Due to the increasing competition in the process industries, there has been a strong need to detect, locate and estimate faults, and to recover the process from faulty states. By providing the operators with supportive information or by recovering the system automatically, fault diagnosis keeps the process running efficiently and safely, thereby bringing enormous economical benefits to the industries.
During the last three decades there has been considerable progress in process automation. Although most chemical plants have automatic systems for shut down and the management of abnormal and unsafe situations, many accidents occur. This clearly demonstrates that safety and fault detection systems are not implemented properly. In order to improve the applicability of fault diagnosis methods, further academic development is needed. Obtaining good fault diagnosis results usually requires a more detailed process model, the development of which is a laborious and time-consuming task. On the other hand, data-driven models are easy to generate from historical data. However, most of these models are able to detect the occurrence of a fault but not to diagnose the fault.
Causal digraph-based fault diagnosis methods attempt to solve this conflict by utilizing process structure knowledge. The latest developments in the causal-directed graph (digraph) based fault diagnosis method guarantee a good physical explanation for fault occurrence and, simultaneously, utilize the time-saving advantage of black-box system identification methods. The causal structure of the process is usually obtained either from process knowledge or through causal analysis of the historical data.
However, the increasing number of process variables means that identification of the causal structure and the cause-effect models is becoming a more and more demanding task. As a result, causal digraph modeling is nowadays faced with a challenge set by the increasing complexity of the processes involved.
The complexity of the process results, in turn, in problems in digraph model construction. This difficulty directly decreases the applicability of the methods. Thus, in order to improve the applicability of the causal digraph fault diagnosis method, it would be beneficial to develop more sophisticated methods to support the construction of the causal structure and cause-effect models.