Fault prediction technology effectively guarantees the smooth and orderly production process and the safety of personnel. However, in actual operation, qualitative and quantitative informations of process data coexist, and the model is complex. In addition, in the production process, there is a timing delay problem when using online collected data for fault prediction. This paper establishes, verifies and applies a fault prediction model based on time dynamic causality diagram(TDCD). In the process of model building, a parameter delay time interval learning algorithm is proposed, that is, the mobile search maximum maximal information coefficient(MIC) algorithm, which fully considers the timing delay problem. In the reasoning process, trend analysis and delay information sorting are added to optimize the reasoning process and reduce the false alarm rate caused by delay time. Finally, the algorithm is validated by using a causal graph network for a flotation process. The proposed strategy is applied to the hydrometallurgical leaching process, and compared with the single-valued/multi-valued uncertain dynamic causality diagram, which shows the advancement and effectiveness of the fault prediction strategy.