Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/gaurav-van/optimizing-rate-of-penetration-in-geothermal-drilling-a-digital-twin-approach

Let’s explore something interesting together. In this project, we developed a machine learning digital twin using Intel-optimized XGBoost and daal4py to simulate and optimize the Rate of Penetration (ROP) in geothermal drilling. We leveraged SHAP for Explainable AI (XAI) to interpret model predictions.
https://github.com/gaurav-van/optimizing-rate-of-penetration-in-geothermal-drilling-a-digital-twin-approach

data-analysis data-science digital-twin explainable-ai geothermal geothermal-energy jupyter-notebook machine-learning python shap xai xgboost

Last synced: 7 months ago
JSON representation

Let’s explore something interesting together. In this project, we developed a machine learning digital twin using Intel-optimized XGBoost and daal4py to simulate and optimize the Rate of Penetration (ROP) in geothermal drilling. We leveraged SHAP for Explainable AI (XAI) to interpret model predictions.

Awesome Lists containing this project

README 

          
# Optimizing-Rate-of-Penetration-in-Geothermal-Drilling-A-Digital-Twin-Approach



Let’s explore something interesting together. In this project, we developed a machine learning digital twin using Intel-optimized XGBoost and daal4py to simulate and optimize the Rate of Penetration (ROP) in geothermal drilling. We leveraged SHAP for Explainable AI (XAI) to interpret model predictions, providing insights into the impact of input features. 



I would like to thank [Digital Twin Design for Renewable Energy Exploration](https://medium.com/intel-analytics-software/digital-twin-design-for-renewable-energy-exploration-2fcc4c545f8d) for providing the foundational concepts and methodologies used in this project.






## **Digital Twin Design for Renewable Energy Exploration. Building Digital Twins with XGBoost.** 



## Introduction



Digital twins are virtual representations of physical objects or systems. These representations allow us to monitor the digital twin when exposed to external/internal variables and predict how it will perform in a production environment.



### Geothermal Energy Exploration



Geothermal energy exploration involves identifying and utilizing the Earth’s natural heat as a sustainable energy resource. Geoscientists focus on locating geothermal systems that have sufficient heat to justify the investment in drilling wells and constructing power plants. These systems can be categorized into:



- **In-field systems:** Located within known geothermal areas.

- **Near-field systems:** Adjacent to known geothermal areas.

- **Deep geothermal systems:** Found at greater depths, often requiring more advanced drilling techniques.



Screenshot 2024-11-01 at 12 26 43 PM



### Drilling Challenges



Drilling for geothermal energy, especially in high-pressure and high-temperature environments, presents several challenges:



- **Specialized Tools:** The extreme conditions require tools that can withstand high temperatures and pressures. This includes drill bits, casing, and drilling fluids designed to operate effectively under these conditions.

- **Temperature Requirements:** For flash and dry steam geothermal power plants, temperatures of at least 350°F (about 177°C) are needed. Such temperatures are typically found at depths greater than 10,000 feet (approximately 3,048 meters).

- **Sensor Durability:** High temperatures and pressures can damage sensors, making it difficult to monitor drilling conditions accurately.



Screenshot 2024-11-01 at 12 27 08 PM



### Control Parameters



To optimize drilling conditions and ensure efficient and safe operations, several control parameters are monitored and adjusted:



- **Weight on Bit (WOB):** This is the downward force exerted on the drill bit. Proper WOB ensures effective penetration of the rock without damaging the bit.

- **Rotations Per Minute (RPM):** This measures the number of complete rotations made by the drill bit per minute. Adjusting RPM helps in managing the rate of penetration and the wear on the drill bit.

- **Flow Rate (FLOW):** This is the volume of liquid or slurry that moves through the drilling pipe per unit of time. Proper flow rate is crucial for cooling the drill bit, removing cuttings from the wellbore, and maintaining well pressure.

- **Torque (TOR):** This is the rotational force applied between the drill string and the rock. Managing torque is essential to prevent sticking of the drill string and to ensure smooth drilling operations.



### Building an AI Digital Twin Model



The goal of creating an AI digital twin model is to simulate the performance of geothermal wells. This involves:



- **Identifying Inputs and Outputs:** Understanding the physical system by identifying key inputs (e.g., WOB, RPM, FLOW, TOR) and outputs (e.g., rate of penetration, temperature, pressure).

