A LOF-based method for abnormal segment detection in machinery condition monitoring

Machinery condition monitoring has entered the era of big data and some research has been done based on big data. Abnormal segments, such as missing segments and drift segments, are inevitable in big data acquired from harsh industrial environment due to temporary sensor failures, network segment transmission delays, or accidental loss of some collected data and so on. Being independent of the machinery condition, the abnormal segments not only reduce the quality of the data for condition monitoring and big data analysis, but also bring a heavy computation load. However, there are few reports to address abnormal segment detection for further data cleaning in the field of machinery condition monitoring. Therefore, an abnormal segment detection method is proposed to improve the quality of big data. First, a sliding window is used to separate the data into different segments. Then, 14 kinds of time-domain features are extracted from each segment and principle component analysis (PCA) is employed to extract the principle components from these features. In addition, local outlier factor (LOF) is calculated based on the principle components to evaluate the degree of being an outlier for each segment. Finally, the data, including a drift segment from a real wind turbine, are used to verify the effectiveness of the proposed method.