Name: JORDANA LUCIA REIS
Publication date: 27/06/2024
Examining board:
Name |
Role |
|---|---|
| FLAVIO MIGUEL VAREJAO | Presidente |
| FRANCISCO DE ASSIS BOLDT | Examinador Externo |
| LUIS ANTONIO DE SOUZA JUNIOR | Examinador Interno |
Summary: This work introduces an approach for detecting brake fluid leakage in gondola wagons
using acoustic signals. Gondola wagons, also known as railroad gondolas, gondola cars,
and open wagons are typically used for transporting dry cargo and they rely on pneumatic
brake systems that directly depends on compressed air components for effective braking.
The proposed method employs deep-transfer learning to identify compressed air leakages
based on sound emissions, mirroring the process performed by human wagon inspection
professionals. Data for model training are acoustic signals captured in waveform files
during rail car inspections. In the proposed model, the audio file undergoes processing
in the time-frequency domain to obtain the mel-spectrogram. These spectrograms are
then used as input data for pre-trained deep convolutional neural networks, including
ResNet50V2, ResNet152V2, Xception and InceptionResNetV2. The results demonstrate
strong performance, achieving an accuracy above of 94% and Xception performing with
96.98% of accuracy. This outcome underscores the model’s capability to effectively identify
sounds associated with air leakage in the brake systems of gondola wagons.