- **Data Collection:** Gathering data from sensors and historical drilling operations to train the AI model.

- **Modeling and Simulation:** Using AI techniques to create a digital representation of the well, which can predict performance under various conditions and optimize drilling parameters.






## Digital Twin Concept



A **digital twin** is a virtual representation of a physical system or process. It uses real-time data and simulations to mirror the behavior and performance of its real-world counterpart. In the context of geothermal energy exploration, a digital twin can be used to model and optimize the performance of geothermal wells.



### How Digital Twins Work



1. **Data Collection**: Sensors and monitoring equipment collect data from the physical system. For geothermal wells, this includes data on WOB, RPM, FLOW, TOR, temperature, pressure, and other relevant parameters.

2. **Modeling**: This data is used to create a digital model of the well. The model incorporates the physical characteristics of the well and the surrounding geological formations.

3. **Simulation**: The digital twin can simulate various scenarios, such as changes in drilling parameters or unexpected conditions. This helps in predicting the performance of the well under different circumstances.

4. **Real-Time Monitoring**: The digital twin continuously updates with real-time data from the physical well. This allows for ongoing monitoring and optimization of drilling operations.

5. **Optimization**: By analyzing the data and simulations, the digital twin can suggest adjustments to drilling parameters to improve efficiency, reduce costs, and enhance safety.



### Benefits of Digital Twins in Geothermal Drilling



- **Predictive Maintenance**: By monitoring the condition of drilling equipment and predicting potential failures, digital twins can help in scheduling maintenance before issues arise.

- **Performance Optimization**: Digital twins can identify the optimal settings for WOB, RPM, FLOW, and TOR to maximize drilling efficiency and minimize wear on equipment.

- **Risk Management**: Simulating different scenarios helps in understanding potential risks and developing strategies to mitigate them.

- **Cost Reduction**: By optimizing drilling operations and reducing downtime, digital twins can significantly lower the overall cost of geothermal energy exploration.



### Application in Geothermal Energy



In geothermal energy exploration, digital twins are particularly valuable because they allow for:



- **Enhanced Decision-Making**: Providing a comprehensive view of the well's performance and potential issues.

- **Improved Safety**: Monitoring high-pressure, high-temperature environments in real-time to prevent accidents.

- **Sustainability**: Ensuring that geothermal wells are drilled and operated in the most efficient and environmentally friendly manner.



By leveraging digital twin technology, geoscientists and engineers can better understand and manage the complexities of geothermal drilling, leading to more successful and cost-effective energy production.






## Exploratory Data Analysis



The **Utah FORGE** (Frontier Observatory for Research in Geothermal Energy) project is a significant initiative aimed at advancing geothermal energy technologies. It serves as a dedicated underground field laboratory where various technologies related to enhanced geothermal systems (EGS) are developed, tested, and accelerated.



### The Dataset

The dataset you mentioned is from the drilling of the **Utah FORGE 58–32 MU-ESW1 well** near Roosevelt Hot Springs. This dataset is collected using an **Electronic Drilling Recorder (EDR)**, which is a system that captures and records various drilling parameters in real-time.



### Key Parameters in the Dataset

The dataset is recorded at one-foot intervals and covers depths from **85.18 feet to 7536.25 feet**. Here are some of the key parameters included:



- **Depth**: The vertical distance from the surface to the point of measurement.

- **Rate of Penetration (ROP)**: The speed at which the drill bit advances through the rock, usually measured in feet per hour.

- **Weight on Bit (WOB)**: The amount of downward force applied to the drill bit.

- **Hookload**: The total weight suspended from the hook, including the drill string and any additional equipment.

- **Temperatures In and Out**: The temperature of the drilling fluid entering and exiting the wellbore.

- **Pump Pressure**: The pressure exerted by the pumps circulating the drilling fluid.

- **Flows In and Out**: The volume of drilling fluid being pumped into and out of the well.

- **H2S**: The concentration of hydrogen sulfide gas, which is a hazardous gas that can be encountered during drilling.



### Purpose of the Data

This data is crucial for several reasons:

1. **Monitoring and Control**: Real-time monitoring of these parameters helps in controlling the drilling process, ensuring safety, and optimizing performance.

