Study On Prediction Model Of Supercritical Carbon Dioxide Solubility In Polymers

Solubility of supercritical carbon dioxide (ScCO2) in polymers is one of the mostimportant physicochemical properties, which has been used successfully in polymermodification, polymer synthesis, polymer blending, preparation of new material, andso on. Solubility studies consist of experiment and prediction, experimental studiesare difficult to implement under many restricted conditions such as time consuming,strenuosity, and material consuming, while most traditional prediction methods havesome shortcomings such as inaccurate, low adaptability, and the correlation is notgood compared with experimental. In this thesis, study on prediction model of ScCO2solubility in polymers based on particle swarm optimization (PSO), clustering method,artificial neural network (ANN), and the adsorption, diffusion, and burying theory iscarried out. The main work and achievements are as follows:(1) Study on improved particle swarm optimization algorithm.The improved PSO algorithm is introduced into the self-adaptive weightadjustment strategy to improve the convergence speed and accuracy, and introducedinto chaos theory to tune the acceleration coefficients and avoid prematureconvergence, hereinafter called CSAPSO algorithm. Examples show that theconvergence speed, accuracy, and diversity of CSAPSO algorithm are better.(2) Study on solubility prediction model based on CSAPSO algorithm.A solubility prediction model combined with CSAPSO algorithm and backpropagation ANN, called CSAPSO-BP ANN is proposed. BP algorithm with a strongability to search local optimum suffers weak ability to search global optimum, whileCSAPSO algorithm complements the BP algorithm. The model is developed usingCSAPSO algorithm to optimize the BP ANN connection weights. Examples showthat, the model has excellent prediction capability and better exploration researchcapability in the prediction example of untrained polymers.(3) Study on solubility prediction model based on clustering method.Three solubility prediction models, called by a joint name, CSAPSO-C RBFANN, are proposed. The connection weights, function center and spreads of radial basis function ANN are optimized by the CSAPSO algorithm and clustering method.Experiments show that the effect of using clustering method to tune function centerand spreads is obvious, and the model has better prediction performance with gooddevelopment research capability in the prediction example of trained polymers.(4) Study on solubility prediction model based on diffusion theory.Inspired by the essence of solubility and diffusion, mass transfer, and PSO, adouble populations PSO algorithm based on the particle diffusion energy anddiffusion probability is developed, then a solubility prediction model, calledDCSAPSO-BP ANN, is proposed. Experiments show that, in the prediction exampleof both trained and untrained polymers, DCSAPSO-BP ANN has superiorperformance and good stability.(5) Study on solubility theoretical calculation model based on the adsorption,diffusion and burying theory.Combined with interface adsorption, diffusion and burying theory, a solubilitytheoretical calculation model is developed based on two processes consist ofadsorption-diffusion and adsorption-burying, the relationship between buriedsolublity and time is estimated by a burying solubility growth factor. Experimentshows that the model describes the influence from solubility experiment conditions,and provides a reasonable reference for extrusion processing experiment.(6) Experimental study on extrusion process.Foaming materials extrusion process experiment under different temperature,pressure, screw speed and gas inlet flow is carried out, the correlation between theperformance of the products and the experiment parameters is analysed. It shows thatthe experiment parameters provided by prediction models are reasonable bycomparing the performance of the products, the solubility prediction models presentimportant references for practical experiment.