用户名 密 码 登 录 忘记密码 免费注册

网站会员中心 | 设为首页 | 加入收藏

 您好,欢迎光临中国轴承行业网         退出登陆

会员中心 | 设为首页 | 加入收藏

国外文献 > 1.轴承 > 5. 监测与诊断 > 故障诊断


关键词搜索:


标 题: 3D PRINTER BEARING FAULT IDENTIFICATION BASED ON FINE-TUNING INCREMENTAL LEARNING
基于精细增量学习的3D打印机轴承故障识别
作 者: Chengqian Zhao
Zhuoxun Liu
Rong Zhao
Zhe Yang
Jianyu Long
Chuan Li
Weiliang Cai
关键词: FINE-TUNING  [微调]
INCREMENTAL LEARNING  [ 增量学习]
FAULT DIAGNOSIS  [ 故障诊断]
出版年: 2025
文集名: 15th Prognostics and System Health Management Conference (PHM 2025)
第15届预测与系统健康管理会议(PHM 2025)
会议名: 15th Prognostics and System Health Management Conference (PHM 2025)
第15届预测与系统健康管理会议(PHM 2025)
届 次: 15th
摘 要: Aiming at the challenge that only normal state data can be acquired in the initial stage of 3D printers and traditional fault diagnosis methods are difficult to adapt to the newly added fault categories, this paper proposes an incremental learning method based on dynamic fine-tuning strategy. The Fine-tuning incremental learning (FIL) approach achieves incremental knowledge migration with limited fault samples by adjusting the parameter update strength of the pre-trained feature extractor. Experiments on a dataset of 3D printer joint bearings show that the Fine-tuning method achieves an average task accuracy of 81.59% in three incremental tasks, and the global forgetting degree is controlled at 2.74%, which effectively mitigates the catastrophic forgetting problem. In industrial equipment health monitoring scenarios, the method is able to continuously learn new failure modes while maintaining the ability to recognise existing failure modes, providing a scalable solution for equipment fault diagnosis over the entire life cycle.

针对3D打印机初期仅能获取正常状态数据,而传统故障诊断方法难以适应新增故障类别这一挑战,本文提出了一种基于动态微调策略的增量学习方法。该微调增量学习(FIL)方法通过调整预训练特征提取器的参数更新强度,在仅有有限故障样本的情况下实现知识的逐步迁移。在3D打印机关节轴承数据集上的实验表明,FIL方法在三个增量任务中平均达到81.59%的任务准确率,全局遗忘程度控制在2.74%,有效缓解了灾难性遗忘问题。在工业设备健康监测场景中,该方法能够在持续学习新故障模式的同时,保持对现有故障模式的识别能力,为设备全生命周期的故障诊断提供了一种可扩展的解决方案。
中文内容为网络AI翻译,专业术语未必准确,仅供参考。

VIP网员

联系我们 | 登录注册 | 会员管理 | 网员服务 | 服务信息 | 广告服务 | 高薪聘请 | 法律声明 | 关于我们

京公网安备 11010202009788    京ICP备2021015424号