Problem
Small hydro turbines are frequently retrofitted into existing mill and dam infrastructure, where operating conditions (head, flow, load) vary seasonally. Tuning the governor and voltage regulator control loops for these plants is traditionally a slow, manual process performed by specialist engineers against manufacturer guidelines — expensive, hard to schedule, and rarely revisited once the plant is running.
L&S Electric sponsored this capstone to explore whether an intelligent control layer could automate that tuning: observing turbine-generator dynamics and converging on control settings that maximize stability and response quality in real time.
Approach
- System identification: characterize the turbine-generator response from sensor data to build a usable dynamic model of the plant.
- ML-driven optimization: apply machine learning and optimization techniques to propose and evaluate controller gains, replacing manual trial-and-error with an automated search over the tuning parameter space.
- Real-time execution: run the tuning loop on embedded hardware so the system can adapt online rather than requiring an offline engineering pass.
- Safety envelope: constrain every proposed setting within validated bounds so automated tuning can never drive the generator outside stable operating regions.
Outcome
The team delivered a working control system prototype demonstrating automated tuning against L&S's use case, presented to the sponsor at capstone design and implementation reviews. The project gave me end-to-end ownership across the stack I care most about — sensor data acquisition, model training, and deployment on constrained embedded targets — under a real industrial customer with real acceptance criteria.
Tech
Source is maintained on GitLab for MSOE capstone delivery; contact me if you'd like a walkthrough of the implementation.