The edge surface warping defect seriously affect the surface quality of strips. In this paper, a technology for diagnosis of warping defects in hot-rolled strip based on data-driven methods is studied. Based on the mechanism analysis of the warping defects, the process parameters affecting the warping defects were sorted out and used as the original input parameters of the defect diagnosis model. Firstly, a diagnostic model that combines the deep belief network and contribution plots of each dimensionality reduction layer is proposed. The deep belief network that integrates each dimensionality reduction layer can predict product defects more accurately and stably than the traditional deep belief network. Meanwhile, on the basis of the pre-judgment model, the method of contribution plot is further introduced to trace the defects, and the comprehensive diagnosis function of model pre-judgment and traceability is realized. Finally, collected the production data from a hot rolling production line for a period of time. Tested the model and predicted a hit rate of 85%. The main influencing factors of edge surface warping defects were determined that the rate of defect decrease with the increase of furnace temperature. When the heating temperature of the second stage of the heating furnace is higher than 1160°C, the incidence of defects is significantly reduced. Defect rate is relatively low within 240min of total furnace time. With the first and third pass phosphorus removal equipment turned on, the incidence of defects was relatively low.