AI for Big Data Problems in Advanced Imaging, Materials Modeling and Automated Synthesis: Poster Session
Sponsored by: TMS Advanced Characterization, Testing, and Simulation Committee
Program Organizers: Mathew Cherukara, Argonne National Laboratory; Subramanian Sankaranarayanan, University of Illinois-Chicago; Badri Narayanan, University of Louisville
Monday 5:00 PM
October 10, 2022
Room: Ballroom BC
Location: David L. Lawrence Convention Center
B-1: Autonomous Closed Loop Synthesis of Gold Nanorods via a Modular Chemical-Handling Robotic Platform: Morgan Chen1; Ari Fiorino1; Aarti Singh1; Reeja Jayan1; 1Carnegie Mellon University
Materials research campaigns often encounter challenges with navigating complex and high-dimensional parameter spaces to uncover scientific insights. Consequently, productivity can be hindered by the practical limitations of obtaining and analyzing vast or sufficient experimental datasets required to obtain a functional understanding of a phenomenon of interest. We build upon a modular programmable chemical-handling device as a hardware platform to leverage machine learning algorithms to accelerate the pace at which scientific conclusions can be extracted from minimal datasets or large experimental parameter spaces. We present a closed-loop autonomous system to optimize the conditions for targeted synthesis of gold nanorods using in-line UV-Vis spectroscopy. Although they have diverse functions in electronic, optical, and biomedical applications, gold nanomaterials are confronted with suboptimal production reproducibility and yield. Therefore, we harness algorithm-driven mechanization to enhance the precision, fidelity, and yield of gold nanorods.
B-2: Logistics Box Recognition in Robotic Industrial De-palletizing Procedure with Systematic RGB-D Image Processing Supported by Multiple Deep Learning Method: Jonghun Yoon1; Jooyeop Han1; Thong Nguyen1; Hyunggyu Kim1; 1Hanyang University
In an automation system of box de-palletization utilizing robots, the quality of vision-based box recognition is considered highly depending on appearance of box containing multiple types of labels and tags. Usually, a large-scale vision dataset is required to detect on diverse complexity conditions, which also need a lot of effort to construct. This paper aims to develop a systematic image processing algorithm to remove unnecessary portion and emphasize the key features. The core of the algorithm is image transformation steps utilizing the consistent generative adversarial network (Cycle GAN) for removing main obstacles of recognition such as adhesive labels or tapes. Also, the depth map-based feature extraction is proposed to emphasize required features such as boundaries of boxes. These pre-processed images are used as inputs to Mask R-CNN for vision-based de-palletizing guideline, which was validated with 400 test cases under conditions of complex surface patterns and spatial arrangement.