摘要
传统神经网络在轴承故障诊断中存在参数选择困难、准确率受限及处理时间过长等问题,为此,提出一种基于注意力机制和贝叶斯优化的改进模型。注意力机制通过对参数权重的分配来增强模型对关键特征的捕捉能力,但参数量的增加会使优化的难度增大;贝叶斯优化通过概率代理模型指导参数的搜索方向,避免对无效参数组合的处理,可大幅缩短调参时间。实验结果表明,改进后的模型准确率提升至99.14%,训练时间大幅缩短。
To tackle the challenges associated with parameter selection difficulties,limited accuracy,and prolonged processing times in traditional neural networks used for bearing fault diagnosis,we propose an enhanced model that integrates an attention mechanism with Bayesian optimization.The attention mechanism improves the identification of critical features by dynamically adjusting parameter weights.However,this increase in parameter complexity introduces optimization challenges.Bayesian optimization mitigates these issues by employing a probabilistic surrogate model to guide the search for optimal parameters,thereby avoiding inefficient parameter combinations and significantly reducing tuning time.Experimental results demonstrate that the improved model achieves an impressive accuracy of 99.14%while substantially decreasing training time.
作者
胡明涛
王风涛
钱居楠
王子豪
李椋泓
HU Mingtao;WANG Fengtao;QIAN Junan;WANG Zihao;LI Lianghong(School of Mechanical and Automotive Engineering,Anhui University of Engineering,Wuhu Anhui 241000,China)
出处
《重庆科技大学学报(自然科学版)》
2025年第4期88-98,共11页
Journal of Chongqing University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金项目“高速轻载工况下圆柱滚子轴承蹭伤机理及振动特性研究”(51905001)
安徽未来技术研究院企业合作项目“CT球管液态金属轴承关键技术研究”(2023QYHZ22)。
关键词
故障诊断
CNN
注意力机制
贝叶斯优化
fault diagnosis
CNN
attention mechanism
Bayesian optimization
