An Adjustable Functional Regression Model for Real-time Degradation Prognostic Under Incomplete Data Scenarios

作者：Kaiyan Zhang, Ling Cao, Xueqing Xing, Tangbin Xia, Zhen Chen, Ershun Pan, Lifeng Xi · 年份：2023 · DOI：10.1109/ieem58616.2023.10406527 · 研究领域：Reliability and Maintenance Optimization、Machine Fault Diagnosis Techniques、Quality and Safety in Healthcare

Nowadays, most prognostic models heavily rely on the complete and intact historical degradation signals to identify underlying deteriorated trends for predicting the lifetime of the engineering system. However, in real-world scenarios, these degradation signals are always collected with inconsistent distributions and incomplete observations, which compromises the ability to establish precise degradation models. Therefore, we have formulated an adjustable functional lifetime regression model with the real-time prognostic capability to tackle distribution shifts and incomplete data. Firstly, feasible degradation curves and informative features are identified through a functional perspective. Consequently, the relationships between the represented features and the lifetime labels are formulated by the innovative regression model with an adjustable functional basis. Finally, by leveraging real-time signals, our method can refine and update the time-to-failure (TTF) results. The experimental results significantly demonstrate the prognostic robustness, evaluation precision, and application prospects of the proposed approach.