基于方差最大化旋转变换的K近邻故障诊断策略

Varimax rotation based K-nearest neighbor fault diagnosis strategy

为了提高FD-KNN针对潜隐变量在非线性和多模态过程中的故障检测能力,提出一种基于方差最大化旋转变换的K近邻故障检测与诊断策略。通过方差最大化方法建立旋转变换将原始数据变换到新的正交空间,在该正交空间中执行FD-KNN方法进行故障检测,并结合贡献图方法给出基于贡献图的故障诊断策略。通过一个非线性模拟实例,证明方法对潜隐变量故障诊断是有效的;同时,在典型非线性工业过程田纳西过程进行测试,与PCA、FD-KNN和PC-KNN方法进行对比,实验结果进一步证明了方法的有效性。

Aiming to improve the fault detection ability of FD-KNN in the nonlinear and multimodal process, this paper proposed a K-nearest neighbors fault detection method based on varimax rotation (Rot-KNN). Firstly, it implemented varimax rotation in observed data set to obtain an orthogonal space. Next, it implemented FD-KNN in the new orthogonal space to detect faults. At last, it proposed a fault diagnosis strategy based on contribution chart. A nonlinear simulation example and the Tennessee Eastman (TE) processes of a typical nonlinear industrial process prove that the method is effective for the latent variable fault diagnosis. The experimental results indicate that the proposed method outperforms the PCA, FD-KNN and PC-KNN.