基于PPCA的旋转机械故障识别算法

摘要

研究汽轮机的旋转机械故障有效提高识别率问题,故障诊断的难点在于传感器采集的离散故障信号无法清晰表现故障特性,存在相互耦合特点以及信号维数截断选择的缺乏标准性.为解决上述问题,根据故障振动信号非线性耦合的特性,提出了一种PPCA的旋转机械故障识别算法,算法采用概率主元分析(PPCA)对故障样本进行主元特征提取,得到较低维数的样本特征,克服了信号维数难以确定的难题,通过样本特征进行训练“一对-投票”的多类SVM分类器.实验结果表明,与传统PCA近邻法和常用SVM算法相比,改进方法不仅有很高的正确识别率,而且对自相关性较严重的类别样本也有较好的识别效果,为机械故障诊断提供了可靠识别方法.%The traditional rolling mechanic fault recognition algorithm suffered from the nonlinear, unstable and low proper recognition rate, which seriously degrade the reliability of the fault diagnosis system. In order to improve the rate of recognition, making use of the features of fault signal's character, this paper proposed an algorithm based on PPCA. After extracted the major features with PPCA, the major features with 25 dimensions can be gained, and then the result can be obtained depending on improved multiple SVM classify. The experiment results demonstrate that this method has the superiority of reliability and stability compared with the traditional PCA near neighbour and SVM algorithm, even if the samples have serious correlative character.