Study On Cascade Modeling Of Spinning Process In Fiber Production

Polyester fibers have good properties and are widely used in the manufacture of industrial products as well as in daily life.The spinning stage of polyester fiber is the last step of the whole process of polyester fiber,and it determines the performance of polyester fiber.Due to the long production line in the spinning stage,and it is extremely vulnerable to external interference,the practicality of data-driven modeling for the whole is insufficient.Therefore,the spinning stage is subdivided to establish data-driven models for each sub-segment,and the global model is obtained by cascade method,which is more suitable for the spinning process.Since the sub-model cascading leads to the cumulative transfer of errors,establishing an error compensation model between the sub-models can effectively compensate the prediction error of the sub-model and improve the predictive ability of the cascaded model.In this thesis,the polyester fiber spinning process is segmented,the data-driven model and error compensation model of each sub-segment are established by machine learning and intelligent optimization algorithm,and the sub-models are cascaded.The prediction of the intermediate material structure,parameters and final performance indexes of the spinning process have certain guiding significance for the production of polyester fiber.The main research work of this thesis is as follows:1.Based on the theoretical analysis and the experience of field engineers,the polyester fiber production spinning process is divided into three sections,namely,windless section,blowing section and natural cooling forming section.The input variables of the windless section are spinning speed,spinning temperature,blowing speed and blowing temperature,and the output variables are polyester fiber orientation,tension,filament radius and filament speed at the entrance of the blowing area.The input variables of the blowing section are polyester fiber orientation,tension,filament radius and filament speed at the entrance of the blowing section,and the output variables are polyester fiber orientation,tension,filament radius and filament speed at the outlet of the blowing section.The input variables of the natural cooling forming are blowing area at the exit of polyester fiber orientation,tension,radius of strips of silk speed,the output variables are the four performance indexes:elongation,breaking strength,EYS1.5 andCV_EYS1.5.2.Data-driven model are used to establish each subsection.The particle swarm optimization?PSO?is prone to fall into premature convergence,the concentration selection mechanism in the immune algorithm is added to the particle swarm update,and the immune particle swarm optimization?IPSO?is applied to optimize the parameters of the extreme learning machine?ELM?.IPSO has combined the advantages of the two intelligent optimization algorithms,and the convergence and predicted accuracy are improved.Since the cascade of sub-models will lead to the accumulation and transmission of errors,it is necessary to establish an error compensation model among the sub-models to compensate the output of the data-driven model in the sub-segment.The error compensation model is a radial basis function?RBF?neural network,and the predictive ability of the cascade model is improved.3.Aiming at the deficiency of using point value data in polyester fiber spinning process modeling,interval data is used to establish the relationship model between production parameters and performance indexes.The gradient momentum term is added into the weights update the upper and lower bound of interval radial basis function neural network?IRBFNN?,get the interval radial basis function neural network based on gradient momentum factor?IRBFNN-GMF?is established.IRBFNN-GMF is applied to the interval data-driven modeling of each sub-section of the spinning process.At the same time,an error compensation model is established between the cascaded sub-segment data-driven models to compensate the upper and lower bounds of each sub-segment data-driven model.The error compensation model is IPSO-ELM,and the predicted performance of the cascade model is improved after the error compensation.4.Design and build a prediction platform for polyester fiber spinning process,which mainly includes three visual interfaces:data selection,point data performance index prediction and interval data performance index prediction.It achieves the selection of training data,point value and interval value performance index prediction,and supports human-computer interaction.The predicted results can be visualized by setting input parameters.It can input the set parameters to display the prediction results visually.