Integration of Data-Level Fusion Model and Kernel Methods for Degradation Modeling and Prognostic Analysis

To prevent unexpected failures of complex engineering systems, multiple sensors have been widely used to simultaneously monitor the degradation process and make inference about the remaining useful life in real time. As each of the sensor signals often contains partial and dependent information, data-level fusion techniques have been developed that aim to construct a health index via the combination of multiple sensor signals. While the existing data-level fusion approaches have shown a promise for degradation modeling and prognostics, they are limited by only considering a linear fusion function. Such a linear assumption is usually insufficient to accurately characterize the complicated relations between multiple sensor signals and the underlying degradation process in practice, especially for complex engineering systems considered in this study. To address this issue, this study fills the literature gap by integrating kernel methods into the data-level fusion approaches to construct a health index for better characterizing the degradation process of the system. Through selecting a proper kernel function, the nonlinear relation between multiple sensor signals and the underlying degradation process can be captured. As a result, the constructed health index is expected to perform better in prognosis than existing data-level fusion methods that are based on the linear assumption. In fact, the existing data-level fusion models turn out to be only a special case of the proposed method. A case study based on the degradation signals of aircraft gas turbine engines is conducted and finally shows the developed health index by using the proposed method is insensitive for missing data and leads to an improved prognostic performance.