基于相对变换PLS的故障检测方法

针对偏最小二乘方法(partial least squares,PLS)在无量纲标准化处理后导致的特征值大小近似相等,难以获得代表性的潜变量等问题,提出了一种基于相对变换PLS( relative-transformation PLS,RTPLS)的故障检测方法.该方法引入马氏距离相对变换理论,通过计算采样数据之间的马氏距离,将原始空间数据变换到相对空间.然后在相对空间进行PLS分解,提取有代表性的潜变量,建立故障检测模型,实现采样数据的在线检测.通过对TE (Tennessee Eastman)过程故障和轧钢机系统力传感器故障的仿真实验验证了所提出方法的有效性和实用性.理论分析和仿真实验均表明,基于RTPLS的故障检测方法能有效地消除量纲的影响,提取具有更大的变化度和代表性的隐变量,增加故障检测的精度和实时性.

Fault detection method based on relative-transformation partial least squares

Aiming at the problems that the eigenvalues are approximately equal and the representative latent variables can not be extracted effectively after dimensionless standardization processing in partial least squares method(PLS),a fault detection method based on relative-transformation PLS(RTPLS) is proposed.The method introduces the relative transformation scheme based on Mahalanobis distance.Firstly,the RTPLS approach transfers the original data space into relative space through computing the Mahalanobis distance between sample data.Secondly,PLS approach is used to decompose the data in the relative space,extract the representative latent variables,build the fault detection model and implement on-line detection of the sample data.In the end,the proposed method was applied to detect Tennessee Eastman(TE) process fault and the force sensor fault of a rolling system in simulation experiments;the effectiveness and applicability of the proposed method are proved with the actual data simulations.Both theoretic analysis and simulation experiment demonstrate that RTPLS based fault detection method can directly remove the effect of dimension,effectively extract the latent variables with more variation degree and representativeness,and then improve the accuracy and on-line performance of fault detection.