Protein Crystallization Image Classification Using Machine Learning

Team members: Jordan Nelson, Karl Yang, Shuyan Zhao, Jingyu Sun, Stanley Kogin

Mentor: Chen Qian

Company: AstraZeneca

AstraZeneca is a global research-based biopharmaceutical company that discovers, develops and provides life-changing medicines to people all over the world. It is one of the many companies to span the entire life-cycle of a medicine from discovery, research and development to manufacturing, supply and commercialization of primary care and speciality care medicines. To develop effective medicines, it is important to first understand the internal structures of DNAs and proteins. Therefore, efforts have been put in to study the structures of DNAs and proteins through crystallographic analysis or X-Ray Diffraction by the company. In the studying of the crystal structures, experimental methods can be complemented with data science approaches to achieve high efficiency and avoid human subjectivity. The purpose of this project is to use the data science approach to design a deep learning algorithm for the classification of various crystallisation inspection images.

There is a large amount of ongoing research into protein structures and how they affect protein function. The structures are large, complex, and amorphous, making characterization challenging. Protein characterization has been more accurate and efficient with crystallizing the protein, which is essential for a variety of biological and biomedical research. Currently, 90% of the characterized protein structures have been determined in this way through x-ray diffraction. The processing methods used in AstraZeneca for crystallizing the proteins are being tested to determine the processing method that makes the most and largest crystals. This has also led to a large amount of images, which cannot be processed manually in a timely manner.

In order to distinguish between samples with crystals and samples without in an efficient manner, a convolutional neural network is used to label each image. The four possible labels are as follows: clear, crystal, precipitate, and other. For training and model accuracy testing, the open-source MARCO dataset is used. The images have been pre-labeled and have been split between a training dataset and a validation dataset. The dataset is then cleaned with an autoencoder using a convolutional neural network to remove duplicate images. The model and data is being run and stored on Microsoft Azure for increased computational time and power. Once an accuracy of above 85% is achieved with the model, the data AstraZeneca has collected will be labeled by the model.

Thermoplastic Chopped Fiber Static Properties Investigation

Team members: Will Gottsch, Devin Mays, John Leitch, Julia Flaherty, Tanya Wang

Mentors: Devin Knowles, Farzad Zafari, Navid Zobeiry

Company: Boeing

The purpose of this capstone design project is to partner with a technical team at Boeing to investigate the static properties of thermoplastic chopped carbon fiber composites and predict how these properties change with variations in the orientations of the fibers. The results of this project will give early information to Boeing to help streamline design processes for complex parts like a clevis. With this data, Boeing can more effectively map out the alignment of complex carbon fiber parts to handle the unique stress conditions of these parts, and ultimately eliminate the need to test individual parts. Additionally, utilizing thermoplastic carbon fiber parts gives Boeing more opportunities for diverse, lightweight, and strong materials to manufacture quality airplanes.

There are four main tasks in this project. First, we developed a test method to determine the minimum mechanical properties of our composite materials. We focused on three alignments; 0°, 90° and random orientation for tension, compression, and open hole tension mechanical testing. These methods were chosen based on their emulation of complex composite part systems like the clevis structure. Second, a square alignment sieve was built out of wooden boards and sheet metal dividers to align the chips in a chosen direction in accordance with the predetermined test matrix. From here, we focused on our third task of working with our Boeing mentors to design and manufacture steel tooling at their facility to heat and press composite panels in our desired mold. These panels are to be composed of Boeing’s own carbon fiber chips (carbon fibers in a PEKK matrix sourced from Toray Composites). When the panels are molded, they can be cut into sample coupons ready for the mechanical testing. The fourth and final task for this project is to perform the tests on the coupons at UW laboratories and record data for each item in the test matrix. If time permits, we will complete a finite element analysis (FEA) based on the data we have obtained from our mechanical tests on a theorized complex carbon-fiber component. This data and further simulations can allow engineers to interpret this information for design applications and uses with other complex part geometry.

