Advanced Projects
Research-grade projects for grad students and ambitious undergrads. These involve cutting-edge techniques like PINNs, reinforcement learning, and multi-sensor fusion.
32 projects
Combine with other filters →Domains at this level
Physics-Informed Neural Net for Aeroelasticity
Train a neural network that respects the laws of physics
Build a physics-informed neural network (PINN) using DeepXDE to solve aeroelastic flutter equations. Combine deep learning with structural dynamics to predict wing flutter boundaries.
Reinforcement Learning for Spacecraft Docking
Train an AI agent to autonomously dock with a space station
Build a reinforcement learning environment for spacecraft proximity operations and train an agent to perform autonomous docking. Uses OpenAI Gymnasium and Stable-Baselines3.
Digital Twin of a Jet Engine Subsystem
Build a living simulation that mirrors real engine behavior
Create a digital twin of a jet engine compressor using MATLAB/Simulink and Python. Simulate performance, inject sensor data, and detect anomalies — the same approach used by GE and Rolls-Royce.
Multi-Sensor Fusion for GPS-Denied Drone Navigation
Navigate a drone without GPS using cameras, IMU, and clever math
Build a sensor fusion system that lets a drone navigate without GPS. Combine visual odometry (OpenCV), inertial measurement (IMU), and optical flow to estimate position in GPS-denied environments.
Satellite Telemetry Anomaly Detection
Catch satellite failures before they happen using unsupervised ML
Apply unsupervised clustering algorithms to real satellite telemetry datasets to detect anomalous behavior patterns before they escalate into failures. Combines signal processing, feature engineering, and scikit-learn to build a production-ready anomaly detection pipeline.
ML Surrogate Model for XFLR5
Replace a CFD solver with a neural network that runs 1000× faster
Generate thousands of XFLR5 airfoil simulations across a parametric design space, then train a neural network surrogate model that predicts aerodynamic coefficients in milliseconds. Enables real-time design optimization that would take days with the full solver.
Vision-Based Autonomous Landing with ArduPilot
Build a drone that finds and lands on a moving platform using only a camera
Develop a complete computer vision pipeline for precision autonomous landing on a moving platform. The system uses OpenCV for target detection and tracking, feeds corrections to ArduPilot via MAVLink, and achieves landing accuracy well beyond what GPS alone can provide.
Neural Network Flight Controller
Replace a PID controller with a neural network trained on flight data
Train a neural network in TensorFlow to replace a classical PID flight controller for a fixed-wing aircraft. The network learns the control mapping from flight dynamics data and is evaluated on stability, disturbance rejection, and robustness compared to a tuned PID baseline.
Differentiable Airfoil Optimization with JAX
Use automatic differentiation to find the optimal airfoil shape via gradient descent
Implement a differentiable panel-method aerodynamics solver in JAX and use automatic differentiation to compute exact gradients of lift and drag with respect to airfoil shape parameters. Perform gradient-based optimization to find airfoils that maximize L/D for given flight conditions.
Fourier Neural Operator for Airfoil Flows
Train a neural operator that predicts full CFD flow fields in milliseconds
Train a Fourier Neural Operator (FNO) in NVIDIA PhysicsNeMo to learn the solution operator for 2D incompressible flow around arbitrary airfoil shapes. The trained FNO predicts complete pressure and velocity fields 1000× faster than OpenFOAM while maintaining engineering accuracy.
ML-Optimized Satellite Constellation Design
Use Bayesian optimization to find constellations that maximize coverage with fewer satellites
Combine Ansys STK's high-fidelity orbital analysis with Python-based Bayesian optimization to explore the satellite constellation design space. Automatically discover configurations that achieve target coverage and revisit time requirements with the minimum number of satellites.
Generative Design Automation with SolidWorks API
Write a Python script that designs, evaluates, and selects structural brackets autonomously
Build a Python automation system that drives the SolidWorks API to programmatically generate structural bracket design variants, run FEA evaluations, extract performance metrics, and select the optimal design — compressing days of manual CAD work into an automated pipeline.
MISRA-Compliant ADCS Software in C++
Write flight-grade attitude control code that meets aerospace safety standards
Implement a complete CubeSat attitude determination and control system (ADCS) in MISRA-compliant C++. Covers quaternion-based attitude determination, reaction wheel control laws, and the software development practices — unit testing, static analysis, and code coverage — required for flight-grade embedded software.
Physics-Informed Neural Nets for Aerodynamics in Julia
Solve the Navier-Stokes equations with a neural network using Julia's NeuralPDE.jl
Use Julia's NeuralPDE.jl to implement physics-informed neural networks (PINNs) that solve the incompressible Navier-Stokes equations for 2D flow over a circular cylinder. Explore the benchmark Von Kármán vortex street problem and compare PINN solutions against reference DNS data.
Multi-Source Predictive Maintenance Pipeline
Combine multiple aerospace datasets to build a model that generalizes across equipment types
Engineer a robust predictive maintenance system by combining multiple Kaggle aerospace and industrial datasets, training models that transfer across equipment types, and deploying a unified health monitoring pipeline. Goes beyond single-dataset ML to tackle real-world generalization challenges.
ML-Enhanced Turbulence Modeling in OpenFOAM
Train a neural network to correct RANS turbulence model errors using LES reference data
Generate high-fidelity LES (Large Eddy Simulation) data in OpenFOAM and train a neural network to correct the Reynolds-Averaged Navier-Stokes (RANS) turbulence model predictions. Implements data-driven turbulence modeling, one of the most active research areas in computational fluid dynamics.
