In order to meet the requirements for the remote adaptive diagnosis of diesel engines, a data-driven fault diagnosis method was proposed and verified during the fault diagnosis of EGR system for China 6 diesel engine. Based on the real-time remote monitoring data of diesel engine test fleet, a data model was built and eleven parameters closely related to EGR were selected as the model characteristic variables. The dimension reduction of raw high-dimensional data was conducted by using the kernel principal component analysis (KPCA). Then the classifier training was carried out in respect of normal and fault data of EGR system by using the kernel fisher discriminant analysis (KFDA). In the end, the fault diagnosis was realized. The experimental results show that the proposed method can diagnose two typical faults of EGR system effectively and the combination of KPCA and KFDA solves the existing problems of dealing with nonlinearity and strong coupling of fault variable for diesel engine with the linear method.