Industry-Sponsored Student Capstone Projects

Ghost shrimp burrowing in Willapa Bay and Grays Harbor damage oyster beds by causing oysters to sink and suffocate, and the pesticide control option, carbaryl, is now banned. In response to Washington state’s direction to research more sustainable pest management for shellfish growers, this project advanced a bubble-based delivery approach for shrimp control agents in saltwater environments. This work was conducted in collaboration with the University of Washington’s Pozzo Laboratory and Massa Products, a leader in ultrasound design and production. The system aimed to bind substances such as clove oil, PyGanic, or Chitosan to stable micro- and nanobubble emulsions that could move into shrimp burrows while oyster beds were submerged. Underwater ultrasound transducers were intended to collapse the carrier bubbles and release the agents directly into the shrimp habitat, enabling treatment at high water rather than only during rare low-tide conditions. The work focused on refining emulsion stability and reproducibility, preparing samples for ultrasonic bursting tests, and considering how the materials could be produced and supplied at a scale relevant to Washington’s oyster industry.

Scaniano is a platform that allows people of all ages to make music without prior experience. Users arrange labeled note cubes on a board, and a scanner is used to convert the blocks to music notes for playback, which can be encoded into QR codes and printed as labels for physical songbooks. Scaniano2 is a continuation of this project, which significantly improved the hardware and features of the prototype. Key upgrades include a custom touch-screen interface and optical music recognition, which allows users to scan both blocks and music sheets with a camera. By integrating the camera and printer interfaces, Scaniano2 provides a scalable, user-friendly workflow wrapped in an intuitive app.

Seattle City Light is replacing manual distribution switches with remotely controllable units, with 77 installed since 2016 and several hundred potential locations identified. Deployment has been limited in part by reliance on fiber optic communications, prompting the need to evaluate alternative approaches (such as cellular) that could reduce cost and accelerate installation while maintaining comparable reliability. This project team's solution supports Seattle City Light's smart switch deployment via data-driven site qualification. Students built a battery-powered LTE signal quality logger measuring RSSI, RSRQ, RSRP and SINR over months optimized for ultra-low power on our custom PCB designed for outdoor deployment. Robust firmware pairs with a scalable website to visualize device health, GPS location, ping time and outage patterns in real time. This end-to-end pipeline reduces installation risk and enables informed deployment decisions for switch locations, accelerating Seattle's shift from legacy grid infrastructure to a connected smart electric grid.

Spada Lake supplies raw water to much of Snohomish County and supports power generation at the Jackson Hydroelectric Project before discharging to the Sultan River and the City of Everett’s filtration system. As water demand increases and snowpack and precipitation become less predictable, interest has grown in storing more water in late spring for use during summer and fall. Because Spada Lake is formed by Culmback Dam and uses a morning-glory spillway, a seasonal spillway-raising gate was considered as a practical way to increase storage. The project aimed to develop an early-stage design concept for a system to raise and lower the spillway seasonally, and to examine the potential value and impacts of that added storage. This included reviewing cost-benefit tradeoffs at different storage heights, identifying needed structural analyses, assessing possible inundation effects around the lake, interpreting climate-related hydrology changes in the Sultan Basin, and summarizing operational considerations for making better use of increased summer water storage.

In partnership with Snohomish PUD, this team developed multiple microgrid designs that aim to reduce utility costs and provide outage support for the critical loads of the remote community of Darrington, WA. HOMER PRO and Helioscope were utilized to recommend optimized battery and solar array sizes. Additionally, design aspects were developed for a microcontroller that maintains system stability between changing operation modes and worst-case scenarios. While designs supporting existing load profiles were technically viable, their associated costs were exorbitant enough that the team also designed and recommended a full HVAC retrofit, enabling a significantly more cost-effective microgrid solution.

As utilities work to incorporate Distributed Energy Resources (DERs) into the distribution grid, new communications and control schemes will need to be developed to fully realize the benefits of distributed generation resources and flexible electrical loads. Communication between devices and utilities are becoming more standardized, but most utilities do not have a framework in place to send price or dispatch signals to non-utility owned devices. Snohomish PUD was looking for this student tam to explore the challenges and feasibility of developing the hardware and software systems needed for the utility to begin utilizing customer-owned DERs for utility benefit. This team developed a simulated client–server environment modeling utility-issued control events and device responses. The system includes device identification, event scheduling, polling-based communication, and status reporting. Results demonstrate the feasibility of standardized DER communication while highlighting challenges in certificate management and interoperability, informing scalable real-world deployment.

Starbucks sought a more tailored way to set up inventory materials at the coffeehouse level, since standard guidance does not reflect differences in individual business patterns and could lead to excess waste or insufficient stock. This project focused on developing a data-driven model to track and quantify inventory materials across coffeehouses by analyzing material flow patterns, historical sales data, coffeehouse-level ingredient usage reports, and observations collected in the Starbucks TRYER Innovation Center. The resulting capability was intended to identify key factors that influenced effective coffeehouse setup and highlight opportunities to improve production throughout the business day. By turning store-level data into actionable inventory insights, the project aimed to support more efficient operations, reduce waste, and improve cost management while maintaining Starbucks’ customer experience standards.

