Algorithm 1: WCIGDG

# Adversarial Training stage

Input: source domain dataset $D^s$, multiple auxiliary domain datasets $D^a={\left\{D_m^a\right\}}_{m=1}^{M_a}$

1: for epoch = 1 to epochs = N do

2: Forward propagation to generate ${\tilde{X}}_t=G({\mathit{X}}_{\mathrm{s}},{\mathit{Y}}_{\mathrm{s}})$.

3: Calculate $L_{\mathrm{C}}$ and $L_{\mathrm{D}}$

4: Backward propagation to update $C$ and $D$

5: Calculate $L_{\mathrm{G}}$

6: Backward propagation to update $G$

Return: Trained generator $G$

# Fault Diagnosis Model Training stage

Input: source domain dataset $D^s$.

1: Forward propagation to generate ${\tilde{X}}_{\mathrm{t}}=G({\mathit{X}}_{\mathrm{s}},{\mathit{Y}}_{\mathrm{s}})$.

2: for epoch = 1 to epochs = N do

3: Calculate $L_{\mathrm{C}}$ based on the classification results of generated data ${\tilde{\mathit{X}}}_{\mathrm{t}}$ and real data from source domain ${\mathit{X}}_{\mathrm{s}}$.

4: Backward propagation to update $\tilde{C}$

Return: Trained classifier $\tilde{C}$