Predictive Drill Bit Intelligence

The Intelligence Layer for the Drill Rig

Mining operations generate continuous MWD signals across every metre drilled. DrillBit AI turns that data into real-time wear predictions — so you pull the bit at exactly the right moment.

↗ Launch Dashboard API Reference
Model Performance — Live
DEPLOYED
F1 Score — DrillBit AI
0.993
+80% vs baseline
Threshold Baseline
0.551
DrillBit AI99.3%
Baseline55.1%
274K
Training Rows
80
Bit Runs
93
MWD Features
System Architecture
How It Works
Full Stack Pipeline
01
MWD Ingestion
Raw downhole signals captured in real time — weight, torque, ROP, rotary speed, standpipe pressure.
→ REST / CSV upload
02
Feature Engineering
93 features engineered from rolling statistics, lag signals, and formation-response ratios.
→ Python · Pandas
03
XGBoost Model
Gradient boosted classifier trained on 274K rows across 80 bit runs. F1 = 0.993.
→ XGBoost · Scikit-learn
04
SHAP Explainability
Every prediction explained. See which signals drove the wear call — not just that wear is imminent.
→ SHAP · Matplotlib
05
Live Dashboard
Real-time monitoring via Streamlit UI. FastAPI backend serving predictions via REST endpoint.
→ FastAPI · Streamlit
Project Context
About
Built in WA.
For the Industry.

Mining operations generate enormous amounts of data and almost none of it is used predictively. DrillBit AI is an attempt to change that — starting with one of the highest-leverage decisions in rotary drilling: when to pull the bit.

Built independently in first semester of a Masters in Mining Engineering at WASM Curtin Kalgoorlie — not for a grade, because the problem was too interesting to leave alone. Industry validation with WA partners is underway.

If you work in drilling, have access to operational MWD data, or are building in the mining technology space — get in touch.

Builder
Sanjay Marappan Somasundaram
Masters Student, Mining Engineering · WASM Curtin Kalgoorlie
Status
Live — deployed on VPS
FastAPI + Streamlit · Ubuntu 24.04
Stack
Python · XGBoost · SHAP · FastAPI · Streamlit · Pandas · Scikit-learn
Next
Validation on operational MWD data
Industry conversations active in WA
Contact
linkedin.com/in/sanjay-somasundaram