Projects

Theme A — Fault detection, diagnosis and prognosis (Aerospace & electromechanical systems)

• Fault diagnosis and prognosis of satellite actuators (reaction wheels and control-moment gyros), including RUL estimation
  Description: model-based and data-driven detection of faults in reaction wheels and CMGs, simulate fault scenarios, then build ML/physics-aware models to detect and predict faults.
  Tasks: implement nonlinear simulator (reaction wheel dynamics), generate fault datasets, train classifiers and prognostic models (RUL via survival models or deep learning), validate with sensitivity/robustness tests.
  Skills/tools: MATLAB/Simulink or Python, system identification, time-series ML (LSTM/transformer), Kalman filters, uncertainty quantification.

• Real-time health monitoring and prognostics for UAV rotors and motors
  Description: online monitoring of rotor/motor health on a small rotorcraft platform, detect imbalance, bearing faults, and estimate remaining life.
  Tasks: acquire vibration/current data from rotor testbed, feature extraction (time/freq), lightweight onboard classifier, RUL prediction pipeline, deploy on SBC (Raspberry Pi/Jetson).
  Skills/tools: embedded Linux, sensor interfacing, FFT/wavelet features, embedded ML, ROS optional.

• Sensor fault isolation and reconfiguration for multi-agent systems (satellites/drones/vehicles)
  Description: detect single/multiple sensor faults and design reconfiguration strategies for continued safe operation of multi-agent systems.
  Tasks: design residual generators or data-driven anomaly detectors, test reconfiguration controllers in simulation, implement consensus-based fault-tolerance.
  Skills/tools: control theory, distributed algorithms, simulation frameworks, model-based fault detection.

Theme B — Intelligent manufacturing, computer vision and Industry 5.0

• Real-time object detection for automated inspection and quality control in manufacturing
  Description: build/benchmark fast detection models for defect detection and part verification on production lines, optimize for inference latency.
  Tasks: create dataset from images/video, train/quantize YOLO/SSD/transformer detectors, integrate into simple PLC or edge-device pipeline, evaluate detection rate and latency.
  Skills/tools: PyTorch/TensorFlow, OpenCV, edge inference (TensorRT/ONNX), dataset augmentation, metrics for production readiness.

• Deep learning for surface-defect segmentation and automated grading
  Description: pixel-wise segmentation of manufacturing defects and mapping to grading/accept/reject decisions.
  Tasks: annotate images, train segmentation networks, post-process to compute defect metrics, build a dashboard for operator feedback.
  Skills/tools: segmentation networks (U-Net, DeepLab), label tools, production metrics, UI prototyping.

Theme C — Systems, control and smart infrastructure

• Fault detection and diagnostics in power systems or battery storage for smart cities
  Description: algorithms to detect anomalies and component faults in grid-connected storage or microgrid components using measurement streams.
  Tasks: create/curate dataset or use simulations, apply model-based residuals and data-driven anomaly detection, evaluate on key grid metrics.
  Skills/tools: power system simulators, state estimation, anomaly detection, Python/MATLAB.

• Predictive maintenance for industrial electromechanical systems (motors, gearboxes, bearings)
  Description: end-to-end predictive maintenance pipeline: sensing, feature engineering, anomaly detection, scheduling maintenance windows.
  Tasks: sensor selection, streaming feature computation, deployability analysis, cost/benefit study.
  Skills/tools: time-series ML, streaming systems, economic modelling, reliability engineering.
  
  

Theme D — Multi-agent systems, autonomy and robustness

• Resilient coordination for multi-UAV/satellite teams under actuator faults
  Description: study how teams reassign tasks and maintain mission objectives when some agents suffer actuator degradation or partial failures.
  Tasks: design fault-aware task allocation, simulate degraded dynamics, evaluate mission success under failure scenarios.
  Skills/tools: multi-agent planning, optimization, Gazebo/Matlab multi-agent simulation, RL optional.

• Simulation-to-real transfer for autonomous inspection drones
  Description: use domain randomization and sim2real methods to move perception/control models from simulation to physical platforms for inspection tasks.
  Tasks: build sim environment, train policies/perception, refine with real flight tests and safety envelope constraints.
  Skills/tools: Gazebo/AirSim, domain randomization, transfer learning, safety verification.