Unsupervised Monitoring of Flocculation Processes based on Recurrence Theory

Continuous monitoring of abnormal conditions during operation is an important requirement to increase the quality, and efficiency of chemical processes, and to optimize operating costs. In this study, fault diagnosis of abnormal conditions is considered for flocculation processes, which due to the complexity of these processes requires more attention. To this end, an unsupervised learning method is developed to diagnose the faults in chemical processes based on recurrence analysis. This method consists of two stages of pre-processing and clustering. The pre-processing stage is carried out by transferring the time series from time space to state space and converting the data into a two-dimensional recurrence plot. Quantitative parameters of recurrence analysis can be extracted from this plot. Then, in the clustering stage, the density-based spatial clustering of applications with noise (DBSCAN) method was used for clustering different operating conditions and diagnosing faults. By comparing the results with conventional methods, such as independent component analysis (ICA) and Kernel ICA (KICA) it was found that the developed method is more powerful and shows the best performance. Application of this method was illustrated throughout a laboratory scale flocculation of silica particles in water. An on-line non-invasive sampling method was used for monitoring the size distribution of particles with a dynamic image analysis sensor.