A PATTERN RECOGNITION METHOD TO IDENTIFY THE ROOT FAULTS DURING ALARM FLOODS

Alarm systems can provide comprehensive monitoring of wind turbines and give useful information to remote technicians. However, most of the current alarm systems are suffering from the alarm flood, which means tens or hundreds of alarms appear in a short period of time. This alarm flood overwhelms the operator and requires extensive domain knowledge to interpret. This paper proposes a pattern recognition method using the alarm floods generated during a fault. The aim is to identify the root faults of alarm floods and assist the operator in decision making. Initially, the FP-growth algorithm is applied to extract the related alarms. The related alarms are expressed as the alarm vectors based on the defined rules in this paper. Subsequently, the key alarm vectors of faults are generated based on the maintenance record. Ultimately, the Hamming distance and the Jaccard distance are used to compare the unknown alarm vectors to the key alarm vectors. The performance of the distances is assessed using three defined indicators. The proposed method is verified using actual data from two wind turbines. The results show that the proposed method can identify the root faults of alarm floods effectively.