Machine Learning Based Approach for Industrial Alarm Detection
Alarm systems, Alarm turning
ncreasing safety restrictions on modern plant operation have resulted in a growing complexity of industrial supervision and control. In this context, alarm systems play a relevant role in meeting the safety standards required in an operational process. Alarm systems are modeled to show any abnormalities that occur in the operation. This requires a good adjustment of the alarm parameters in order to make them more useful. A misconfiguration of these parameters can induce an incorrect decision of the operator, leading to an accident. These adjustments are not obtained simply and are capable of generating various problems such as false alarms, missed alarms and noisy alarms. In order to achieve a well-performing alarm system, an approach is proposed in this paper that uses machine learning techniques to decide when an alarm should be activated or not. System inputs are the current and past values of the process variable associated with past values of the alarm variable. To validate the proposed approach, a case study using the Tennessee Eastman Process industrial plant simulator was conducted to determine an efficient alarm signal. The results indicate that this approach is promising in creating an alarm without undesired behavior.