Explainable-AI_Drilling

Drilling Process Parameter Optimization and Explainable AI (XAI) for ROP Prediction

📌 Project Overview

This project investigates the relationship between drilling process parameters using two distinct datasets. The primary objectives are:

• Developing Explainable AI (XAI) models using SHapley Additive exPlanations (SHAP) with XGBoost and Random Forest.
• Predicting the Rate of Penetration (ROP) using machine learning models.
• Optimizing process parameters for ROP using various optimization techniques, including Bayesian Optimization, Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), and Simulated Annealing (SA).

📊 Datasets Used

1️⃣ Real-Time Drilling Data & Computed Petrophysical Output (CPO) Log Data

• Source: Well number 15/9-F-15 in the Volve Oil Field, North Sea.
• Features:
  • Depth: Depth of drilling
  • WOB: Weight on Bit
  • SURF_RPM: Rotation Per Minute (RPM)
  • ROP AVG: Rate of Penetration (Target Variable)
  • PHIF: Porosity
  • VSH: Volume of Shale
  • SW: Water Saturation
  • KLOGH:

2️⃣ XAI Drilling Dataset

• Source: TransAI 2023 Conference Paper
• Size: 20,000 drilling operations (rows) with 10 features, 1 binary main failure label, and 4 binary subgroup failure modes.
• Features:
  • ID: Unique identifier for each data point
  • Cutting Speed (vc): Speed at which the drill bit moves through material (m/min)
  • Spindle Speed (n): Rotational speed of the drill bit (1/min)
  • Feed (f): Depth the drill bit penetrates per revolution (mm/rev)
  • Feed Rate (vf): Speed of material feed to drill bit (mm/min)
  • Power (Pc): Power consumption (kW)
  • Cooling (%): Cooling levels (0%, 25%, 50%, 75%, 100%)
  • Material: Type of material being drilled (Steel, Cast Iron, Non-Ferrous Metal)
  • Drill Bit Type: Type of drill bit used
  • Process Time (t): Duration of drilling operation (s)
  • Main Failure: Binary indicator of major drilling failure
• Subgroup Failures:
  • Build-up Edge Failure: Material accumulation on the cutting edge
  • Compression Chips Failure: Formation of compressed chips
  • Flank Wear Failure: Wear on the drill bit flank
  • Wrong Drill Bit Failure: Incompatible drill bit usage

🛠 Methodology

1. Data Preprocessing
   • Data cleaning and handling missing values.
   • Feature engineering and scaling for better model performance.
2. Machine Learning Models for ROP Prediction
   • XGBoost and Random Forest models developed.
   • SHAP used for Explainable AI insights.
3. Optimization Techniques for Process Parameter Optimization
   • Bayesian Optimization
   • Genetic Algorithms (GA)
   • Particle Swarm Optimization (PSO)
   • Differential Evolution (DE)
   • Simulated Annealing (SA)
4. Evaluation & Validation
   • Model performance metrics (R², RMSE, MAE)
   • Comparison of optimization techniques for ROP improvement.

🚀 Installation & Setup

1. Clone the repository:

   git clone https://github.com/yourusername/drilling-optimization.git

2. Install required dependencies:

   pip install -r requirements.txt

3. Run the preprocessing and modeling scripts:

   python train_model.py

📈 Results & Findings

• Explainable AI (SHAP) provided insights into the impact of each feature on ROP.
• Bayesian Optimization and Genetic Algorithms performed best in optimizing process parameters.
• Failure mode analysis from the second dataset helped identify key factors affecting drilling efficiency.