基于时频特征和PCA-KELM的液压系统故障诊断

为了解决液压系统泄漏、堵塞和气穴等多类型故障下特征提取和模式识别困难的问题,提出基于时频特征和PCA-KELM的液压故障智能诊断新方法。首先利用统计分析和总体平均经验模态分解方法,构造高维混合域初始特征向量,从不同特征指标、不同分析角度对不同种类液压故障进行表征和刻画;然后通过主成分分析对多维初始特征向量进行降维和特征二次提取,将高维相关变量转化为低维独立的主特征向量;最后利用PCA主元构造的主特征向量输入核极限学习机网络中,实现故障类型的识别。实验结果表明,混合域初始特征向量能全面准确地描述故障特征,PCA提取的主特征向量摒弃了冗余信息且简化了分类器结构,KELM网络诊断速度快、分类准确率高。

Fault diagnosis of hydraulic system based on time-frequency characteristics and PCA-KELM

To tackle the different faults in hydraulic system, e.g. leakage, blockage, cavitation and so on, which it makes difficult to make feature extraction and mode recognition, a new fault diagnosis approach based on time-frequency characteristics and PCA-KELM was proposed. Firstly, high-dimensional mixed domain feature vectors were obtained by statistical analysis and ensemble empirical mode decomposition (EEMD) so as to describe different faults from various feature indexes and domains. Then, principle component analysis (PCA) was used to reduce the dimensionality of feature vectors, which turns high-dimensional variables into low-dimensional variables. Finally, these final features were input into kernel extreme learning machine (KELM) to identify faults. The results show that the fault characteristics are described comprehensively and accurately by mixed domain feature vectors. PCA can eliminate superfluous data and simplify classifier structure. KELM can increase diagnosis speed and meet the recognition rate requirement.