Automating material image analysis for material discovery
- PDF / 3,452,696 Bytes
- 11 Pages / 612 x 792 pts (letter) Page_size
- 82 Downloads / 252 Views
rtificial Intelligence Prospective
Automating material image analysis for material discovery Chiwoo Park, Department of Industrial and Manufacturing Engineering, Florida State University, Tallahassee, FL 32310, USA Yu Ding , Department of Industrial and Systems Engineering, Texas A&M University, 3131 TAMU, College Station, TX 77843, USA Address all correspondence to Yu Ding at [email protected] (Received 22 January 2019; accepted 2 April 2019)
Abstract Advancements in temporal and spatial resolutions of microscopes promise to expand the frontiers of understanding in materials science. Imaging techniques produce images at a high-frame rate, streaming out a tremendous amount of data. Analysis of all these images is time-consuming and labor intensive, creating a bottleneck in material discovery that needs to be overcome. This paper summarizes recent progresses in machine learning and data science for expediting and automating material image analysis. The discussion covers both static image and dynamic image analyses, followed by remarks concerning ongoing efforts and future needs in automated image analysis that accelerates material discovery.
Introduction Material scientists have long desired to image, with high-spatial and temporal resolutions, material structures, and interactions in chemical, material, and biologic processes, because visual information with sufficient details is vital to discovery as emphasized in many scientific activities.[1] Major research efforts have been made to develop material imaging tools, including various kinds of electron microscopes (EM)[2] and scanning probe microscopes (SPM).[3] The wide use of these microscopes leads naturally to the exponential increase in material imagery data, which in turn calls for efficient and effective analysis methods to extract useful information from the raw images for aiding designs, discoveries, and decision-makings in materials science. An EM operates with an electron beam, which illuminates materials to create an image of materials. A scanning electron microscope (SEM) creates images based on electrons reflected by material surfaces, whereas a transmission electron microscope (TEM) creates images based on electrons transmitted through material samples. When the electron beam in a TEM rasters across the sample, the resulting technique is called scanning transmission electron microscopy (STEM). Historically, EM techniques were capable of capturing only a static image, because an EM requires samples to be placed in a high-vacuum environment, meaning that the samples are typically frozen or dried before imaging. Recent progress in environmental in situ EM, including liquid phase EM and in situ gas EM, allows direct, high magnification, and real-time imaging of material structures or objects during experiments, capturing their evolution at high-imaging speeds.[4–11] By their very nature, these imaging strategies generate material images at a high-frame rate, streaming out a tremendous amount of image data.
An SPM makes use of the dynamic interaction between a
Data Loading...
