A Deep Learning-Based Log-Mel Spectrogram-Utilized Fault Diagnosis Algorithm for In-Wheel Motor-Driven Vehicles

Key Points

• The study aims to propose a fault diagnosis technique using deep learning for in-wheel motor-driven vehicles to ensure driving safety.
• Attached 3-axis accelerometers on vehicle components including upper arm and brake caliper.
• Collected vibration data under normal and bump road conditions.
• Defined normal and fault states based on performance issues like inverter damage.
• Converted vibration signals to log-mel spectrograms for analysis.
• Classified data using a transfer learning-based CNN model.
• Achieved an average classification accuracy of 98.80% across four driving condition classes.
• Successfully identified normal and faulty operating conditions.