【作者】 程平；

【导师】 王洪艳；

【作者基本信息】 吉林大学 ， 应用化学， 2007， 硕士

【摘要】 本论文研究了MATLAB神经网络工具箱在聚丙烯酸酯乳液和水性涂料性能预测中的应用。利用MATLAB神经网络软件对聚丙烯酸酯乳液和涂料的硬度、附着力、耐冲击性和反射率等性能进行了预测。考察了BA、MMA和St的组成对乳液性能的影响,且依次讨论了隐含层节点数、网络层间的传递函数和训练目标值等计算参数对预测结果的影响,得出了最佳的隐含层节点数、网络层间的传递函数和训练目标值。由钛白粉和轻钙粉与乳液构成了43个配方的涂料,对钛白粉与BA、MMA和St的不同组合对涂料性能的影响进行了讨论。对12个涂料样品的四种性能进行了预测,收到了较好的预测结果。实验证明用人工神经网络算法可以解决复杂涂料体系多种性能同时预测的问题,为涂料研究提供了一个新的简便的研究方法。更多还原

【Abstract】 Waterborne coating is composed of film forming matter(namely emulsion), pigment, solvent and various addition agents. The content of each component all may have effect on the whole coating system. So the coating system should be carefully designed, and the optimal scheme is chose out from all possible design systems. All these processes need masses of experiments. The neural network is usually introduced to assist design study for simplifying the design process of coating and studying the properties of coating more deeply.The coating is a complicated system. When writing the program of artificial neural network for predicting the properties of coating and emulsion, we often can’t make the calculated results satisfy the needed precision request. The software parcel of MATLAB provides a very mature neural network toolbox, which has many advantages such as being simple and easy to study it, high efficiency, calculated function and figure expression function, etc. In this paper, a back propagation(BP) network of MATLAB neural network toolbox was used to train and predict the properties of polyacrylate emulsion and related waterborne coating.1. Prediction of properties of polyacrylate emulsionTo study the effect of the composition of butyl acrylate(BA), methyl methacrylate(MMA) and styrene(St) on the properties of emulsion, a three-layer BP artificial neural network was chose to train and predict the two properties hardness and adhesion. The dosages of BA, MMA and St were as the input nodes, the number of hidden layer node was decided through experiment, the hardness and adhesion were as the output nodes. The appropriate hidden layer node number, transfer function and training goal value, and preferable predictive results were obtained through training and simulating the network. The average absolute value of relative errors between predictive and measured values of hardness is 5.90%, and the prediction accuracy of adhesion is 100%.The study results show that BP network of MATLAB being used to predict the hardness and adhesion of emulsion coating is feasible. The BP network may also be applied into prediction of other properties of emulsion.2. Prediction of properties of waterborne coatingA three-layer BP network was also used during the process of prediction of properties of waterborne coating. To study the effect of BA, MMA, St and titanium pigment on the four properties of coating—hardness, adhesion, impact resistance and reflection index, their dosages were as the input nodes; the number of hidden layer node was decided through experiment, and the properties of coating were as output nodes. Every two properties of coating were predicted under the decision of transfer function, training goal and max-epochs. Compared with the separated prediction of each property, the simultaneous predictions of every two properties were better. The results showed that when hardness and adhesion were simultaneously predicted, the average absolute value of relative errors between predictive and measured values of hardness is 7.78% and the prediction accuracy of adhesion is 91.67%; when and were simultaneously predicted, the average absolute value of relative errors between predictive and measured values of impact resistance is 1.38% and the average absolute value of relative errors between predictive and measured values of reflection index is 0.46%. In addition, the four properties hardness, adhesion, impact resistance and reflection index were also simultaneously predicted. The results showed it is no better than the results of simultaneous prediction of every two properties. The reason may be that the more the output nodes is, the bigger effect among of them. This may result in the bigger errors. And the fewer hidden layers may be another reason.The study results show that BP network can predict the properties of coating, and simplify complicated problem. So BP network fits for the modeling of multi- variable and complicated problem. Various addition agents also have effect on the coating properties, so the predictive result errors are slightly increased without considering the effect of those addition agents.更多还原