Multi-space PCA with its application in fault diagnosis

Traditional PCA method can detect “big” failures with obvious signs of abnormality effectively. But it does not seem to apply for failures with smaller signs drowned in the noise or “big” failures. Meanwhile, there is still not a clear and consistent explanation for the impact of the PCA subspace decomposition on the fault detection capability. In this paper, aiming at fault diagnosis with small signs, a method of multi-space principal component analysis is proposed based on the research on the effect of subspace decomposition on the capability of fault diagnosis, which is applied into the process monitoring. Case studies validate the effectiveness of the proposed approaches.