Process Singular Value Recognition Based PAC Approaches To Multi-Mode Process Monitoring

Conventional PCA method is susceptible to intractable difficulties inmonitoring a process experiencing mode transitions. In this context, thispaper presents a modified recursive PCA approach which combineson-line process singular value recognition algorithms and contributes tomode transition process monitoring. Specifically, an enabling algorithmof process singular value recognition is proposed, which accounts foraccurately identifying ongoing mode transition of a process beforelaunching recursive PCA algorithms to monitor the transition phases. TEprocess is employed as a case study, which demonstrates the benefits ofthe contribution.In practical industrial processes, changes of process variableset-points usually force to produce multi-mode process running, whereconventional PCA monitoring could inevitably produce false alarms. Inthis context, this paper presents a novel multi-mode process monitoringmethod which combines online process singular value recognition andrecursive PCA monitoring technology. Therein, two problems are effectively solved. First, transitions of a multi-mode process can beidentified using singular value recognition algorithms. Second, differentstages of the multi-mode process can be monitored by means ofappropriate PCA techniques. The main contribution of this thesis could beaddressed as follows.1. An enabling data-driven online singular value recognitionalgorithm is introduced which could effectively identify different stagesof a multi-mode process.2. An improved PCA method is suggested to monitor multi-modeprocesses containing transition and smoothly running stages. Specifically,the smoothly running stages can be monitored using conventional PCAwhile the transition phases can be dealt with using recursive PCA. Thisnovel idea ensures a better adaptability of PCA methods to multi-modeprocesses.3. TE process is employed as a test-bed for verification of thecontribution, where, a multi-mode process is managed to happen and thenovel PCA approaches are launched accordingly, achieving satisfiedresults. In addition, a process operation monitoring system is developedby using VB language, which helps make inroad into industrial sectors.