Centrala begrepp
Leveraging signal processing and machine learning techniques to automate the detection of rotor blade defects in UAVs during pre and post-flight operations.
Sammanfattning
The paper presents a system for detecting rotor blade defects in Unmanned Aerial Vehicles (UAVs) using vibration analysis and machine learning. The key highlights are:
-
Data Collection:
- Experiments were conducted on a Helipal Storm 4 UAV with four types of rotor blades: normal, cracked, trimmed, and scratched/misshapen.
- Vibration data was collected using two ADXL345 accelerometer sensors mounted on the UAV.
- 8 experiments were conducted on the normal blade and 3 on each of the 6 defective blades.
-
Feature Extraction:
- Time-domain features (amplitude, mean, standard deviation) were extracted from the vibration data.
- Frequency-domain features were extracted using Short-Time Fourier Transform (STFT), Wavelet Packet Transform, Spectral Centroid, and Spectral Skewness.
- A total of 27,786 features were extracted per record.
-
Machine Learning:
- The dataset was split into training (70%) and testing (30%) sets, with stratification to maintain the same normal-to-defective ratio.
- Four machine learning algorithms were evaluated: Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and k-Nearest Neighbors (KNN).
- Principal Component Analysis (PCA) was used for dimensionality reduction.
-
Results and Analysis:
- Without PCA, all models achieved over 99% accuracy in detecting defective rotor blades.
- PCA on the STFT features improved the performance of the tree-based algorithms (DT and RF), but negatively impacted the SVM and KNN.
- Feature isolation analysis showed that the STFT features were the most important, followed by time-domain and wavelet features.
- Feature importance analysis using the DT and RF models revealed that the STFT features, particularly those related to the x-axis, were the most influential in the classification decision.
The work demonstrates the effectiveness of using vibration analysis and machine learning for automated rotor defect detection in UAVs, and provides insights into the critical features and their importance for this task.
Statistik
The amplitude of the vibration signal is calculated using the formula: A = (max - min) / 2.
The mean of the vibration signal is calculated using the formula: x̄ = (Σxi) / n, where xi is a sample value and n is the number of samples.
The standard deviation of the vibration signal is calculated using the formula: σ = sqrt((Σ(xi - x̄)^2) / n), where xi is a sample value, x̄ is the mean, and n is the number of samples.
Citat
"The work presented in this paper deals with UAV health monitoring from the hardware perspective as it explores the use of signal processing and ML techniques to identify rotor blade defects through a vibrational analysis."
"Using two mounted sensors, vibrational data from the UAV is constantly relayed and processed. The processing of this data consists of the extraction of time and frequency domain features that are used to train ML models to classify the performance as normal or defective."