Multi-temporal sequential pattern mining based improvement of alarm management systems

Even in a case of a simple failure, modern process control systems can cause a vast number of alarms. Due to the overload of the operators these alarm floods may result in tragedical accidents. Alarm management systems can suppress correlated and predictable alarms to reduce the workload of the operators. Since the process units of complex production systems are strongly interconnected, the signals defined on different process variables generate complex multi-temporal patterns. We propose a multi-temporal sequence mining based approach to extract these patterns and form alarm suppression rules. We demonstrate the applicability of the concept in a vinyl-acetate production technology. The results illustrate the multi-temporal analysis of events defined on process variables can detect causes of alarm, and prevent alarm floods by pro-actively suppressing alarms based on the extracted sequences of events.