Prediction of the Styrene Butadiene Rubber Performance by Emulsion Polymerization Using Backpropagation Neural Network

e effect of the amounts of initiator, emulsi�er, and molecular weight regulator on the styrene butadiene rubber performance was investigated, based on the industrial original formula. It was found that the polymerization rate was increased with the increased dosage of initiator and emulsi�er, and together with replenishing molecular weight regulator will make the Mooney viscosity of rubber meet the national standard when the conversion rate reaches 70%. e backpropagation neural network was trained by the original formula and ameliorated formula on the basis of Levenberg-Marquardt algorithm, and the relative error between the simulation results and experimental data is less than 1%. e good consistency shows that the BP neural network could predict the product performances in different formula conditions. It would pave the way for adjustment of the SBR formulation and prediction of the product performances.


Introduction
Scienti�c researches try to increase emulsion polymerization monomer conversion and reaction rate by improving the formula on the basis of styrene butadiene rubber's good performance for many years [1,2].But because of too many formulas, the researchers face the advantages such as large amount of experiments, and the mutual in�uence of formulas when alter single additive amount or feeding mode and use orthogonal test.e experimental data cannot clearly re�ect the in�uence of each factor in formula [3].
e scienti�c workers are looking for a model to simulate the relationship between the polymerized styrene butadiene rubber performance and formula, what can provide reference to adjust formula and predict the product properties [4][5][6][7].e arti�cial neural network (ANN) is composed of a large number of neurons by communicating through the adjustable metric.It has many features, such as distributed information memory, massively parallel processing and the self-adapt learning function.e ANN is widely used in pattern recognition, information processing, intelligent control, system modeling and other �elds.Especially the Error Backpropagation Training (BP network) can approximate every continuous function and has good ability for nonlinear mapping.e layer number, processing elements number, learning coefficient, and other parameters can be settled by case.e BP network plays a very important role in many application �elds because of its �exibility [8,9].e effects of the amounts of initiator, emulsi�er, and molecular weight regulator on the performances of styrene butadiene rubber by original formula were investigated.Based on the original formula and optimized formula, the input vector of BP network train was the proportion of original initiator, emulsion, molecular regulator, and the conversion rate; the target vector was the combining styrene content and Mooney viscosity of the product of butadiene styrene rubber.e network was trained by using the Levenberg-Marquardt (L-M) model.e product properties of different formulas were simulated by the established model.Jilin petrochemical company.Activator solution (assay 0.6%) was prepared in lab.

Polymerization Reaction. e polymerization was reacted in 1.4 L reactor what is manufactured by
Songling chemical company, �antai, China, the polymerization �ow chart as shown in Figure 1.e solutions were prepared �rstly for accurate amount, and they were added to the reactor in quantization by order of sequence.e butadiene was calculated by mass �ow meter and added into the reactor.e oxidizer, emulsi�er, and molecular regulator were added by feed tank.e reaction was terminated by 1%  hydroquinone solution aer regular sampling.e samples were stored in ice bath for analysis.

3.�.�. �n��ence o� �nitiator A�o�nt on �oly�erization �ate.
e in�uence of different initiator amount on polymerization rate was investigated.e reaction time of 70% monomer conversion was chosen to show the rate of polymerization for avoiding the impact of sampling.e results were shown in Figure 2. Figure 3 shows that the combining styrene content did not mainly change, but the Mooney viscosity decreased with the increasing of amount of initiator.
e results show that the polymerization rate can be increased by increasing the amount of initiator signi�cantly.e polymerization time was reduced to 8.5 hours form 12 hours when the amount of initiator of original formula was increased by 40%.Every 10% increase of initiator amount, the reductions of reaction time were different, the higher amount of initiator, the less reduction of reaction time.Because the rate of polymerization is depended on the amount of free radical.e more initiator, the more free radical, and the collision probability will increase.

