A distributed monitoring method for non-stationary processes with application to float glass production

Process monitoring is a vital part of ensuring the normal and stable operation of modern industrial systems. The distributed method considers the local adjacency between sub-processes and the information interaction between sequences, which is a more effective process monitoring strategy for industrial systems containing multiple sub-processes. This paper proposes a new one-way transfer distributed process monitoring method based on the idea of dynamic-static fusion for non-stationary processes. Firstly, the static and dynamic information in the local sub-processes are extracted by PLS-PCA and QSFA-SFA methods respectively, and the performance of the local monitoring model is further improved by using dynamic-static fusion. Then, it is assumed that the information between the sub-processes is transmitted sequentially when considering the information interaction, which is also consistent with the current situation of most large-scale process industries in assembly-line production. Local monitoring statistics and control limits are designed, and global monitoring is achieved by Bayesian fusion. Finally, actual data from a float glass production line from a company verifies the effectiveness of this framework, which is better than the traditional single monitoring method.