论文部分内容阅读
由于一维时间序列重构G)P算法提取故障征兆的鲁棒性较差,特别是受实测信号中敏感声噪的影响,提出一种基于去趋势波动分析(FDA)与核主分量分析(KPCA)融合的降噪方法,提出基于多重分形谱的特征值提取算法,通过伪相图技术判定权重因子阈值,优选参量并比较原G)P算法缺陷,结合转子系统3种常见故障,分析该法提取特征值的稳定性和准确性,结果证明诊断效果好。
Due to the poor robustness of one-dimensional time-series reconstructed G (P) algorithm to extract fault symptoms, especially the sensitive noise caused by the measured signal, a new method based on the trend analysis (FDA) and kernel principal component analysis KPCA) fusion, the eigenvalue extraction algorithm based on multifractal spectrum is proposed. The thresholds of weighting factors are determined by the pseudo-phase diagram technique, the parameters are compared and the defects of the original G) P algorithm are compared. Combining the three common failures of the rotor system, Method to extract the stability and accuracy of eigenvalues, the results show that the diagnosis is effective.