Code for "Propeller Damage Detection: Adapting Models to Diverse UAV Sizes" Submitted to IEEE Robotics and Automation Letters.

Abstract

This study introduces a methodology to train a model for detecting propeller damage in unmanned aerial vehicles (UAVs), and demonstrates how this model can be modified for use in various UAV sizes through a novel domain adaptation method anchored in the vehicles' physics. The damage detection model is trained by simulating propeller failure, specifically by chipping away the tip of a propeller. A key feature is that the detection model leverages only inertial sensors that are standard in commercial UAVs, making it broadly applicable without the need for additional hardware.

Video

https://www.youtube.com/watch?v=i1G_ICewvB0

To run

1. Clone this repository.

2. Install Python3 dependencies, matplotlib, numpy, pytorch, scikit-learn, scipy, seaborn, torch, lightning, tqdm

3. Run model_train.py with the following configuration for each experiment:

Experiments

Experiment VI.A

Use the quadcarbono_datamodule inside model_train.py

data_module = quadcarbono_datamodule

Experiment VI.B

Use the affine_datamodule inside model_train.py

affine_datamodule = affine_DataModule(
    source_path="data/garage_manual/",
    target_path="data/quadcarbono/",
    source_fs=222,
    target_fs=222,
)
data_module = affine_datamodule

Experiment VI.B.1

For the naive use no scaling, set self.scale_factor = 1.0 inside affine.py

Experiment VI.B.2

For the scaled experiment, set self.scale_factor = 0.734 inside affine.py

Experiment VI.B.3

For this experiment use the data in quadcarbono_weight_shift as test instead of Hasymm164_inertial

Experiment VI.C

Use the affine_datamodule inside model_train.py

affine_datamodule = affine_DataModule(
    source_path="data/quadcarbono/",
    target_path="data/quadcarbono_1000/",
    source_fs=222,
    target_fs=222,
)
data_module = affine_datamodule

Experiment VI.C.1

For the naive use no scaling, set self.scale_factor = 1.0 inside affine.py

Experiment VI.C.2

For the scaled experiment, set self.scale_factor = 1.12 inside affine.py

Experiment VI.C.3

For this experiment use quadcarbono_water as test

Experiment VI.D

affine_datamodule = affine_DataModule(
    source_path="data/quadcarbono/",
    target_path="data/hexaF550/",
    source_fs=222,
    target_fs=222,
)
data_module = affine_datamodule