Classification of Iron Ionization States Using Ensemble Learning on NIST Spectral Data
Summary A machine learning-based classification study designed to distinguish between Singly Ionized (Fe II) and Doubly Ionized (Fe III) states of Iron atoms. By implementing an automated web scraping pipeline on the NIST Atomic Spectra Database and engineering physics-based features, this project compares probabilistic and ensemble learning models, achieving near-perfect classification accuracy with Random Forest.
Description This project addresses the challenge of automating the identification of atomic spectral lines in large-scale physics databases. Manually classifying spectral signatures is prone to error and time-consuming. To solve this, I developed a robust data mining and classification framework:
- Automated Data Pipeline: Built a dynamic web scraper using Python to extract raw spectral data from the NIST server. Implemented complex Regex (Regular Expression) cleaning to handle measurement uncertainties and noise in raw wavelength/intensity data.
- Physics-Informed Feature Engineering: Beyond standard features, I derived a new physical attribute, "Energy Difference (ΔE)," representing the gap between upper and lower energy levels, which significantly improved model separability.
- Comparative Model Analysis: Evaluated three distinct algorithms: Naive Bayes (Baseline), Support Vector Machines (Kernel-based), and Random Forest (Ensemble). Validated results using 10-Fold Cross-Validation to ensure generalization.
- Results: The Random Forest model outperformed others, achieving a 99.16% Accuracy and 99.15% F1-Score, successfully distinguishing complex spectral patterns where the baseline model struggled.
Tech Stack
- Languages: Python 3.x
- Machine Learning: Scikit-learn (Sklearn)
- Techniques: Random Forest, SVM (RBF Kernel), Naive Bayes, K-Fold Cross Validation, Web Scraping, Feature Engineering
- Data Science: Pandas, NumPy
- Visualization: Matplotlib, Seaborn, Heatmaps
Remaining Useful Life Prediction of Turbofan Engines Using LSTM Networks with Piecewise Linear Target Labeling
Summary
A deep learning-based predictive maintenance framework designed to estimate the breakdown time of aircraft engines. By utilizing Long Short-Term Memory (LSTM) networks and a Piecewise Linear Target Labeling strategy on NASA's C-MAPSS dataset, this model overcomes the limitations of traditional regression approaches, achieving high accuracy in predicting engine failures.
Description
This project addresses the critical challenge of Prognostics and Health Management (PHM) in the aviation sector2. Standard regression models often fail to predict the "healthy" stage of an engine, introducing noise into the training process. To solve this, I developed a robust deep learning solution:
- Piecewise Linear Strategy: Implemented a target labeling technique (clipping RUL at 125 cycles) to stabilize the learning process during the engine's healthy operation phase, significantly reducing early-stage prediction errors3333.
- Deep LSTM Architecture: Designed a stacked LSTM network (128 & 64 units) with Dropout regularization to capture long-term temporal dependencies in complex multivariate time-series data4.
- Data Processing: Utilized a Sliding Window approach (50 time-steps) and Min-Max normalization to process data from 14 distinct sensors (pressure, temperature, speed)5555.
- Results: The model achieved a Root Mean Squared Error (RMSE) of 14.03 and an R2 score of 0.88 on the FD001 test set, demonstrating superior performance over baseline models6.
Tech Stack
- Languages: Python 3.8
- Deep Learning: Keras, TensorFlow
- Techniques: LSTM (Long Short-Term Memory), Time-Series Analysis, Sliding Window, Regression
- Data Science: Pandas, NumPy, Scikit-learn
- Visualization: Matplotlib, Seaborn
Link
DataDiet Daily AI Newsletter
Summary A minimalist, automated newsletter aggregator designed to "cut the noise" and deliver essential tech developments directly to your inbox.
Description In an era of information overload, staying updated on technology can feel overwhelming. DataDiet was built to solve this problem by focusing on signal over noise. It is an automated system that curates the most critical developments across five key sectors: AI, Space, Gaming, Mobile, and Gear.
The project prioritizes a "Clean UI" philosophy. I intentionally removed timestamps to eliminate FOMO (Fear Of Missing Out) and utilized distinct color-coded categories for rapid visual scanning. The goal is to provide a digestible, focused briefing on what truly matters in tech, accessible via a clean web interface or delivered as an email newsletter.
Tech Stack React / Next.js, Tailwind CSS, Python (BeautifulSoup, requests, feedparser), Vercel Deployment.
