Modeling and Testing of Temporal Dependency in the Failure of a Process System

The complexities of process plants
are increasing because of process
integration and plant-wide optimization. Failure models of a process
system (henceforth referred to as process-accident models) should
be able to capture the inherent dependence among process components
and their associated variables and also the temporal dependencies
among failures. This work demonstrates the suitability and applicability
of process-accident models in capturing temporal dependence using
process data. Performances of process-accident models are investigated
to establish their competitive advantages as well as their limitations.
Using experimental data from a pilot plant, the performances of three
widely used accident models, namely, the fault tree (FT), the dynamic
fault tree (DFT), and the dynamic Bayesian network (DBN), are evaluated
in predicting abnormal events. Normal and abnormal process data is
collected and used in studying the three different models to assess
the process-accident probability. The study confirmed the DBN model
to be the most appropriate accident-modeling approach because of its
flexible structure and ability to capture spatial and temporal dependencies.