结合异常检测算法的轴承故障检测研究

发布时间：2018-03-23 02:30

本文选题：故障诊断　切入点：小波包能量谱　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：滚动轴承是机械设备中常见且十分重要的通用零件,轴承发生故障往往会造成严重的生命财产安全问题。所以,对滚动轴承进行故障检测研究是国内和国外工业领域的热门话题。实际工程中受实际数据采集环境影响,常出现祥本不平衡、样本数量有限问题,本文针对这两大问题提出异常检测算法结合支持向量机的一种新的双步诊断方法。首先对轴承的振动特性进行分析:在对轴承的基本结构、振动特性和轴承的动力学特性进行分析后,确定采用振动信号的故障诊断方式。接着采用小波包能量谱特征提取方式提取轴承振动信号:采用小波包变换对轴承信号进行分解重构,得到能量谱特征向量,小波包变换不仅可以去噪,而且可以对高频段的故障振动信号进行分解细化,有效地提取故障轴承的信号特征。在轴承信号经过小波包变换处理后,分析轴承样本识别诊断问题。首先研究基于多分类支持向量机(SVM)的单步故障诊断方法:分析不同核函数下多分类支持向量机的诊断效果,确定最佳核函数。针对样本不平衡的情况,发现优化后的SVM分类器在该情况下诊断性能仍有不佳。在研究单步故障诊断方法后进一步研究基于结合异常检测算法的双步故障诊断方法:基于异常检测算法的特点,对小波包能量谱的特征信息进一步优化,确定先采用异常检测算法进行故障检测再采用SVM分类器进行故障分类的双步故障诊断模型。该方法可以将大量的正常轴承先一步检测出来,大大减少后续支持向量机的分类负担。最后建立单步与双步故障诊断模型的评价准则:根据实际工业中不同误判带来的损失是不相同的,创建一种结合工业实际损失与故障诊断误判概率的误判损失评价准则。该准则不仅更全面而且更加符合实际情况。基于该准则,针对样本不平衡、样本数量有限的情况,验证了基于异常检测算法结合支持向量机的双步故障检测模型的有效性和优越性。
[Abstract]:Rolling bearing is a common and very important universal part in mechanical equipment. Bearing failure often causes serious life and property safety problems. The research of rolling bearing fault detection is a hot topic in domestic and foreign industrial field. Affected by the environment of actual data collection, the problem of unbalance and limited sample size often occurs in practical engineering. In this paper, a new two-step diagnosis method, which combines anomaly detection algorithm with support vector machine (SVM), is proposed to solve these two problems. Firstly, the vibration characteristics of bearings are analyzed. After analyzing the vibration characteristics and the dynamic characteristics of the bearing, The fault diagnosis method of vibration signal is determined, and then the bearing vibration signal is extracted by wavelet packet energy spectrum feature extraction. The energy spectrum characteristic vector is obtained by decomposing and reconstructing the bearing signal by wavelet packet transform. Wavelet packet transform can not only denoise, but also decompose and refine the fault vibration signal in high frequency band, and extract the signal characteristics of the fault bearing effectively. After the bearing signal is processed by wavelet packet transform, First, the single-step fault diagnosis method based on multi-classification support vector machine (SVM) is studied, and the diagnosis effect of multi-classification support vector machine under different kernel functions is analyzed. Determine the best kernel function. In case of sample imbalance, It is found that the performance of the optimized SVM classifier is still poor in this case. After studying the single-step fault diagnosis method, the two-step fault diagnosis method based on the anomaly detection algorithm is further studied: based on the characteristics of the anomaly detection algorithm. The characteristic information of wavelet packet energy spectrum is further optimized. A two-step fault diagnosis model using anomaly detection algorithm and SVM classifier is determined. This method can detect a large number of normal bearings in one step. The classification burden of subsequent support vector machines is greatly reduced. Finally, the evaluation criteria for single-step and two-step fault diagnosis models are established: according to the loss caused by different misjudgments in actual industry, An evaluation criterion of misjudgment loss combining industrial actual loss and fault diagnosis misjudgment probability is established. The criterion is not only more comprehensive but also more in line with the actual situation. Based on this criterion, the sample is unbalanced and the sample size is limited. The effectiveness and superiority of the two-step fault detection model based on anomaly detection algorithm and support vector machine are verified.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH133.33

【参考文献】