Given that numerous variables exist in industrial processes, it is difficult to make out what real relationships are among the variables. In the principal component analysis (PCA) approach, the loading matrix can reveal inner relations between variables and components, and different components contain different information about a certain variable. Therefore, this study proposes a novel method that respectively selects principal components (PCs) for each variable according to the loadings. The PCs containing more information about a certain variable are selected to construct the subspace for the corresponding variable, and then support vector machine data description (SVDD) technique is adopted to examine the variations in all subspaces. Additionally, a corresponding contribution plot is developed to identify the root cause. Finally, two case studies, a numerical example and the Tennessee Eastman (TE) system, demonstrate the effectiveness of the proposed method, with other PCA-based methods listed for comparison.