Boeing Mechanical Fixture of Thermoplastic Composite

Team members: Alexander Hicker, Shulong Mo, Johnathan Emerson, Jennifer Prasetyo

Mentors: Mohamed Azdamou, Pradeep Krishnaswamy, Ashley Tracey, Larry Ridgeway.

Company: Boeing

Thermoplastic matrix composites have gained attention as an alternative to thermosets due to their lack of shelf life, recyclability, and their ability to melt and resolidify. Aerospace companies such as Boeing are interested in repairing these parts using epoxy and other thermoset materials, but require additional surface treatment due to their lower surface energy. Energetic surface treatment methodologies like atmospheric plasma, laser treatment, and UV light have shown promise as a robust method to increase the surface energy before structural bonding. Currently, these surface treatments are controlled by bulky robotic arm systems which are often impractical in terms of costs and accessibility to parts on aircrafts. The objective of this project was to design and build a prototype mechanical fixture/guidance system that is less expensive and able to attach to the aircraft in any orientation. The device was designed to raster the nozzle over the surface utilizing numerical controls. The flexibility of the structure allows for surface adaptation to maintain constant height between the surface and nozzle. These design considerations are visualized in CAD models and will be tested through a built prototype at Boeing facilities.

Investigation of Transport Erector Bearing In-Service Intervals

Team members: Jayson Haury, Jerry Hung, Walid Mouss, Xitlalit Sanchez-Martinez, Emerson McNamee

Mentors: Tuesday Kuykendall - Senior Manager of Material and Processes Laboratories, Brian Dykas - Senior Material and Process Engineer

Company: Blue Origin

Blue Origin is a privately owned spaceflight company based in Kent, Washington. Their mission is to build a road to space using safe and reliable reusable launch vehicles. With this, the company aims to aid in the discovery of unlimited space resources to better the quality of life of people everywhere.

Within the aerospace industry, transport erectors are used to maneuver a shuttle or rocket from a horizontal position into a vertical one. These highly loaded critical structures are often operated in coastal environments, and are therefore prone to corrosion. This investigation focuses on the bearings found within these complex systems. The overall structure of the transport erector was left vague for the purposes of our investigation, thus custom test methods were developed to observe the materials’ behavior when subjected to a corrosive environment. By studying the corrosion and wear response of metals that are commonly used in bearing systems, there is the potential for increased cost saving opportunities, not only in the manufacturing of the part but also in the maintenance performed throughout its lifetime. In order to successfully simulate in-service conditions, carefully selected metals were subjected to heat treatment, corrosion testing, hardness testing, and tribology testing. From this data, in-service inspection intervals were developed to be used in the maintenance of a bearing part found on a transport erector.

3D Forming of Thin Metal Sheets

Team members: Renjie Song, Ryan Van Der Hoeven, Janet Diep, Jason Samson, Cole Hofstrand, Derek Chang

Mentor: Jed Brich, Janicki Industries, Inc.

Company: Janicki Industries

Janicki Industries is a prominent and privately owned Engineering and Manufacturing company in the PNW. Their specialty is advanced composite materials and exotic metals for uses with aerospace, marine, energy, space, military, transportation, and architecture. Their R&D lab pushes the boundaries of composite fabrication materials and techniques based on customer needs.

We, as Materials Science and Engineering students at the University of Washington, welcome the opportunity to work with Janicki in our capstone project of finding metals/alloys that have a similar coefficient of thermal expansion (CTE) to either carbon fiber or glass fiber reinforced polymers. These metal candidates should be able to form complex, three-dimensional shapes. Finding new applicable metals can allow for a reduction of material waste, energy waste, and cost. From material and forming process selection to experimental verification, we are excited to share our findings and candidate metals. Our final verification experiment is forming the metal into a complex shape using a ceramic mold we created.

