Abstract: Not satisfied with the traditional monovariant statistical process control in tobacco strip processing, a failure monitoring approach based on relative variation analysis was proposed via analyzing the running characteristics of key processing equipments. The approach adopted attribution expansion to convert the three-dimensional data of batch, time and attribution characteristics into two-dimensional data and applied principal component analysis method to statistical modeling for failure monitoring under reference mode. According to the potential process fluctuation of alternate mode and its relative variation against the reference mode, the principal component subspace and residual subspace of monitoring model for reference mode were dissolved separately, then the statistical modeling for failure monitoring under alternate mode were carried out in the dissolved four subspaces. Off-line validation was conducted based on actual running data of equipments, the results showed that: comparing with traditional failure monitoring methods, such as pre-control diagram and process capability index, the proposed method revealed the dynamic variations of process variables and correlation between variables for different modes and batches in depth, and detected equipment failures timely and effectively.