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

Figure 14

Visualization of the learned features in different methods via t-SNE: (a) CNN, (b) CNN-LSTM, (c) Bi-LSTM, (d) WRN-16-2, (e) TST, (f) TFT, (g) BAFT, and (h) MTFST (proposed).

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)