Abstract
As the premise of ensuring safe production and improving economic benefits, process monitoring and fault diagnosis technology plays a vital role in process industry. Traditional industrial process monitoring and fault diagnosis mostly focus on fault detection and identification, and seldom involve quantitative assessment of faults. Considering that different dimensionality reduction methods may affect the monitoring effect, this paper proposes a fault detection and quantitative assessment method based on global and local information features to reduce the influence of data preference for dimensionality reduction model. Firstly, kernel principal component analysis and sparse locally linear embedding are used to reduce the dimension of the data, and the features preserving global and local structural information are obtained respectively. Secondly, finding the projection matrices through canonical correlation analysis makes the correlation between global features and global featured maximum, and the features transformed by the projection matrices are fused. Fusion features not only retain the difference between single features, but also eliminate the redundant information. Then, support vector data description is utilized to monitor the fused data to complete fault detection and quantitative assessment. Finally, a fault detection and quantitative assessment experiment on Tennessee Eastman process is conducted to demonstrate the effectiveness of the proposed method.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call