2. **Training Models**: The data can be used to train machine learning models to predict and optimize drilling performance, detect anomalies, and improve overall efficiency.

3. **Research and Development**: It supports research into new drilling technologies and methods, contributing to the advancement of geothermal energy extraction.

4. **Application in Enhanced Geothermal Systems (EGS)**: Enhanced Geothermal Systems involve creating artificial reservoirs in hot rock formations by injecting fluid to enhance permeability. The data from the EDR helps in understanding the subsurface conditions and optimizing the drilling process to create these reservoirs effectively.



### Overview of Data

![image](https://github.com/user-attachments/assets/b3af0baa-6a1e-4c48-8425-2ab932c1d997)






## Analysis and WalkThrough



### Libraries



```python

import pandas as pd 

import matplotlib.pyplot as plt

import seaborn as sns

import os

import time

import sys

import joblib

import shap

import numpy as np

import xgboost as xgb

from sklearn import linear_model

from sklearn.model_selection import GridSearchCV

from sklearn.model_selection import train_test_split

sns.set(style="darkgrid")



import warnings

warnings.filterwarnings('ignore')

```



### Utilities to handle various aspects of XGBoost and Daal4Py modeling and optimizations.



- **prepare_data**: Splits the dataset into training and testing sets.

- **linreg**: Trains and evaluates a linear regression model, optionally using Intel's optimizations.

- **XGBHyper_train**: Trains an XGBoost model with hyperparameter tuning.

- **XGBReg_train**: Trains an XGBoost model with specified parameters.

- **XGB_predict**: Makes predictions using a trained XGBoost model and calculates the mean squared error.

- **XGB_predict_daal4py**: Makes predictions using a trained XGBoost model with daal4py optimizations.

- **XGB_predict_aug**: Makes predictions using a trained XGBoost model with optional daal4py optimizations.

- **data_proc**: Processes raw data by applying a rolling mean and optional interpolation, and plots the processed data.



### Data Pre-Processing 



For digital twins, we must first define the causal relationships within the modeled system. To achieve this, we will separate the inputs and outputs of the drill rig. This leaves WOB, TOR, RPM, and FLOW as inputs and ROP as our sole output. **ROP defines the amount of downward force applied to rock during drilling.** Lastly, we will remove the first 1000ft of data to focus on areas where drilling is slow and can be improved, like zones of high pressure and temperature.



```python

geothermal_drilling_data_1000ft = geothermal_drilling_data_raw.loc[geothermal_drilling_data_raw['Depth'] > 1000, :]

geothermal_drilling_data_1000ft.head(10)

final_features = ['ROP(1 ft)', 'WOB (k-lbs)', 'Surface Torque (psi)', 'Rotary Speed (rpm)','Flow In (gpm)']

geothermal_data_final = data_proc(geothermal_drilling_data_1000ft, final_features, depth_var='Depth', dropna=True, rolling_win=10, interpolate='cubic')

```

![image](https://github.com/user-attachments/assets/e25450e9-c26b-4032-a94f-54a4458f4d6b)



### Feature Analysis 



**Pearson correlation matrix**



- **Positive Relation:** Flow in, rotary speed, and ROP show moderate positive correlations. This makes sense because fluid flow through the drill string helps spin the bit, leading to higher rotation speeds and rates of penetration. 

- **Negative Relation:** Between ROP and Depth. As depth increases, the rate of penetration (ROP) decreases due to harder rock formations, higher pressure and temperature, and equipment wear.

- **Positive Relation:** Between WOB and Depth. As depth increases, the weight on bit (WOB) increases to counteract the reduced ROP and overcome greater resistance from deeper formations.



**Pairplot Analysis**



Next, we will perform a pair plot analysis, which uses a “small multiple” approach to visualize the univariate distribution of all variables in a dataset along with all of their pairwise relationships. From the diagonal histograms, we see various complex multi-modal and skewed distributions. The kernel density maps associated with depth indicate a complex distribution of clusters for each control parameter, representing multiple rock types and the parameters used to drill through them. 



A pair plot analysis is particularly useful for visualizing these relationships, as it reveals complex, multi-modal, and skewed distributions through diagonal histograms. The kernel density maps for depth highlight clusters for each control parameter, indicating the presence of various rock types and the drilling parameters employed. 



