“…Various approaches to robust PCA have been proposed over the past few decades. − They can be classified into six groups: − (1) taking robust covariance estimators instead of the empirical covariance matrix, , (2) projection pursuit techniques, , (3) a combination of the ideas of groups 1 and 2, (4) robust subspace estimation, (5) elliptical and spherical PCA, and (6) low-rank matrix approximation . Although robust PCA methods have been widely investigated, there are few applications of robust PCA to process monitoring. , In practice, data collected in industrial processes frequently contain outliers. The faulty data, low-quality data, and data obtained during irregular operating phases (e.g., startup or shutdown periods) can all be viewed as outliers. , Therefore, any data-driven process monitoring scheme should take into account the possible presence of outliers in the training data.…”