A fault prediction system based on improved workflow was proposed to reduce the complexity of model construction and parameter configuration,which was a workflow-based framework to visually design,execute and deploy machine learning experiments.By using this platform,not only experts can use various fault prediction algorithms and tools more easily,but also non-experts without machine learning knowledge and experience can participate in the construction of machine learning experimental models.The feasibility of the system architecture design was verified through the fault prediction experiment of the surface drying furnace equipment.