【作者】 张旭；

【导师】 于洋；

【作者基本信息】 沈阳工业大学 ， 仪器科学与技术， 2025， 硕士

【摘要】 滚动轴承作为旋转机械的核心部件,其运行状态直接决定设备的安全性与可靠性。然而设备运行过程中由于客观原因,滚动轴承会出现各种故障,带来安全隐患。声发射技术具有对故障的高灵敏度特性和高频信号捕捉能力,在早期轴承故障诊断中拥有独特优势。然而,声发射信号的高频、时变和非平稳特性也对信号处理提出了更高的要求。因此,构建一个能够有效提取故障特征并准确识别故障类型的诊断模型,对于实现轴承早期故障的精准识别、提升设备运行可靠性具有重要意义。针对滚动轴承故障声发射信号早期故障特征微弱导致诊断精度低的问题,本文提出一种基于完全集成经验模态分解自适应噪声(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)结合多维故障表征因子和改进多尺度排列熵(Improved Multiscale Permutation Entropy,IMPE)的特征提取方法。该方法通过计算CEEMDAN分解得到的各分量多维故障表征因子,有效筛选出最能代表故障特征的分量,克服了分量选择的盲目性,并使用小波阈值降噪消除残余噪声。最后计算该分量的改进多尺度排列熵值构造特征向量,用于后续故障诊断。针对滚动轴承故障识别分类问题,提出一种多策略增强改进麻雀搜索算法(Multi-Strategy Enhanced Sparrow Search Algorithm,MESSA)优化支持向量机(SVM)的故障诊断模型。针对使用麻雀算法容易陷入局部最优的问题,MESSA算法对初始化和更新方式提出改进,并在14个基准测试函数的实验中表现出优越性,显著优于灰狼优化与粒子群算法。最后采用MESSA-IMPE-SVM模型进行轴承状态诊断研究,实验结果显示,IMPE相比传统的多尺度排列熵,能更有效地区分不同故障类型。同时MESSA优化的SVM模型在轴承状态诊断中的准确率达到97.5%,证明该方法诊断效果出色。进一步与不同算法对比,相较于麻雀算法、粒子群优化算法和灰狼优化算法准确率分别提升4.2%、7.5%和1.67%。证明了本文提出的故障诊断模型在特征提取与故障识别方面的有效性,为滚动轴承的精准故障诊断提供了可靠的支持。更多还原

【Abstract】 Rolling bearings,as core components of rotating machinery,directly determine the safety and reliability of equipment.However,due to objective factors during equipment operation,rolling bearings can develop various faults,introducing safety hazards.Acoustic emission technology,with its high sensitivity to faults and high-frequency signal capture capabilities,offers unique advantages in early bearing fault diagnosis.Nevertheless,the high-frequency,time-varying,and non-stationary characteristics of acoustic emission signals impose higher requirements on signal processing.Therefore,constructing a diagnostic model capable of effectively extracting fault features and accurately identifying fault types is of significant importance for precise identification of early bearing faults and enhancing equipment operational reliability.Addressing the issue of low diagnostic accuracy caused by weak early fault characteristics in rolling bearing acoustic emission signals,this thesis proposes a feature extraction method based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN combined with multidimensional fault characterization factors and Improved Multiscale Permutation Entropy,IMPE.This method effectively screens components that best represent fault characteristics by calculating multidimensional fault characterization factors of each component obtained through CEEMDAN decomposition,overcoming the blindness in component selection,and employs wavelet threshold denoising to eliminate residual noise.Finally,the improved multiscale permutation entropy values of these components are calculated to construct feature vectors for subsequent fault diagnosis.For rolling bearing fault identification and classification,a Multi-Strategy Enhanced Sparrow Search Algorithm,MESSA optimized Support Vector Machine,SVM fault diagnosis model is proposed.Addressing the problem of the sparrow algorithm easily falling into local optima,the MESSA algorithm improves initialization and update methods,demonstrating superior performance in experiments on 14 benchmark test functions,significantly outperforming Grey Wolf Optimization and Particle Swarm Optimization algorithms.The MESSA-IMPE-SVM model is subsequently employed for bearing condition diagnosis research.Experimental results indicate that IMPE,compared to traditional multiscale permutation entropy,can more effectively differentiate between various fault types.Concurrently,the MESSA-IMPE-SVM model achieves a diagnostic accuracy of 97.5%in bearing condition diagnosis,demonstrating excellent diagnostic performance.Further comparison with different algorithms shows accuracy improvements of 4.2%,7.5%,and 1.67%compared to the Sparrow Algorithm,Particle Swarm Optimization Algorithm,and Grey Wolf Optimization Algorithm,respectively.This validates the effectiveness of the proposed fault diagnosis model in feature extraction and fault identification,providing reliable support for precise fault diagnosis of rolling bearings.更多还原