Fig 1.
Waveforms and spectrograms of vibration signals from healthy and broken gearboxes.
Fig 2.
The multi-scale convolutional neural network with depth-wise feature concatenation architecture.
Fig 3.
Comparison of vibration signals from four sensors in healthy and broken gearbox states.
Fig 4.
Spectrogram images generated from vibration signals of healthy and broken gearbox conditions.
Table 1.
Performance and time comparison of different models on the Gearbox Fault Diagnosis dataset (mean std).
Table 2.
Statistical significance of MixNet accuracy compared to other models.
Fig 5.
Confusion matrices for different models in gearbox fault diagnosis: MixNet, GoogLeNet, SqueezeNet, AlexNet, and MLPC, results from the first run of 10 independent runs.
Fig 6.
Training loss of the investigated models over calculation times.
Fig 7.
Comparison of deep learning models for gearbox fault diagnosis: Accuracy (higher is better) vs. Training Time (lower is better).