What a Student ML Challenge Taught Me About the Future of Geothermal and the Imaging Industry—-Jane Onu

By Tosin Clegg

On March 21, 2025, my teammates and I placed second in the SPE–Fervo Machine Learning Challenge, an international competition that asked participants to predict Bottom-Hole Circulating Temperature (BHCT) from authentic drilling-log data. The event ran alongside the SPE Gulf Coast Section (GCS) Data Science Convention at the ExxonMobil campus in Spring, Texas bringing students, engineers, and data scientists into the same room to solve a real operational problem in geothermal drilling.
Why BHCT and why it matters beyond drilling
BHCT is the temperature at the drill bit while fluid is circulating. Get that number wrong, and you risk overheating tools, damaging cement jobs, shortening the life of the bottom-hole assembly, or creating delays that cost real money. Get it right, and you can proactively manage mud programs, protect temperature-sensitive electronics, and keep operations within safe envelopes. In other words, BHCT is not just a number; it’s a control knob for safety and efficiency and the literature backs that up.
The challenge had two phases: build a generalized model that works across wells, then adapt it to new wells mirroring the reality of taking a solution from one pad or basin to another. We engineered features that captured the physics we know drive BHCT (circulation time, depth, flow rate, mud properties, surface vs. downhole temperature indicators) and emphasized robustness over leaderboard luck.
Geothermal is finally having its moment. But drilling hotter, deeper, and faster strains traditional workflows. Imaging and sensing technologies—DTS/DAS cables, temperature logs, and thermal management models are expanding what we can “see.”
At the same time, ML prediction of BHCT is maturing from research to field-ready capability, with studies demonstrating real-time forecasting that can extend tool life, cut non-productive time, and improve safety margins. Put together, this is a powerful flywheel: better data → better models → smarter operations → better data again.