Link Live Project
KUKA Robot Dynamics Analysis
Summary: A comprehensive mathematical analysis of the motion dynamics, forces, and torques for KUKA industrial robots.
Description: This project focuses on the dynamic modeling of industrial robotic arms. By analyzing the physical properties and motion constraints of KUKA robots, I calculated the necessary forces and torques required for specific trajectory executions. The study bridges theoretical robotics (Lagrangian and Newton-Euler formulations) with practical engineering requirements, providing a foundation for efficient motion control and motor sizing in industrial automation.
Tech Stack: Python, Robotics Dynamics, Physics Modeling, MATLAB/NumPy
Link: View on GitHub
6-DOF Robot Arm Forward Kinematics Simulator
Summary: A Python-based simulation tool designed to visualize the Forward Kinematics of a 6-Degrees-of-Freedom (6-DOF) robot arm.
Description: I developed a custom simulator to visualize the movement and positioning of a 6-axis robotic arm. The software utilizes Forward Kinematics (FK) algorithms to calculate the precise position and orientation of the end-effector based on given joint angles. This tool serves as a verification platform for motion planning algorithms, allowing for the simulation of complex robotic tasks in a virtual environment before physical deployment. It demonstrates strong proficiency in spatial transformation matrices and algorithm implementation.
Tech Stack: Python, Forward Kinematics, Simulation, 3D Visualization, Linear Algebra
Link: View Simulator
Real-Time Sensor Monitoring & ML-Based Fault Detection
Summary: An IoT-integrated system designed to detect operational anomalies in industrial machinery using Machine Learning.
Description: Designed to enhance industrial safety and efficiency, this system monitors sensor data in real-time to identify potential failures before they occur. I implemented Machine Learning algorithms (Classification & Anomaly Detection) to analyze data streams from sensors. The system successfully differentiates between normal operating conditions and fault states, enabling predictive maintenance strategies. This project demonstrates the practical application of AI in Industry 4.0 contexts.
Tech Stack: Machine Learning, Python, IoT, Data Analysis, Sensor Fusion
Links: View on GitHub
Predictive Modeling of Airfoil Noise Using Machine Learning
Summary: A regression-based ML study to predict aerodynamic noise levels generated by airfoils for quieter aviation technology.
Description: In this study, I utilized the NASA Airfoil Self-Noise dataset to build predictive models for aerodynamic noise. By applying various regression techniques, including Linear Regression and Random Forests, I analyzed the relationship between frequency, angle of attack, chord length, and sound pressure levels. The project highlights the use of data science to solve complex engineering problems in aerodynamics and acoustics.
Tech Stack: Python, Scikit-learn, Pandas, Data Visualization, Regression Analysis
Links: View on GitHub
End-to-End Autonomous Driving: Lane Detection & Steering Control
Summary: A deep learning-based system for autonomous vehicle navigation, utilizing computer vision for lane tracking and real-time steering angle prediction.
Description: This project addresses the core challenges of autonomous driving, specifically lane keeping and trajectory following. Using a Convolutional Neural Network (CNN) architecture inspired by NVIDIA's end-to-end learning model, the system processes raw camera input to map visual data directly to steering commands. The implementation integrates advanced image preprocessing, data augmentation, and regression modeling to enable the vehicle to autonomously navigate and stabilize itself within a simulation environment.
Tech Stack: Python, OpenCV, TensorFlow/Keras, Deep Learning (CNN), Computer Vision, NumPy.
High-Altitude Model Rocket Design & Simulation
Summary: A G-class motor model rocket project designed in OpenRocket, optimized for aerodynamic stability (1.3 cal) and capable of reaching an altitude of 987 meters.
Description: This project involves the engineering and simulation of a custom model rocket based on aerodynamic principles and flight mechanics. Using OpenRocket software for computational analysis, the rocket's Center of Pressure (CP: 65.3 cm) and Center of Gravity (CG: 58.1 cm) were optimized to ensure a safe static stability margin of 1.3 cal.
The design features a Haack series fiberglass nose cone to minimize drag and trapezoidal plywood fins for flight stability. Simulation data indicates that, powered by a G73-P motor with 142 Ns total impulse , the rocket achieves a rail exit velocity of 15.3 m/s , a maximum velocity of 204 m/s , and an apogee of 987 meters. The system is designed for a safe parachute recovery with a landing velocity of 7.01 m/s after a total flight time of 151 seconds.
Tech Stack: OpenRocket, Aerodynamics, Flight Mechanics, Physics Simulation, CAD/Technical Drawing, Avionics Integration.