Skip to Main Content
The paper presents our experience in using sparse principal components (PCs) (Zou, Hastie and Tibshirani, 2006) for visualization of gearbox diagnostic data recorded for two bucket wheel excavators, one in bad and the other in good state. The analyzed data had 15 basic variables. Our result is that two sparse PCs, based on 4 basic variables, yield similar display as classical pair of first two PCs using all fifteen basic variables. Visualization of the data in Kohonen's SOMs confirms the conjecture that smaller number of variables reproduces quite well the overall structure of the data. Specificities of the applied sparse PCA method are discussed.
Date of Conference: 18-21 Sept. 2011