Data serves as the crucial resource for intelligent fault detection, most notably within the realm of large, low-speed diesel engines. The substantial manufacturing and maintenance expenditures attached to these engines pose significant data acquisition challenges, leading to the scarcity of data samples. In addition, given the inherent privacy protection measures within diesel engine databases, data exchange becomes difficult, making it hard to fuse multi-party data and train models. Addressing these challenges, this paper conducts an in-depth study on the federated learning-based fault diagnosis model specific to diesel engines. This study improves the fault diagnosis accuracy within the convolutional model framework by establishing a fault diagnosis method for diesel engine fuel systems based on federated learning (FLMD). This study provides a novel approach to fault diagnosis in the field of diesel engine faults under data island conditions, achieving a diagnostic accuracy of 95.6%, possessing significant theoretical and practical value.