To deal with the problem of weak fault characteristics and misdiagnosis in one-time fault diagnosis, a cross-validation enhanced digital twin-driven fault diagnosis methodology for is proposed. The models for fault diagnosis using Bayesian networks are constructed, integrating both virtual and real data, specifically designed for addressing composite faults and minor faults. A multiple diagnosis mechanism of diagnosis, verification and re-diagnosis is established through data exchange between the digital twin model and the fault diagnosis model. This fault diagnosis method is applied to a subsea production system in the South China Sea. The results indicate that compared to a one-time fault diagnosis, this method can identify minor faults and composite faults in the subsea production system accurately.