The process industry is increasingly oriented towards automation. Use is made of both automated systems and human operators to control processes. For this control, operators can use computer systems to execute actions and conduct procedures. Automation of procedures can improve performance and safety, while reducing workload. Identifying procedures suitable for automation currently relies heavily on manual inspection and knowledge of operational staff.

This thesis proposes a method for the analysis of stored process data to identify opportunities for procedural automation in processing plants. The goal of this analysis is the identification of action patterns in operator responses to alarms that are predictable. Action patterns which are predictable can present good opportunities for automation.

The method utilizes the physical layout of the process, combined with a statistical analysis of events occurring in the event log to identify the relevant events in response to each alarm. Sequential pattern mining is then applied to those events. This thesis introduces a novel pattern type called ’independent frequent patterns’ to quantify patterns in response to an alarm. The pattern mining results are condensed into a value for the predictability of the response of an alarm.

In order to ensure validity of the patterns resulting from the method, a training and test set were constructed, consisting of alarms and the response patterns known to be the correct responses to those alarms. The method was designed using the training set and validated using the test set, by applying the method to the alarms in those sets and observing if the correct response patterns were identified. Then, the method was applied to alarms in the event log of which the correct response was not yet known, to rank those alarms on their potential for automation.

The thesis is conducted in collaboration with Shell Energy and Chemicals Park Rotterdam, and the method is applied to data supplied by them. The most common alarm messages are evaluated for the response pattern required to resolve them, the predictability of that pattern and the benefit of automating the response. From this, a ranked list of automation opportunities was constructed, which was then further discussed with operational staff.