Researchers of Xi'an Jiaotong University use machine learning to combine domain knowledge to quickly develop high
As the complexity of material composition and microstructure increases, data technologies such as machine learning are being used by researchers in the development of new materials. However, the lack of material data caused inaccurate machine learning model predictions, which affected the efficiency of material development. Thankfully, not only data in material science, but also rich "domain knowledge", such as empirical models and thermodynamic theory. How to effectively combine data algorithms with "domain knowledge" to further accelerate the development of new materials is a common challenge facing researchers. Recently, researchers from the strength teaching and research section of the School of Materials Science and Technology of Xi’an Jiaotong University have proposed that “domain knowledge†can be used to effectively narrow the search space for unexplored materials. The idea of ​​using machine learning to develop new materials in the search space is further accelerating the efficiency of machine learning algorithms in searching for target materials. The researchers took the development of lead-free ferroelectric ceramics with high energy storage density under low electric field as a case study. Using the "domain knowledge" of the composition of the transition region in the composition-temperature phase diagram to have higher energy storage density, we successfully designed and synthesized ferroelectric ceramics with high energy storage density in a simple system. However, this strategy is only applicable to simple systems and cannot guide researchers to quickly search for target materials in complex systems (multi-doped systems). This prompted researchers to use two data-driven strategies to solve optimization problems in complex systems. Strategy I uses pure data-driven methods to directly search for target materials in a huge unknown space (about 9 million possibilities), and found ferroelectric ceramics with high energy storage density in the sixth experimental iteration. Strategy II first uses "domain knowledge" to pre-select the material composition in the transitional area in the unknown space, and then searches in the pre-selected small space (about 700,000 possibilities), and finds that it has good performance through only one experimental iteration Target material (the energy storage density is higher than the optimal value found in Strategy I). The above results indicate that the strategy of combining "domain knowledge" and machine learning is more efficient than the pure data-driven research strategy. At the same time, it also provides inspiration for the performance optimization of other complex material systems containing a large number of extreme values. The research result was recently "Accelerated Search for BaTiO3-Based Ceramics with Large Energy Storage at Low Fields Using Machine Learning and Accelerated Search for BaTiO3-Based Ceramics with Large Energy Storage at Low Fields Using Machine Learning and Experimental Design), published in the comprehensive journal "Advanced Science" (Advanced Science). Professor Xue Dezhen and Professor Ding Xiangdong of the School of Materials Science and Engineering of Xi'an Jiaotong University are co-corresponding authors of this paper, doctoral student Yuan Ruihao is the first author, and the State Key Laboratory of Metal Material Strength of Xi'an Jiaotong University is the first corresponding author of this paper. The research was jointly funded by the National Natural Science Foundation of China, the National Key R & D Program, and the 111 Project.
Pressure Gauge for ships
Pressure Gauge For Ships,Ship Pressure Gauge,Ships Pressure Gauge,Ships Oil Pressure Gauge wuxi kaifeng pressure gauge co., ltd , https://www.wxkfmanometer.com