Multi-Sensor Fault Diagnosis of Underwater Thruster Propeller Based on Deep Learning

Area of Science:

• Marine engineering
• Robotics
• Artificial intelligence

Background:

• Unmanned underwater vehicles require reliable fault diagnosis to prevent catastrophic failures.
• Propeller damage is a common issue in underwater thrusters, necessitating effective health monitoring.

Purpose of the Study:

• To develop and evaluate a fault diagnosis method for underwater thruster propellers.
• To assess the effectiveness of deep learning models in analyzing sensor data for propeller health.

Main Methods:

• A deep convolutional neural network was employed for fault diagnosis.
• Current signals from a Hall element and sound signals from a hydrophone were acquired.
• Fast Fourier Transform (FFT) was used to convert time-domain signals to the frequency domain for neural network input.

Main Results:

• Multi-signal input to the neural network demonstrated higher accuracy than single-signal input.
• A hybrid approach, training two signal types in separate networks before merging, yielded the highest accuracy at 99.88%.
• The proposed method reliably indicated propeller health conditions.

Conclusions:

• Deep learning models, particularly CNNs, are effective for underwater thruster propeller fault diagnosis.
• Fusing multiple sensor signals significantly improves diagnostic accuracy.
• A separated network architecture for signal fusion offers superior performance for propeller health monitoring.