of Word2vec, the intelligent learning system can infer the value of any given word Tags and the keyword classification it applies to, so researchers can build their system around about 640,000 papers instead of just 100 papers, so researchers can maximize their data sets. The tip of the iceberg, however, because They do not have a standard for evaluating the performance of unlabeled data, so the accuracy of their test system can only be based on Rely on labeled data. In these tests, the system was able to identify paragraphs containing keywords with an accuracy of 99%, and label the words in these paragraphs with an accuracy of 86%.

Professional tensile tester manufacturers also understand that when you're working with tensile tester manufacturers product, it's important to understand that quality of textile testing equipment always matters.

GESTER is the vital link in the supply chain, adding value with efficient and cost-effective service and solutions for our customers and our suppliers.

Simultaneously being able to offer not only the product but also the service, gives the customer a quality 'one-stop-shop' service.

GESTER International Co.,Limited constantly discovers the demands of global market for developing a wide range of products applied in different use.

Increasing consumer awareness and rising concern about improving tensile tester manufacturers are driving the market of products.