Research Article

Multiscale Time-Frequency Sparse Transformer Based on Partly Interpretable Method for Bearing Fault Diagnosis

Table 2

The performance of different methods under the XJTU-SY dataset.

MethodSignal processingAverage accuracy (%)

CNNRaw vibration signal90.42
CNN-LSTMRaw vibration signal88.79
Bi-LSTMTFR (SWT)89.47
WRN-16-2TFR (STFT)90.53
TSTRaw vibration signal93.37
TFTTFR (SWT)96.43
BAFTRaw vibration signal97.29
MTFST (proposed)TFR (STFT)99.34