Laser Powder Bed Fusion (LPBF) has made it possible to produce metal parts with complex internal channels, hollow features, and other geometries that were difficult or impossible to create with conventional machining, but these parts often have internal contamination and rough surfaces that limit their usefulness. This project explored Sugino’s Cavitation Abrasive Surface Finishing (CASF) as a way to improve those rough surfaces in LPBF titanium parts and to better define where the process is most effective. The effort focused on building comparative data for multiple alloys, including titanium, Inconel, stainless steel, and aluminum, and on identifying treatment conditions that improved surface finish without reaching saturation or overtreatment. The student team also examined whether combining CASF with a workplace-safe, non-toxic chemical fluid could further remove additive manufacturing debris and smooth internal surface texture, and included initial scoping of how the process might treat blind internal passageways and cavities where abrasive flow was constrained. Together, the project aimed to establish a clearer testing and optimization basis for finishing complex additively manufactured metal features that had previously been difficult to access and improve.

This project, sponsored by T-Mobile, developed a 5G-based real-time drone video analysis system integrated with agentic LLMs. Drone video and telemetry data are transmitted via 5G to an edge node, where YOLOv12 performs object detection and Gemma3-4B generates captions. GPT-based models handle higher-level summarization and reasoning. Results, including detections and key flight/network KPIs, are stored in a database and visualized through an operator dashboard. The system supports natural language queries via an agent orchestrator and features an end-to-end, real-time, asynchronous processing pipeline.

Modern power grids face increasing load demands, forcing utilities to improve efficiency instead of pursuing costly infrastructure upgrades. This team was tasked with evaluating Dynamic Line Ratings (DLRs) as a feasible alternative. This evaluation was done by developing a tool that calculated theoretical DLR ratings through an analysis of historic weather data. These ratings were then compared to Tacoma’s existing methodology. The team also identified transmission routes that would benefit most from DLR implementation. The project found that DLR technology would increase line ratings by a meaningful margin, thereby allowing Tacoma to safely meet load demands without requiring expensive upgrades.

This project presented a cross-platform mobile application for industrial cable dimension measurement. The system integrated AI segmentation (trained on a synthetic dataset) and advanced depth sensing, utilizing a multi-sensor fusion pipeline to support both LiDAR and non-LiDAR devices. To further enhance robustness, the team developed an independent calculation framework, effectively mitigating errors from imperfect AI masks. This approach eliminated single-point failures, delivering stable results for professional quality control.

Automotive paint, e-coat, and powder coat ovens typically rely on standalone Regenerative Thermal Oxidizers to abate Volatile Organic Compounds (VOC) emissions, adding significant capital cost and factory space requirements. This project developed a conceptual integrated Regenerative Thermal Oxidizer (RTO) approach within oven heater boxes to reduce system complexity, cost, and footprint while supporting applicable NFPA, air district, and building code requirements. The concept used regenerative thermal oxidation as the basis of design, with targets for VOC destruction, thermal efficiency, cold-start readiness, and modular airflow capacity across different oven applications. The work defined the proposed system through engineering calculations, computational fluid dynamics (CFD) analysis, conceptual layouts, P&IDs, functional diagrams, a parametric 3D design, and a cost comparison against a traditional standalone abatement system.

Retrieving objects from the seafloor is typically done by human divers or remotely operated vehicles, which makes the process costly, time consuming, and dependent on specialized operators. This project aimed to develop a prototype “smart hook,” an ROV-inspired module mounted at the end of a winch line that could help automate attachment to subsea objects before they were lifted to the surface. The concept focused on cooperative objects designed for retrieval by the system, with the hook lowered near an approximate known location and then using limited horizontal motion and visual homing to search for, approach, and attach to the target. The prototype was intended to include core autonomy and control functions, along with manual control modes for testing, using a reconfigured BlueROV platform and a hardwired tether rather than wireless communication. The system was intended to support faster testing and recovery of subsea scientific instruments, with future versions potentially extending to less structured objects such as subsea debris.

The Autonomous Vent Finder project utilizes Gaussian regression, machine learning, and computer vision to locate deep-sea hydrothermal vents using an autonomous underwater vehicle. These vents discharge plumes that spread thousands of kilometers. They sustain distinctive deep-sea ecosystems and have large impacts on global ocean biogeochemistry. As the AUV traverses the plume, it collects plume-related data using onboard sensors and estimates the vent location. By training the algorithm to adapt to the changing conditions of the plume, the AUV can re-align itself to locate the vent more efficiently than traditional methods, which rely on pre-programmed paths and post-dive data analysis.

This team developed a robust machine-learning framework that corrects long-term drift in low-cost salinity sensors. Salinity data is vital for monitoring ocean health, tracking climate change, and managing sustainable aquaculture, yet high-quality equipment is often very expensive. While these low-cost sensors offer a scalable alternative compared to more expensive conductivity-temperature-depth instruments, they suffer from physical degradation like membrane hydration and ionophore leaching. By analyzing time-series data, which includes features such as voltage, impedance, and temperature, this model automatically separates true environmental variations from apparent changes caused by sensor drift.