The Bidirectional Intelligent Optimization Model Of Carbon Fiber Drawing Process

Carbon fiber is a new kind of polymer material and widely used in military civil and other aspects with its high modulus,low density,high specific performance,no creep and other excellent characteristics.So various countries in the world is rapid developing of carbon fiber products.However,the instruments and equipment for producing carbon fiber are wide various,the preparation process is complex,the production process is diverse and the product accuracy requirement is high.So the preparation process is full of frustrations.Once the development of new products,even rich experienced employees also need a lot of time to constantly try and repeatedly modify.So it is urgent to reveal the complex relationship between the process parameters and the properties of carbon fiber products.Therefore,in this paper,we established the bidirectional intelligent optimization model between the production process and product performance index in the drawing process of carbon fiber precursor based on machine learning and intelligent algorithm to solve the problem.The main contributions of this paper are as follows:?1?First of all,after consulting a large number of Chinese and English literature,the production process parameters are selected as air drawing ratio r₁,coagulation drawing ratio r₂,hot water drawing ratio r₃,boiling water drawing ratio r₄,dry heat drawing ratio r₅ and saturated steam drawing ratio r₆.Moreover,the performance indexes of the carbon fiber product quality include: linear density,carbon fiber strength and elongation at break.Then,this paper introduces the source of data,the purpose and the process of data pretreatment.?2?Next,the forward prediction data model is established based on the least square support vector machine?LS-SVM?algorithm.And using the particle swarm optimization?PSO?algorithm to optimize the two parameters C and ? of LS-SVM.Because of PSO-LS-SVM consuming a lot of time,so the PSO algorithm is optimized according to the power law.In the PSO algorithm,the generation of sub particles is improved.The proposed algorithm could speeds up the optimization speed and saves time.?3?Then,the inverse optimization model is established based on the Elitist Non-DominatedSorting Genetic Algorithm?NSGA-??on the basis of the forward prediction model to provide production parameters in the production of new carbon fiber products.Due to the uncertainty of mutation process of NSGA-? algorithm,the global optimal operator is selected as the guide operator to guide the mutation process in order to get the diversity of the solution and to reduce the error of the optimization target.This paper uses K-means clustering algorithm and the inverse proportion distribution to select the global optimal operator,and introducing the average-distance-amongst-points in population initialization process to indirectly control the distribution of population initialization.The proposed algorithm can not only improve the diversity of the solution,but also reduce the solution of the optimal target error.?4?After that,the bidirectional optimization visualization platform is established based on GUI MATLAB according to the forward prediction model and the inverse optimization model.The user can choose historical data,forward prediction model or the inverse optimization model according to needing,and then enter the corresponding value.The system will train and learn based on the user's choice of historical data to predict or optimize according to the user input values to achieve human-computer interaction.Finally,this paper discusses the optimization of the research,and puts forward the new research content aiming at the content of this paper.