Fault Diagnosis Research On Penicillin Fermentation Process Based On FDA

The batch process is an important mode of production in modern process industry. Due to its flexibility, it is widely used in fine chemical engineering, food, pharmacy and other industries. Penicillin fermentation is a complex biochemical reaction system, and the stability of the fermentation process determines the efficiency of penicillin fermentation. Therefore, it has a major practical significance to do research on the fault detection and the fault diagnosis of penicillin fermentation processes, as well as major referential significance for relevant applications of other biomechanical reaction processes.Fisher Discriminant Analysis (Fisher Discriminant Analysis, FDA) is a dimension reduction technique, which is widely used in the field of pattern recognition. When FDA is used for modeling, it uses the historical data under both normal working conditions and fault conditions, thereby providing its good fault diagnosis ability. Based on the research on penicillin fermentation processes and the access to amounts of literature, this paper completes the establishment of a window-based multi-FDA model for the fermentation process of penicillin and makes a fault diagnosis for the fermentation process of penicillin.This paper proposes a window-based multi-FDA model method, that is, on the premise of understanding the dynamic characteristics of the batch processes, selecting the appropriate time window, and using data close to some sampling points for the establishment of the FDA model. Compared with the multi-model FDA method, this method not only takes full advantage of limited fault data resources, but also takes process dynamic characteristics into account. Aiming at the fermentation process of penicillin, in contrast a number of research results, we choose the following eight variables for modeling and fault diagnosis:air flow, mixing power, substrate flow, dissolved oxygen concentration, substrate concentration, carbon dioxide concentration, pH value and temperature. As for the selection of windows, different principles to choose windows are adopted for the characteristics of different stages of the fermentation process of penicillin. Then, an FDA model is built on in each window, so as to obtain the optimal discriminant matrix of the window model and the discriminant score vector of each type. During on-line fault diagnosis, the required fault diagnostic model is determined according to the time periods of data, and then fault diagnosis is made based on the principle of the minimum Euclidean distance of the discriminant score vectors, so as to estimate the state of the current working conditions of production.Finally, through the simulation studies in the fermentation process of penicillin, it is proved that window-based multi-FDA model method has significantly improved fault diagnosis performance compared with the multi-model FDA method.