Process Optimization · Applied Machine Learning · Data-Driven Modeling
- Process Optimization & Modeling: Developing physics informed and data-driven methodologies to optimize complex industrial and manufacturing systems.
- Bayesian Methods & Uncertainty: Active learning, Gaussian processes, and distribution free conformal prediction for robust decision-making under uncertainty.
- Applied Machine Learning & Control: Reinforcement learning, time-series forecasting, and computer vision for autonomous control and in situ monitoring.
| Domain | Tools & Technologies |
|---|---|
| Programming Languages | Python, MATLAB, R |
| Machine Learning & AI | PyTorch, Scikit-Learn, Conformal Prediction, Time-Series Analysis, Random Forest, Neural Networks |
| Data Science & Statistics | Pandas, NumPy, Bayesian Statistics, Design of Experiments (DoE), Data Visualization |
| Engineering & Simulation | Simulink, CFD Simulation Analysis, OpenCV (Computer Vision), IoT Sensor Processing |
| Project | Domain & Stack | Highlights & Outcomes |
|---|---|---|
| 🔬 Thermal Monitoring for AM | Python · OpenCV · Scikit-Learn · Conformal Prediction | In situ infrared video monitoring during FFF 3D printing. Implements Conformal Prediction for anomaly detection, ROI PCA modeling ( |
| 🌌 SLM Spatter & Spatial Statistics | R · Python · OpenCV · Functional Data Analysis | In situ video image analysis of spattering dynamics in Selective Laser Melting (SLM). Combines computer vision segmentation, Ripley's K-function, MANOVA, PCA/QDA quality classification, and Functional Data Analysis (FPCA). |
| 🦾 Compliant Robot Arm Control | Python · PyTorch · Scikit-Optimize · DDPG | Coupled 2-DoF mass spring damper arm exploiting dynamic resonance. Combines Bayesian Optimization (chirp sweep) with DDPG and energy based reward shaping to reach out of reach targets. |
| 📦 Parcel Singulation via RL | MATLAB · Reinforcement Learning Toolbox · Simulink | Decentralized PPO agent on a 5×5 Actuator Matrix System (AMS). Employs multi-parcel curriculum learning (2 to 10 parcels) and inertia transfer learning to singulate parcels at 100% accuracy. |