一种基于注意力机制的滚动轴承域适应故障诊断方法

本发明涉及一种基于注意力机制的滚动轴承域适应故障诊断方法，对采集到的滚动轴承振动监测信号，由一维可分离式卷积并嵌有通道注意力机制和长度注意力机制的特征提取器提取到深层故障特征；构建局部注意力域适应模块和全局注意力域适应模块筛选可迁移性好的信号及信号片段，从而提高模型的泛化能力，使模型能够更好地应对变工况故障诊断问题；与迁移学习智能故障诊断算法对比，该算法考虑到不同信号及信号片段的可迁移性不同，提高了模型的可解释性；应用多种轴承振动数据进行实验，都验证了该算法的性能稳定性好，能够在各类工况变化条件下依然保持较为优秀的诊断结果。

This invention relates to a domain adaptation fault diagnosis method for rolling bearings based on attention mechanisms. For acquired rolling bearing vibration monitoring signals, deep fault features are extracted via a feature extractor integrated with 1D separable convolution, channel attention mechanism and length attention mechanism. Local attention domain adaptation modules and global attention domain adaptation modules are then constructed to screen signals and signal segments with favorable transferability, thereby enhancing the model's generalization ability and enabling it to better tackle fault diagnosis tasks under variable working conditions. Compared with intelligent fault diagnosis algorithms based on transfer learning, the proposed algorithm accounts for the varying transferability across different signals and signal segments, thus improving the model's interpretability. Experiments conducted on multiple bearing vibration datasets verify that the proposed algorithm exhibits excellent performance stability and can maintain relatively outstanding diagnostic results even under various working condition variations.

搜集汇总

背景与挑战

背景概述

该数据集是一项专利资源，提出了一种基于注意力机制的滚动轴承域适应故障诊断方法。该方法通过一维可分离式卷积结合通道和长度注意力机制提取振动信号特征，并利用局部和全局注意力域适应模块提升模型泛化能力，以应对变工况下的故障诊断问题，实验验证了其性能稳定性。