Projects
Each project represents a unique challenge that I have tackled, and I am excited to share the stories behind them with you.
Smart Semiconductor Manufacturing
MFG 598
• Architected and deployed a complete 5-layer IIOT system for real-time semiconductor wafer defect detection, using wafer dataset as device layer, integrating MQTT messaging, edge-based ML inference and PostgreSQL database as cloud layer.
• Created streamlit-based interactive dashboard acting as application layer, with real-time process parameter visualization, defect prediction, alert acknowledgment with multi-line production status resulting in enhanced operator decision-making.
• Achieved 94% wafer defect prediction accuracy with 70% recall value by training and deploying an edge-based XGBoost model on historical wafer data with <50ms inference latency.
Project Link:
Sensor to Dashboard
MFG 598
-
Architected complete IIOT data pipeline integrating Raspberry Pi 4 with Sense HAT sensors, PostgreSQL database, and Streamlit dashboard for real-time sensor monitoring.
-
Developed Python-based script reading temperature, humidity, pressure, and gyroscope data every 2 seconds from Sense-Hat with automated data insertion to PostgreSQL with timestamp.
-
Built interactive Streamlit visualization dashboard featuring real-time gauges, time-series graphs, and 3D orientation displays for data-driven decision making.
Project Link:
ML based Generative Modeling of Airfoil Geometries
MAE 551
• Replaced computationally expensive airfoil CFD simulations with ML-based generative model using autoencoder and inverse neural network in PyTorch, reducing airfoil design iteration time from hours to seconds.
• Processed and standardized 200 airfoil simulations from AirfRANS dataset using CST parameterization, training autoencoder to compress 18D geometry space to 6D latent representation with 0.018 validation MSE loss.
• Successfully generated airfoil designs from target aerodynamic parameters (drag/lift coefficients, pressure) by mapping 1007-dimensional performance space to 6D latent codes using inverse neural network with 0.3106 validation MSE loss.
Project Link:
Embedded Temperature Control System with Disturbance Rejection
RAS 550
• Developed real-time feedforward + PID temperature control system in Simulink for TCLab heater with 5V dc fan acting as measurable disturbance, deployed to Arduino via automatic code generation with 0.5s sampling rate.
• Performed open loop system identification with 40% heater power and developed first-order thermal model to characterize heater dynamics and calculated feedforward gain from measured cooling effect by fan.
• Achieved 50% reduction in disturbance using feedforward + PID as compared to baseline PID control.
Project Link:
Autonomous Line Following Drone
RAS 546
• Modeled an efficient control architecture in Simulink, using RGB color space conversion to create binary image for enhanced detection of colored line. Implemented custom submatrix blocks to analyze specific pixel regions of binary image coming from the drone’s camera for navigation decisions (left, right, forward).
• Implemented blob analysis block for circle detection, and used the area of the circle as a threshold parameter for landing signal. Incorporated a state flow machine with five different states for path planning.
• Achieved autonomous blue line tracking and center-circle landing with an accuracy of ±2 mm using limited computational resources on the mambo drone platform, demonstrating robust performance in real-world flight conditions.
Project Link:
3 DOF Robotic Arm with Kinematic Modeling
MAE 547
• Collaborated with a team of five members to design a low cost, programmable, 3 DOF robotic arm to perform autonomous pick-and-place operation of a small cubical object.
• Designed the links and gripper parts in SOLIDWORKS, 3-D printed using PLA and PETG materials, and integrated MG5465 servo motors and electro-mechanical control interfaces for precision joint actuation.
• Developed a robust inverse kinematic algorithm in Python that actuated the servo motors using PWM signals from a Raspberry Pi 5 microcontroller to execute the task with 98% accuracy and repeatability.
Project Link:
State-Space Modeling and Control of a Levitating Train
MAE 506
-
Developed state-space control system for magnetic levitation train dynamics by deriving nonlinear equations of motion from first principles, linearizing about equilibrium using Taylor expansion, and proving full controllability/observability.
-
Stabilized inherently unstable maglev system by designing PD state feedback controller that transformed marginally stable dynamics (pure imaginary eigenvalues) into asymptotically stable response meeting strict performance specs: 4% overshoot, 0.24s settling time.
-
Validated controller performance through MATLAB simulation of 2250kg train with 40mm levitation gap, demonstrating elimination of sustained oscillations and achievement of precise position control using optimally tuned gains.
Project Link:
FEA Based Thermal Modeling of 3D Printing Process
MAE 545
-
Developed FEA thermal model in ANSYS to simulate fused deposition modeling (FDM) process for ABS and PC materials, implementing element birth-and-death technique to replicate sequential layer-by-layer deposition
-
Analyzed thermal behavior through steady-state and transient simulations, revealing PC reached 280°C nozzle temperature with 8.26 W/mm² peak heat flux versus 235°C and 6.56 W/mm² for ABS, explaining differences in part quality and warping behavior.
-
Identified critical process insights through transient analysis demonstrating ABS requires tighter parameter control due to faster, more variable cooling rates while PC shows better interlayer bonding.
Project Link:
FEA Code Development and Validation for Perforated Titanium Plate
MAE 503
-
Developed custom 2D finite element code in MATLAB to analyze stress concentration in titanium plate with central hole, implementing T3, Q4, T6, and Q8 element types with Gaussian quadrature integration and least-squares stress projection achieving < 5% error.
-
Evaluated computational efficiency vs. accuracy trade-offs for linear (T3, Q4) and quadratic (T6, Q8) elements across mesh refinements, proving higher-order elements capture curved geometry and stress gradients with significantly fewer degrees of freedom.
-
Performed 3D validation study comparing plane stress assumptions with full 3D solid analysis, showing plane stress introduces 15% error in out-of-plane components near geometric discontinuities, establishing guidelines for when 3D analysis is required.
Project Link:
