Quantitative selection of sensor data based on improved permutation entropy for system remaining useful life prediction

Condition monitoring is an effective tool for diagnosing and predicting the system fault or failure. One class of method in system condition monitoring is based on the condition data (i.e., data-driven methodology). However, not all the collected condition data can be utilized for the data-driven methodology. Hence, the selection of reasonable condition data is crucial for the application of the data-driven methodology. This is especially useful for the system which has the characteristics of degradation. In such system, the condition data that have the increasing or decreasing trend are desirable. This article proposes the quantitative selection of sensor data for system remaining useful life prediction. The main advantage of the proposed metric to select sensors is that the information theory is adopted. Hence, the selection of sensors can be determined by the proposed quantitative metric. Two case studies which include one simulation data set and one practical data set are carried out to evaluate the effectiveness of the proposed metric. The detailed experiments prove the advantage of the proposed approach.