The principle and ability of de-correlation of Principal component analysis (PCA) are explained in this paper. SNR between neighboring components are introduced as a criterion for data compression and components cut-off. The influence of mixture matrix condition number on PCA is analyzed. When using PCA in mechanical system, the influences of excitation and measurement locations are analyzed, and a study on the influence of source properties is presented. Experiments are carried out on a beam rig. PCA is used to identify source number of several cases include correlated white noise sources and uncorrelated white noise sources. The results show that when observation number is equal to or larger than source number, accurate prediction of the number of uncorrelated excitation sources in a multiple input multiple output(MIMO) system could be obtained by PCA. Based on this framework, the pre-processing could make the blind identification more reliable.