A marine diesel engine fault diagnosis technology based on digital twin enables precise prediction and efficient maintenance of equipment status through virtual simulation and real-time data fusion. To address the information gap between physical and virtual spaces and the limited fault modes of the twin, which hinder the direct application of diagnostic models in industrial environments and the identification of unknown faults, this paper proposes a multi-condition fault diagnosis method for marine diesel engines based on dynamic triplet adversarial enhancement and adaptive selectivity. A digital twin model of the marine diesel engine is constructed to obtain twin data under different operating conditions. A data augmentation method based on dynamic triplet adversarial joint loss is designed to mitigate distribution discrepancies between twin data and physical space data, thereby enhancing the quality of the twin data. Additionally, a selective domain adaptation method based on adaptive weights is developed to identify known and unknown fault modes under various conditions without prior information. A marine diesel engine digital twin system is established to achieve precise engine control and virtual-real synchronization based on real-time operational data. Experimental results demonstrate that the proposed method significantly improves fault diagnosis accuracy and enhances the recognition of unknown faults, verifying the feasibility and engineering value of the proposed framework.