A Fault Monitoring Method Based on Transfer Subspace Learning

Fault monitoring is an important measure to ensure the safety, stability and automatic control of production process. The data-driven methods have been intensively researched and arisen as an active research area in process monitoring. But the current challenge is how to accumulate operational experience in production process and use it effectively. We propose a novel fault monitoring method based on transfer subspace learning, which maps data from both old process (labeled data) and new process (unlabeled data) to a common space where all data is of the same distribution and hence enables the new process to learn knowledge from an old one. Moreover, the proposed method uses a new strategy of low rank and sparse processing to retain the global and local characteristics of data. Finally, we have carried out the simulation analysis of fault monitoring for the Tennessee - Eastman process (Tennessee Eastman TE) and the grinding and grading process respectively, the simulation results verify the effectiveness and practicability of the proposed method.