Health indicator construction based on Double attribute feature deviation degree and its application into RUL prediction

The construction of health indicator (HI) and the determination of failure threshold are crucial steps in predicting remaining useful life (RUL), traditional method ignores the importance of various feature parameters to HI construction at different operating times, and there is the issue of difficulty in determining personalized failure threshold. To address these issues, this study proposes a method for HI construction based on double attribute feature deviation degree and a method for determining the failure threshold based on the similarity. Firstly, the advantageous features are selected from multiple domains, and double attributes of numerical and contribution attributes are assigned to each advantageous feature, in which the contribution attribute is used to describe the contribution degree of the feature parameter in the process of constructing HI. Then the HI is constructed based on the deviation of the initial state from the degraded state. Subsequently, a curvature dynamic time warping distance considering the shape information of the degradation trend trajectory is proposed to measure the similarity, and the failure threshold is determined by constructing a failure threshold adjustment function. Finally, RULs are estimated by the long short-term memory neural network. The effectiveness of the proposed method is validated using real datasets.