(504h) A Novel Data-Driven Causality Digraph Modeling of Large-Scale Industrial Processes

Standard formal mathematical tools are usually not suitable to address the modeling of large-scale industrial processes since it is hard to obtain the precise mathematical model for such systems. This is due to the fact that the physical laws of large-scale industrial processes are often unknown or too complex. In the big data era, the model of such systems may be developed by using the big data tools. However, the causal digraph model often changes in such systems because of the operation of workers, process control or disturbances. The causality is usually instantaneous. Discovering useful information for modeling is difficult as there are too many interference in the big data. In this paper, we proposed a method to identify the causal relationships between the process variables to get a digraph model in the large-scale industrial processes. The process data set are splitted into a few small data sets. The causal relationships are identified separately from each small data sets. The results of each small data sets are merged to get a graphical model. This model can be used in process monitoring, root cause and hazard propagation analysis. A case study on a Hydrocracking unit is presented to illustrate the application of the proposed method.