Industrial equipments have been widely used in modern industry, which indicates that the final product quality highly relies on the operational status of industrial equipments. Therefore, it is of high importance to detect the quality-related faults in equipments where non-linearity and non-stationarity are considered as two crucial characteristics in operating data. In this paper, high order and low order information in process data are discussed in details where a two-step latent variables extraction method is implemented using maximum mutual information and partial least square (PLS). In order to handle multimode process monitoring issue, a common-specific information strategy is designed. Combined with the high order and low order information extraction, the original data are further separated into four subspaces: common high, common low, specific high and specific low order information, respectively. Secondly, an online monitoring algorithm is developed, which helps identify mode switch or fault and improves the performance of multi-mode process monitoring. Finally, monitoring performance is further demonstrated on a real mechanical arm where the result shows the superiority of the proposed method.