Improvements to the T2 statistic for multivariate fault detection

Multivariate fault detection has been a requirement in various industrial systems to ensure safe operations and to improve product quality. The T2 statistic is a widely used fault detection index for monitoring unusual changes in the mean of multivariate data. In spite of its advantages, the T2 statistic also suffers from the amplification and masking effects, especially when the number of variables to be monitored is large. The reason of amplification effect is that the T2 statistic often uses an enlarged control limit that is determined in the space of all variables to detect the faults caused by only a small part of variables. The masking effect is because the faulty components of the T2 statistic are diluted by the normal (small) components to yield a T2 value within the control limit. Both amplification and masking effects reduce the fault detection capability of the T2 statistic. To address this problem, improvements are made to the T2 statistic in this paper. Based on an orthogonal decomposition of...