Scholarlink

<p>Accurate prediction of the remaining useful life (RUL) of rubber V-belts is essential for enhancing reliability, reducing maintenance costs, and optimizing equipment utilization. This study proposes a hybrid methodology that inte-grates physical property models and dynamic time-series characteristics for V-belt RUL prediction. A dual-path fea-ture extraction framework identifies key physical properties—tensile strength, reference force elongation, elastic modulus, and hardness—and dynamic operational features, such as slip rate and tension force. These features are fused using support vector regression (SVR) with adaptive parameter optimization to derive two comprehensive degradation indicators. The fused features are incorporated into a bivariate Wiener process model, which synergisti-cally captures physical and fatigue degradation behavior for RUL prognostics. The model parameters are adaptively estimated via the maximum likelihood estimation method, enabling accurate and flexible predictions. Experimental validation under standard operating conditions demonstrates that the proposed method achieves prediction accuracy within ±5 % for both MAE and RMSE metrics, while exhibiting robustness and generalizability across different V-belt specifications. This work offers a novel, physics-informed fusion approach and lays a solid foundation for improving V-belt reliability, scheduling predictive maintenance, and optimizing lifecycle management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Physical and dynamic time-series feature fusion based remaining life prediction of V-belts using SVR and Wiener process

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Changyi Lei,
  • Jun Wu,
  • Wenqian Yang

摘要

Accurate prediction of the remaining useful life (RUL) of rubber V-belts is essential for enhancing reliability, reducing maintenance costs, and optimizing equipment utilization. This study proposes a hybrid methodology that inte-grates physical property models and dynamic time-series characteristics for V-belt RUL prediction. A dual-path fea-ture extraction framework identifies key physical properties—tensile strength, reference force elongation, elastic modulus, and hardness—and dynamic operational features, such as slip rate and tension force. These features are fused using support vector regression (SVR) with adaptive parameter optimization to derive two comprehensive degradation indicators. The fused features are incorporated into a bivariate Wiener process model, which synergisti-cally captures physical and fatigue degradation behavior for RUL prognostics. The model parameters are adaptively estimated via the maximum likelihood estimation method, enabling accurate and flexible predictions. Experimental validation under standard operating conditions demonstrates that the proposed method achieves prediction accuracy within ±5 % for both MAE and RMSE metrics, while exhibiting robustness and generalizability across different V-belt specifications. This work offers a novel, physics-informed fusion approach and lays a solid foundation for improving V-belt reliability, scheduling predictive maintenance, and optimizing lifecycle management.