Thesis work in the area of machine learning on drilling data:

Bit Wear Detection Model

Supervisors: Epiroc and Johannes Andreas Stork (Örebro University)

Goal:

Develop and evaluate machine learning models that can estimate the health state of drill bits based on sensor data collected from multiple drill rigs. The objective is to provide a “bit health score” that helps operators decide when a bit should be replaced — improving efficiency and reducing downtime.

Background:

Modern drilling rigs collect large amounts of time-series data from various sensors. These signals contain valuable information about equipment performance and wear. By analyzing this data with machine learning, it is possible to predict tool degradation before failure occurs — a key concept in predictive maintenance.

Thesis Tasks:

The student will:

• Analyze sensor time-series data from drill rigs.
• Investigate feature extraction techniques (e.g., mean, standard deviation, FFT coefficients, and other frequency-domain features).
• Use these features as input to lightweight ML models such as Logistic Regression, Random Forests, Decision Trees, Naive Bayes, or Support Vector Machines.
• Compare the performance of different model architectures in classifying drill bit health.
• Assess model interpretability and discuss practical deployment aspects.

Expected Outcome:

A comprehensive analysis identifying which model type and feature set best predict drill bit health, along with recommendations for real-time or on-rig implementation.

Student Profile:

Interest in machine learning, data analysis, and industrial applications.
Basic programming skills in Python (pandas, scikit-learn, etc.).
Curiosity about data-driven maintenance strategies and sensor data interpretation.

For more information, please contact Maria Krantz (Epiroc, maria.krantz@epiroc.com) and Johannes Andreas Stork (Örebro University, johannesandreas.stork@oru.se)