A Method of False Alarm Recognition Based on k-Nearest Neighbor

Authors:
  • FGFei Guan
    Fei Guan
  • JSJunyou Shi
    Junyou Shi
  • XMXiaodong Ma
    Xiaodong Ma
  • WCWeiwei Cui
    Weiwei Cui
  • JWJie Wu
    Jie Wu
Publisher website
Google Scholar
Abstract
Built-in test (BIT) is widely used in state monitoring and fault diagnosis, and it has a good effect on improving testability of a system. With the structure and function of system more and more complicated, false alarm emerges and prevents wider application of BIT. The mechanism of false alarm is hard to analysis for a complicated system, and the traditional methods for false alarm recognition such as repeated test cannot meet the requirement. To suppress false alarm, a method based on k-nearest neighbor (KNN) is proposed. In this method, the states of a system are classified into 3 types: normal state, false-alarm state and faulty state, and the origin data are pre-processed by k-means algorithm to abandon noisy points. Then KNN is used to recognize false alarm based on the processed data. Comparing the methods based on intelligent algorithms, this method is more appropriate in multi-classification, and used fewer samples. Finally, a system of fault injection is taken as an example, and the result shows that this method is effective.