Study On Soft Sensor Technology For Quality Control Of Polypropylene Production Process

Melt index is an important quality variable in polypropylene production process. However, it is unmeasured on-line directly. Soft sensor technology provides a feasible way to solve the problem, and contributes to realize quality control of polypropylene production process. According to the analysis of process characteristics and mechanisms, soft sensor modeling of melt index by utilizing process information included mechanistic knowledge and industrial data was studied. The instantaneous melt index and cumulative melt index models were built, and grade transition strategy was studied based on cumulative melt index model.The primary research content is as follows:(1) Soft sensor technology was analyzed and summarized comprehensively. The multi-model fusion modeling method was reviewed, and it was divided into data driven fusion modeling method and hybrid modeling method. The design ideas, model structures and research statuses of the multi-model fusion modeling method were presented in detail, and future directions of research development and engineering application were discussed. The research status of soft sensor modeling of polyolefin melt index was commented under above classification of soft sensor modeling method, and it provides as foundation for subsequent study.(2) According to the characteristics of polypropylene production process, the whole production process was divided into steady grade production process and grade transition process. And, differences and relationships of instantaneous property and cumulative property of melt index were analyzed. Three kinds of instantaneous melt index models were built respectively by mechanistic modeling method, Elman neural network modeling method and ensemble Elman neural networks modeling method. These models were applied to develop soft sensor of melt index in steady grade production process. The simulated results shown that the ensemble Elman neural network model which overcome the deficiency of mechanistic model’s nonlinear ability and instability of Elman neural network model can achieve better predictive accuracy.(3) The cumulative melt index model of dual-loop reactor was developed by mechanism analysis. And, the model was simplified by reasonable assumption. A self-adaptive ensemble Elman neural network model was proposed to estimate unknown instantaneous melt index in the simplified model, and the hybrid model of cumulative melt index base on serial structure was built. Grade transition strategy of polypropylene production process was studied based on the hybrid model. Firstly, hydrogen concentration’s influence on melt index was investigated. Secondly, based on one-step transition strategy, that the appropriate overshooting of hydrogen was beneficial to shorten transition time was found. Finally, a multi-stage transition strategy closed to real production process was proposed. The simulated result shown that the transition time can be shorted effectively by this strategy.The research on melt index modeling and grade transition strategy research can promote quality monitoring and control in polypropylene production process, and provide the guidance for grade transition operation. The proposed multi-model fusion modeling method can offer the references for research and development of soft sensor technology on process industry, and be helpful to improve the operation level of process industry, and satisfy the high-throughput and low-consumption demand of production process, thereby enhance the competitive power of process industry.