![image](https://github.com/user-attachments/assets/92910ed0-fd64-49e4-b960-5a0963c3047d)



**Score Metric**



Since RMSE is essentially standard deviation plus the random and systemic errors of our data and model, it is a good goal to achieve an **RMSE far below our ROP standard deviation (23.36 feet/hour).** `geothermal_data_final.describe()`






## Model Building - Building our Digital Twin — daal4py and XGBoost Optimized Training and Inference



**Configurations**



```python

i_flag = 1

model_file = 'test0.pkl'



drilling_data = geothermal_data_final

input_cols = ['WOB (k-lbs)_P', 'Surface Torque (psi)_P', 'Rotary Speed (rpm)_P','Flow In (gpm)_P']

output_col = 'ROP(1 ft)_P'

```

### 1. Linear Regression RMSE Score benchmark

We will start by training a linear regression model and evaluating it to establish an RMSE reference point and ensure that our XGBoost model will outperform a simple linear regressor.



```python

x_train, x_test, y_train, y_test = prepare_data(

    drilling_data, input_cols_list=input_cols, output_var=output_col)



print("===== Running Benchmarks for Linear Regression =====")

train_time, pred_time, MSE = linreg(

    x_train, x_test, y_train, y_test, i_flag)

print("Training time = ", train_time)

print("Prediction time = ", pred_time)

print('MSE:', str(round(np.mean(MSE), 3)))

print('Root MSE:', str(np.sqrt(round(np.mean(MSE), 3))))

```

Screenshot 2024-11-01 at 12 53 25 PM



### 2. XGBoost Daal4Py Model 



Using XGBoost and automated hyperparameter optimization through GridSearchCV and xgb methods (Figure 12). The XGB_prediction_daal4py custom function from the digital-twin AI Reference Kit is utilized to convert the XGBoost model to daal4py.



```python

x_train, x_test, y_train, y_test = prepare_data(

    drilling_data, input_cols_list=input_cols, output_var=output_col, shuffle=False)

loop_ctr = 5

parameters = {'nthread': [1],

              'learning_rate': [0.02],  # so called `eta` value

              'max_depth': [3, 5],

              'min_child_weight': [6, 7],

              'n_estimators': [750, 1000],

              'tree_method': ['hist']}

print(

    "===== Running Benchmarks for XGB Hyperparameter Training =====")

train_time, trained_model, model_params = XGBHyper_train(

    x_train, y_train, parameters)

print("Training time = ", train_time)

if i_flag:

    prediction, pred_time, MSE = XGB_predict(

        x_test, y_test, trained_model, loop_ctr, i_flag)

    prediction, pred_time_daal4py, MSE_daal4py = XGB_predict_daal4py(

        x_test, y_test, trained_model, loop_ctr, i_flag)

    print("Prediction time = ", pred_time)

    print("daal4py Prediction time = ", pred_time_daal4py)

    print('Root MSE: ', str(np.sqrt(round(np.mean(MSE), 3))))

    print('daal4py Root MSE: ',

                str(np.sqrt(round(np.mean(MSE_daal4py), 3))))

else:

    prediction, pred_time, MSE = XGB_predict(

        x_test, y_test, trained_model, loop_ctr, i_flag)

    print("Prediction time = ", pred_time)

    print('Root MSE: ', str(np.sqrt(round(np.mean(MSE), 3))))



joblib.dump(trained_model, "model_xgbhpo.pkl")

```

Screenshot 2024-11-01 at 12 54 32 PM






## Explainable AI (XAI) and SHAP



### Introduction to Explainable AI (XAI)

Explainable AI (XAI) refers to methods and techniques that make the behavior and predictions of machine learning models understandable to humans. The goal of XAI is to provide transparency, interpretability, and trust in AI systems by explaining how models make decisions. This is crucial for ensuring accountability, fairness, and compliance with regulations, especially in critical applications like healthcare, finance, and autonomous systems.



### SHAP (SHapley Additive exPlanations)

SHAP is a popular XAI method that uses cooperative game theory to explain the output of machine learning models. The core idea is to allocate credit for a model’s prediction among its input features by considering the impact of each feature when it is present, missing, or combined with other features. SHAP values provide a unified measure of feature importance, making it easier to understand and compare the contributions of different features.