3.�.2. �n��ence o� ���lsion A�o�nt on �oly�erization �ate.
e in�uence of different emulsion amount on polymerization rate was investigated.e reaction time of 70% monomer conversion was chosen to show the rate of polymerization for avoiding the impact of sampling.e results were shown in Figure 4. Figure 5 shows that that the combining styrene content maintain stability, and the Mooney viscosity decreased with the increasing of amount of initiator, but the Mooney viscosity value is also far away the Chinese national standard what is 46 ∼ 58.
e results show that the polymerization rate can be increased by increasing the amount of emulsion signi�cantly.e polymerization time was reduced to 8.5 hours form 10 hours when the amount of initiator and emulsion of original formula were increased by 20%.Every 5% increase of emulsion amount, the reductions of reaction time were different, the higher amount of emulsion, the less reduction of reaction time.
�.�.�.�n��en�e o� �o�e���a� Re���ato� ��o�nt on �o���e�� ization Rate.e 10%, 16%, 20% molecular regulator was added to reactor aer the reaction 0 hour, 1 hour, 2 hour, respectively, when the amounts of initiator and emulsion of original formula were increased by 20%.e reaction was terminated when the monomer conversion was 70% for avoiding the impact of sampling.e Mooney viscosities of �occulated rubber were determined.e results were shown in Figure 6.e results show that the polymerization rate can be increased by increasing the amount of emulsion signi�cantly.e polymerization time was reduced to 8.5 hours form  10 hours when the amount of initiator and emulsion of original formula were increased by 20%.Every 5% increase of emulsion amount, the reductions of reaction time were different, the higher amount of emulsion, the less reduction of reaction time.e �gure shows that the higher amount of molecular regulator, the lower Mooney viscosity when the initial addition amount was 100% and the polymerization times were same� the Mooney viscosity decreased �rstly and increased subsequently with the delay of addition time when the additional amount were same, and it reached a nadir when the addition time was aer reacting for 1-2 hours.e Mooney viscosity of rubber can meet the Chinese national standard GB8655-88 when increasing the amount of initiator to 120% original, increasing the amount of emulsion to 120% original, the initial amount of molecular regulator was 100% original, and adding 20% more aer reacting 1-2 hours.
To sum up, the polymerization rate of butadiene styrene rubber can be accelerated by 30% via increasing the initiator and emulsion amount.e Mooney viscosity of 70% conversion rate rubber can meet the Chinese national standard together with adding additional molecular regulator.

e Foundation of BP Neural Network Model.
According to the product of butadiene styrene rubber, the actual situation was simulated predictively (the summary affection of formula condition to rubber was shown in Table 1).ere were two hidden layer BP neural network, the �rst layer is linear, re�ecting the in�uence of each condition on the product.e second layer is nonlinear, re�ecting the in�uence of each interaction factor, and both used the sigmoid logarithmic type function model.e linear transfer function was used for the output layer.e hidden layer had 10 neurons and the output layer had 2. e BP neural network is shown in Figure 7.
On the basis of original formula and optimized formula, the input vector of BP network train were the proportion of original initiator, emulsion, molecular regulator, and the conversion rate (such as the data of 1-16 in Table 1), the target vector was the combining styrene content and Mooney viscosity of the product of butadiene styrene rubber.e network was trained by using the Levenberg-Marquardt (L-M) [10] model.Figure 9 shows that the error of simulation result and experiment data is less than 1%.ey have good consistency.e correlation coefficient  2 of experiment and predicted values is 0.985.e comparison between the simulation results of 17#, 18#, and experiment is shown in Table 2, and the error

F 3 :
Effect amount of initiator on the combining styrene content and Mooney viscosity.

F 4 : 5 F 5 :
Effect amount of emulsi�er on the emulsion polymerization rate.Effect amount of initiator on the combining styrene content and Mooney viscosity.

F 6 :
of molecular weight regulator (hr) Mooney viscosity, ML 100 ∘ C 1+4 Effect of molecular weight regulator on the Mooney viscosity of rubber.

5 eF 9 :
SM binding rate of predetermination (%) e SM binding rate of experiment (%) (a) e SM binding rate comparison of BP neural network simulation results and experimental values.(b) e Mooney viscosity comparison of BP neural network simulation results and experimental values.
Neural Network Model.e predictive performance was achieved aer 1093 trains by Levenberg-Marquardt algorithms,   1  10 −4 .e relationship of error function and training number is shown in Figure 8.
Characterization.e total solid content (TSC, m%) was analyzed by Chinese SH/T 1154-1999 standard method.e Ca-H 2 SO 4 method was used for �occulation.e pH of emulsion was adjusted by H 2 SO 4 to 3, T 2: e comparison of experimental and simulation data.