No abstract

Many difficulties are encountered when analyzing hard-to-identify underlying causes of disturbances in distillation processes. Direct model-based fault diagnosis methods often prove to be time-consuming because they iteratively solve the direct model to obtain representative disturbance parameters. In this paper, an inversion method is proposed for identifying fault causes in the field of distillation diagnosis by mapping measured data against disturbance types. The proposed method consists of an offline phase and an online phase. In the offline phase, both normal and abnormal samples are simulated based on a simplified dynamic mechanistic model, and they are used to establish an inversion model for each disturbance with regard to each characteristic representation quantity using a genetic algorithm (GA). In the online phase, neural network, negative selection algorithm (NSA), and support vector machine (SVM) algorithms are used to determine the disturbance type, and the disturbance quantity is obtained by solving the corresponding inversion model. The proposed method is applied to high purity binary distillation column and benzene-toluene distillation column test cases. Case studies demonstrate that the inversion model-based diagnosis method can effectively identify disturbances and thus provide theoretical support for the application of quantitative fault diagnosis technology in the chemical industry.