Degradation model of multi-signals based on a novel feature-selection criterion

In most data-driven prognostics approaches, features extracted from measurements are used in the model to indicate the degradation process and determine the reliability in real time. However, many features with physical meaning commonly exhibit no variation until a failure occurs, which leaves little time to conduct maintenance strategies or replacement policies. Hence, this research presents a novel feature-selection criterion, which enables to select a feature with an obvious trend throughout the entire life, thereby avoiding the problem mentioned. In addition, for reliability estimation and condition monitoring, an innovative model-building method based on identical statistical features extracted from multi-signals is developed, in which the features are considered to be dependent and governed by a copula function. An example is provided to illustrate the application of the proposed two-step methods. The result shows that this method is more convincible and realistic for reliability estimation.