Abstract： By analyzing the vibration process of cold rolling and using the structure model of the rolling system, a dynamic rolling force of the rolling vertical system was built, with the consideration of the influence of rolling vibration. A data-driven fault diagnosis was proposed based on industrial field data by using ensemble empirical mode decomposition (EEMD) and support vector machine (SVM), with the focus on the generalized fault, which were mostly caused by variations of process parameters under complex working conditions. According to the decoupling effect on measured rolling force data with the EEMD algorithm, the intrinsic mode function (IMF) component was defined as fault eigenvector and chosen as the input into the classifier of vector machine. Then, the vibration-related fault of cold rolling mills could be diagnosed by  the distinction between the normal state and the fault state by SVM.