Semi-qualitative Temporal Episodes Prognosis for Process Supervision

This paper deals with a prognosis method that is used in an operation support system. The role of prognosis in a supervisory system is analysed. Then, a supervision method based on trend analysis is explained. First the on line extraction of semi-qualitative episodes from a noisy signal is described. The method uses a segmentation algorithm, a classification of the segments into seven temporal shapes and a temporal aggregation of episodes. Discontinuities are preserved, which is important in case of failure. Then, the estimation of relevant variables future evolution is described, which is based on the last semi-qualitative episode. A two tanks system is used as a pedagogical example to illustrate the method potential. An industrial application is then described shortly.