ML-Driven Generative Design Pipeline
Combine Fusion 360 generative design with ML to automate optimal design selection
Build an end-to-end pipeline that runs Fusion 360's generative design engine, exports the resulting design variants, evaluates them with ML-based scoring models, and automatically selects the Pareto-optimal designs — turning a manual design review into a fully automated optimization workflow.
ML Surrogate for Full Aircraft Analysis
Generate thousands of aircraft configurations in OpenVSP and train a neural network for rapid preliminary design
Script OpenVSP to generate thousands of full aircraft configurations spanning the preliminary design space, run vortex lattice aerodynamic analysis on each, and train a neural network surrogate that predicts aerodynamic performance in milliseconds — enabling real-time MDO during concept exploration.
Edge-Deployed Satellite Detection with YOLO
Train, quantize, and deploy a YOLO model for real-time satellite detection on Jetson Nano
Train a YOLO object detection model on satellite image datasets, apply TensorRT quantization to reduce model size and inference time, and deploy on a Jetson Nano edge device for real-time detection. Demonstrates the full ML deployment pipeline from training to embedded hardware.
Synthetic Training Data Generation in Omniverse
Generate photorealistic synthetic datasets for aerospace computer vision using NVIDIA Omniverse
Use NVIDIA Omniverse Replicator to create a photorealistic synthetic data generation pipeline for aerospace computer vision tasks. Randomize lighting, materials, backgrounds, and object poses to generate annotated training data that is cheaper, faster, and more diverse than real-world data collection.
Turbine Blade Multiphysics Analysis
Perform coupled thermal-structural-vibration analysis of a turbine blade under real operating conditions
Conduct a complete multiphysics analysis of a gas turbine blade in Siemens Simcenter: steady-state thermal analysis under combustion gas and cooling air conditions, one-way coupled structural analysis for thermal stress, and constrained modal analysis for vibration frequencies under centrifugal load — matching the workflow used in engine certification.
Hybrid Physics-ML Digital Twin
Combine a physics-based reduced-order model with ML correction for real-time structural health monitoring
Build a hybrid digital twin in Ansys Twin Builder that combines a physics-based Reduced Order Model (ROM) of a structural component with a machine learning correction layer. The hybrid twin runs in real time, processes incoming sensor data, and provides health state estimates with uncertainty bounds for structural health monitoring.
Graph Neural Network for Microstructure-Property Prediction
Turn microscope images into graphs and predict how strong the metal is
Represent alloy or composite microstructure images as graphs — with grains as nodes and boundaries as edges — and train a graph neural network to predict mechanical properties like yield strength and fatigue life. A research-grade application of geometric deep learning to materials science.
ML Model for Combustion Instability Detection
Train a neural network to hear when a combustor is about to go unstable
Train a time-series classifier — LSTM or 1D-CNN — on pressure oscillation data to detect thermoacoustic instability in combustion chambers. Addresses a critical safety problem in rocket engines and gas turbines with deep learning on sequential sensor data.
Engine Sound Anomaly Detection with Autoencoders
Teach a neural network what normal sounds like — then catch everything else
Train a convolutional autoencoder on mel-spectrograms of normal engine audio to detect anomalous sounds using reconstruction error as the anomaly score. Addresses the practical challenge of detecting unknown fault types without labeled failure data.
ML-Optimized Flight Route Planning Around Weather
Route aircraft around storms and turbulence with graph search and ML
Build a system that ingests weather forecast grids and uses graph search with an ML-learned cost model to find fuel-optimal routes that avoid turbulence and convective weather. Compare against great-circle routes to quantify fuel and safety improvements.
Process Parameter Optimization for Additive Manufacturing
Use Bayesian optimization to tune 3D printing for aerospace-grade quality
Build a Bayesian optimization pipeline that tunes additive manufacturing parameters (laser power, scan speed, layer thickness) to minimize porosity and maximize tensile strength. Apply surrogate modeling and intelligent search to a real aerospace manufacturing challenge.
Radar Target Classification with Deep Learning
Classify drones, birds, and aircraft from synthetic radar signatures
Train a CNN on synthetic radar range-Doppler maps to classify aerial targets (drone, bird, aircraft, clutter). Generate realistic synthetic data using radar cross-section models and signal processing fundamentals.
Contrail Prediction and Avoidance with ML
Predict where contrails form and reroute flights to avoid them
Build an ML model predicting contrail formation probability from atmospheric conditions and satellite imagery. Explore how small route adjustments can significantly reduce aviation's non-CO2 climate impact.
Predict Pilot Fatigue Risk from Flight Schedule Data
Model the invisible threat to flight safety with ML
Model cumulative fatigue using the SAFTE/FAST framework, then train an ML model on schedule features to predict fatigue risk scores. Tackle one of aviation safety's most challenging human factors problems.
ML-Based Gust Load Prediction for Wing Structures
Replace expensive time-domain simulations with a trained sequence model
Train a sequence model (LSTM or Transformer) on time-series gust encounter data to predict peak wing root bending moment. Compare against classical 1-minus-cosine gust analysis for a real-world loads problem.
Neural Network Surrogate for Thermal Protection System Design
Accelerate reentry vehicle thermal design with deep learning
Train a neural network on parametric FEA results to predict temperature distribution through a multi-layer thermal protection system for reentry vehicles. Use the surrogate for rapid design-space exploration.
Other Levels
High School
Hands-on projects for students with basic math and science. No prior coding experience required — every project walks you through setup from scratch.
36 projects →Undergraduate
Engineering and CS projects for college students. Expect to work with real industry tools, write substantial code, and produce results you can show in interviews.
32 projects →