| Autor: |
Sun, Jiechen, Zhou, Funa, Hu, Xiong, Wang, Chaoge, Wang, Tianzhen |
| Zdroj: |
Signal, Image & Video Processing; Nov2024, Vol. 18 Issue 11, p7953-7968, 16p |
| Abstrakt: |
Accurate Remaining Useful Life (RUL) predictions of drive motors operated under multiple working conditions are crucial for ensuring the safe operation of electric vehicles (EVs). Traditional prediction methods ignore the influence of working condition categories on the degradation process and lack multi-dimensional and multi-grained differential fusion approaches for degradation features at different stages, resulting in limited improvement of the RUL prediction performance under multiple working conditions. Therefore, a macro- and micro-spacetime feature-preference gated recurrent unit (MMFPGRU) is proposed for RUL prediction of electric motors under multiple working conditions. First, a global temporal trend attention mechanism, temporal trend level division, and an updating unit are designed to extract macro- and micro-temporal trend information from monitoring data. Second, a multi-grained feature interaction module is designed to mine the macro- and micro- spatial features of different degradation stages. Ultimately, a feature-selection gate is designed to enhance the extraction of micro-temporal trend information and to select key spacetime features of the degradation, thereby improving the accuracy of RUL prediction. Moreover, the correspondence between working conditions and degradation processes is established by encoding multiple categories of working conditions to help capture the effects of changes in working conditions on motor degradation. The effectiveness and superiority of the MMFPGRU in predicting the RUL under multiple working conditions are confirmed by simulating scenarios of electric motors under various working conditions using the C-MPASS dataset. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink