Presenter Profile
Kristen Hernandez
PhD Student
Case Western Reserve University, Department of Materials Science and Engineering
Kristen Hernandez holds a Bachelor of Science degree in chemistry at Youngstown State University with an emphasis on solid state materials and analytical chemistry. Her interest was recognized by the local chapter of the American Chemical Society. Kristen honed her expertise in analytical chemistry during her five-year tenure at Element Materials Technology as analytical chemist and metallographer for the product certification to national and international standards for aerospace application. Her certification for evaluations involves a wide range of materials and characterization techniques. Currently, she is pursuing a Ph.D. in Material Science and Engineering at Case Western Reserve University (CWRU), where she contributes to the Material Data Science for Stockpile Stewardship Center of Excellence (MDS3-COE) which emphasizes an interdisciplinary approach to common material challenges by blending material science and computer data science skills. Kristen's passion and background drive her ongoing development of computational skills to address real-world material science challenges.
TALK TITLE
Advanced Manufacturing applications in Additive Manufacturing for Automated In-situ and Ex-situ feature extraction to predict defect formation
KEYWORDS
Advanced manufacturing, real time monitoring, automated pipelines
ABSTRACT
Metal-based additive manufacturing (AM) requires active monitoring for assessing the quality and reliability of printed parts and waste reduction. Real-time monitoring allows for impromptu defect detection, making it feasible to reject parts, correct errors, and prevent releasing faulty products. Techniques such as pyrometry and high speed camera imaging are common and well-explored for in-situ part monitoring through the use of cloud computing, machine learning, and other Industry 4.0 Advanced manufacturing innovations. This work focuses on developing an automated pipeline for feature extraction of pyrometry measurements and melt pool geometric properties as well as identifying features found in micro-X-ray Computed Tomography images of completed parts. Automated processes enable massive throughput of characterization techniques following Advanced manufacturing objectives for AM.
This monitoring techniques output qualitative and quantitative information about the conditions of the laser powder bed fusion (L-PBF) build, which are spatiotemporally coordinated. Examining the evolution of melt pool geometry to detect possible part failures requires large-scale data analytics to retain large amounts of data and perform calculations with multimodal variables. We leveraged open-source computer vision algorithms such as deep learning tools You-Only-Look-Once(YOLO) and U-Net alongside traditional image processing techniques from scikit image Python package to characterize melt pool properties. In this study, we examine 715 individual prints, with 283 GB data using high-performance computing (HPC) and Hadoop distributed file system (HDFS). This study found that a melt pool that exceeds an area of 0.04 mm2 and pyrometry signal over a threshold (375 pyrometry units) were more likely to generate defects. Different image extraction techniques show varying accuracy and predictive description which allows for more optimized approaches for predicting the print reliability of L-PBF AM parts.