Condition Monitoring Time Series Retrieval Based on Feature Patterns

Abstract

This paper analyzes the problems existing in condition monitoring time series of engineering machinery and equipment. A series eigenvector extraction model is developed, which can be used to retrieve several representative shape eigenvectors showing the sub-series variation trend from the long time series. Based on the eigenvectors output from the model, a feature pattern discovering model is constructed to discover the shape feature pattern of eigenvectors. Based on the obtained feature pattern set and the original condition monitoring time series, we adopt the design concept of inverted index from Internet search engines to build the pattern index database. In this database, the user input time series can be retrieved and the corresponding results can be output. It enables users to quickly locate desired time series similar to their input, making data analysis significantly more convenient for users.