#### Key Concepts:

- **Shapley Values**: Originating from cooperative game theory, Shapley values fairly distribute the "payout" (model prediction) among the "players" (input features) based on their contributions.

- **Feature Impact**: SHAP values indicate how much each feature contributes to the difference between the actual prediction and the average prediction.



### Waterfall Plot

The SHAP waterfall plot is a visualization that shows how each feature contributes to moving the model output from the base value (average model output) to the final prediction. It provides a detailed breakdown of the impact of each feature on a single prediction.



#### How to Interpret:

- **Base Value**: The average model output over the training dataset.

- **Feature Contributions**: Features that push the prediction higher are shown in red, while those that push it lower are shown in blue.

- **Final Prediction**: The sum of the base value and all feature contributions.



### SHAP Heatmap

The SHAP heatmap provides a high-level overview of the impact of all features across multiple predictions. Unlike the waterfall plot, which focuses on individual predictions, the heatmap shows the collective influence of features on the model’s outputs.



#### How to Interpret:

- **Color Intensity**: Indicates the magnitude of the SHAP values, with brighter colors representing higher impact.

- **Feature Importance**: Helps identify which features consistently have a strong influence on the model’s predictions.

- **Patterns and Clusters**: Reveals patterns and clusters in the data, indicating how different features interact and affect the model’s behavior.



```python

explainer = shap.TreeExplainer(trained_model)

shap_values = explainer(x_test)

```



```python

shap.plots.waterfall(shap_values[1])

```



When ROP = 16.757 feet/hour, this is the impact of each feature:

- Flow In (GPM) brings the model down from the base value by -5.13 ROP units.

- WOB (k-lbs) brings the model down from the base value by -3.13 ROP units.

- Surface Torque (psi) brings the model down from the base value by -0.77 ROP units.

- Rotary Speed (rpm) brings the model up from the base value by +0.43 ROP units.



![image](https://github.com/user-attachments/assets/e4d60e8f-8103-4466-a4d1-e3b291ddcd6c)



```python

shap.plots.waterfall(shap_values[10])

```



When ROP = 46.66 feet/hour, this is the impact of each feature:

- Flow In (GPM) brings the model up from the base value by +11.59 ROP units.

- WOB (k-lbs) brings the model up from the base value by +2.98 ROP units

- Surface Torque (psi) brings the model up from the base value by +4.15 ROP units.

- Rotary Speed (rpm) brings the model up from the base value by +2.57 ROP units



![image](https://github.com/user-attachments/assets/567dfdde-1689-476e-8970-a5c56bb8b4d7)



**SHAP Heatmap**



The SHAP Heatmap serves a similar purpose as the waterfall plot. However, instead of looking at the impact sample by sample, it shows the high-level impact of all features across our model’s outputs. Again, this visualization can help us understand how different features impact our model. In the case of the sharp increase in ROP around instance (sample) 1750, we see very high SHAP values for both Flow In and WOB, indicating that these two features are primarily responsible for the resulting increase in ROP.



```python

shap.plots.heatmap(shap_values[0:])

```



![image](https://github.com/user-attachments/assets/1da3d04d-6106-410e-a8a9-347d11da6d2b)



- **Feature Impact Varies with ROP:** The influence of features like Flow In, WOB, Surface Torque, and Rotary Speed changes with different ROP levels. At lower ROP, these features tend to have a negative or smaller positive impact, while at higher ROP, their impact becomes more positive and significant.



- **Optimizing Drilling Parameters:** By understanding these impacts, drilling operations can be optimized. For example, adjusting Flow In and WOB based on the current ROP can lead to more efficient drilling.



- **Data-Driven Decisions:** These insights enable data-driven adjustments to drilling parameters, improving overall efficiency and effectiveness.






## Conclusion 



In this project, we leveraged Intel-optimized XGBoost and daal4py libraries to develop a machine learning digital twin for simulating Rate of Penetration (ROP) in drilling operations. This digital twin, represented by our predictive model, provides valuable insights by simulating ROP based on various input features. To enhance trust and interpretability, we integrated an Explainable AI (XAI) workflow using SHAP. This allowed us to clearly understand the impact of each input feature on the model’s predictions, ensuring transparency and confidence in our digital twin’s outputs. Through this approach, we demonstrated the effectiveness of combining advanced machine learning techniques with XAI to optimize and understand complex drilling processes.