Scholarlink

<p>Polymer-based composite gears offer advantages such as lightweight structure, low friction, and self-lubricating behavior. However, their wear and failure mechanisms strongly depend on material composition and operating conditions. Conventional maintenance schedules and fatigue life prediction models often fail to accurately estimate actual service life. This study integrates wear information into gear failure mode identification and remaining useful life (RUL) prediction using vibration and acoustic signals combined with artificial intelligence (AI). A comprehensive health monitoring platform for 3D-printed polymer composite gears captures surface degradation and vibration data, which are applied to three AI models: support vector regression (SVR), one-dimensional convolutional neural network (1D-CNN), and long short-term memory (LSTM) for fault classification and lifetime prediction. The framework combines wear-informed signal analysis with deep learning to achieve real-time gear condition monitoring, improve maintenance planning, reduce unplanned downtime, and enhance intelligent fault detection in industrial transmission systems. Results reveal three distinct stages of gear operation: running-in, steady wear, and failure. During the steady stage, vibration amplitude remains between 11 and 11.5&#xa0;m/s², while Y-axis vibration, ranging from 13.3 to 13.7&#xa0;m/s², shows the highest sensitivity to wear. The integrated triaxial amplitude gradually increases from 0 to 10,000&#xa0;s, remains stable between 10,000 and 20,000&#xa0;s, and rises sharply after 20,000&#xa0;s, indicating fatigue failure. Among the AI models, LSTM achieves the best RUL prediction performance, with a root mean square error of 229.20&#xa0;s and a coefficient of determination of 0.9983, outperforming 1D-CNN and SVR. These findings demonstrate that wear-informed deep learning effectively captures temporal degradation features for accurate gear failure prognosis. AI-based life prediction promotes sustainable manufacturing aligned with SDG 9 and SDG 12 by enabling proactive maintenance, resource efficiency, and industrial sustainability.</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

Wear-informed gear failure mode identification and life prediction using vibration signals with AI integration

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Chil-Chyuan Kuo,
  • Zi-Huan Wang,
  • Yu-Heng Cai,
  • Armaan Farooqui,
  • Song-Hua Huang

摘要

Polymer-based composite gears offer advantages such as lightweight structure, low friction, and self-lubricating behavior. However, their wear and failure mechanisms strongly depend on material composition and operating conditions. Conventional maintenance schedules and fatigue life prediction models often fail to accurately estimate actual service life. This study integrates wear information into gear failure mode identification and remaining useful life (RUL) prediction using vibration and acoustic signals combined with artificial intelligence (AI). A comprehensive health monitoring platform for 3D-printed polymer composite gears captures surface degradation and vibration data, which are applied to three AI models: support vector regression (SVR), one-dimensional convolutional neural network (1D-CNN), and long short-term memory (LSTM) for fault classification and lifetime prediction. The framework combines wear-informed signal analysis with deep learning to achieve real-time gear condition monitoring, improve maintenance planning, reduce unplanned downtime, and enhance intelligent fault detection in industrial transmission systems. Results reveal three distinct stages of gear operation: running-in, steady wear, and failure. During the steady stage, vibration amplitude remains between 11 and 11.5 m/s², while Y-axis vibration, ranging from 13.3 to 13.7 m/s², shows the highest sensitivity to wear. The integrated triaxial amplitude gradually increases from 0 to 10,000 s, remains stable between 10,000 and 20,000 s, and rises sharply after 20,000 s, indicating fatigue failure. Among the AI models, LSTM achieves the best RUL prediction performance, with a root mean square error of 229.20 s and a coefficient of determination of 0.9983, outperforming 1D-CNN and SVR. These findings demonstrate that wear-informed deep learning effectively captures temporal degradation features for accurate gear failure prognosis. AI-based life prediction promotes sustainable manufacturing aligned with SDG 9 and SDG 12 by enabling proactive maintenance, resource efficiency, and industrial sustainability.