A novel method for real-time alarm root cause analysis

Abstract With complexities in the control systems, it is natural that the chemical plants may become vulnerable to faults in practice. Alarm systems were designed to raise an alarm when a fault is detected. But modern alarm systems often produce large amounts of false or nuisance alarms which leads to alarm floods. Operators receive far more alarms than they can handle. To reduce the alarm floods, we developed a novel method which combined the Time-delayed Convergent Cross Mapping (TCCM) and Bayesian Network (BN). In this paper, the BN is constructed with application of TCCM to process data and expert knowledge. The alarm data is used to estimate the conditional probability among the nodes of the BN. The BN is used to find the alarms propagation paths and root cause at real-time. Finally, the method is applied to the Tennessee Eastman model to illustrate the applicability.