Improving BaTiO3 Dielectric Capacitor Operating Temperature

Team members: Kristine Lam, Kavish Chandra, Sarah Little, Chase Mersberg, Savannah Camacho

Mentors:
Industry Mentor: Jeff Day
Faculty Mentor: Jihui Yang represented by Parker Steichen

Company: MWD Technology & Innovation Center LLC

MWD Technology & Innovation Center LLC & CalRamic Technologies LLC specialize in processing low and high voltage ceramic capacitors. These companies seek to push the limits of current class 2 dielectrics by processing dielectrics operable above 225˚C with greater than 10 M-ohms of resistance. Such devices would make waves in aerospace systems, satellite systems, military systems, geothermal detection devices, etc. The UW team is working with current MWD BaTiO3 dielectrics, testing the effects of compositional changes and modified sintering conditions on the device resistivity using electrical impedance spectroscopy. The changes to processing that were carried out were selected with the intent of limiting oxygen vacancy conductivity, which is one of the key contributors to leakage current in dielectrics at high temperatures. Ultimately, the goal of the project was to raise operating temperature for BaTiO3 dielectrics in capacitors without reducing charge storing ability and to build a framework for future efforts.

Large Part Light Weighting

Team members: Keli’i Clark, Logan McKinney, Lucas Palacio, An Pham, Genya Shimada, Brooke Taylor

Mentor: Jordan Kiesser

Faculty adviser Aniruddh Vashisth

Company: PACCAR

The main design objective of this project is to select a new composite material for the roof structure of the Peterbilt 579 Ultraloft truck manufactured through PACCAR. Our goal for the new roof structure is to select a composite material that will provide better mechanical properties than the current randomly oriented glass fiber sheet molded composite (SMC) while not spending more than $3 per pound of weight saved in overall structure. Success of this project is achieved through mechanical testing of selected carbon fiber composites to compare to the original SMC material. A cost analysis on the manufacturing process for the new composite will be generated to determine how much weight we will be saving. We will also be performing a roof crush test simulation through finite element analysis (FEA) to demonstrate our new composite will meet and exceed the properties of the old material. Overall, this new roof composite will not cost more than $3 per pound of weight saved on the roof structure and have a better performance compared to the original SMC material.

Toray Data Management System

Team members: John Foster, Vivian Huynh, Erin Mee, Yulun Wu

Mentors: Ben Rutz, PhD.

Faculty adviser: Aniruddh Vashisth

Company: Toray Composite Materials America, Inc.

Toray is a cutting edge carbon fiber composites manufacturing company continuously developing high performance materials for companies like Boeing whose aircraft have increasingly utilized composites for their high strength to weight ratio. The composite development Toray is involved in generates vast amounts of data. This valuable data is currently only being stored in Windows files system and cross-experiment analysis is only possible through Excel. Our team was tasked with creating a data management tool to improve data accessibility and streamline cross-experiment analysis. Our solution is capable of storing raw data, recording extracted values from data sets (e.g. Tg values), and analyzing sample sets based on designated parameters (e.g. “look at all composite samples with tensile moduli above ___ MPa”) all through a user-friendly UI. To store and retrieve the data, MySQL was utilized to design a relational database. This was achieved utilizing MATLAB in conjunction with a MySQL database. MATLAB app designer was used to create custom UIs to input, output, and analyze data stored in the MySQL database. The database was designed following relational database normalization forms to reduce redundancy and ensure data integrity. This improved data management has the potential to be the groundwork for advanced data science technologies at Toray which are increasingly becoming essential in the research and development of modern materials.

Connecting School of Medicine and College of Engineering Researchers using Machine Learning

Team members: John Foster, Vivian Huynh, Erin Mee, Yulun Wu

Mentors: Dr. Yi Wang, Dr. Luna Huang

Company: UW College of Engineering

The Office of Sponsored Programs (OSP) has noticed a significant amount of duplicate or similar proposals that were competing for funding. The OSP suggested a project to identify faculty from the College of Engineering and School of Medicine with similar research interests and introduce them to each other to facilitate research collaboration between the two departments. We collected a database of 1,800 researches and 100,000 papers and applied machine learning to create a representation of their interests and determine research similarities. With these representations, we found that past collaborations can be correlated to each collaborator’s research background. These correlations can be used to encourage future collaboration between researchers with similar backgrounds but no prior connection.