Utilizing fake data (simulated based on mechanism models or generated through data-driven models) for data enhancement is a popular approach to solve the problem of fault diagnosis with small samples.