Dubdub Ibrahim
Department of Chemical Engineering, King Faisal University, Al-Hassa 31982, Saudi Arabia.
Polymers (Basel). 2022 Jun 28;14(13):2638. doi: 10.3390/polym14132638.
Pure polymers of polystyrene (PS), low-density polyethylene (LDPE) and polypropylene (PP), are the main representative of plastic wastes. Thermal cracking of mixed polymers, consisting of PS, LDPE, and PP, was implemented by thermal analysis technique “thermogravimetric analyzer (TGA)” with heating rate range (5−40 K/min), with two groups of sets: (ratio 1:1) mixture of PS and PP, and (ratio 1:1:1) mixture of PS, LDPE, and PP. TGA data were utilized to implement one of the machine learning methods, “artificial neural network (ANN)”. A feed-forward ANN with Levenberg-Marquardt (LM) as learning algorithm in the backpropagation model was performed in both sets in order to predict the weight fraction of the mixed polymers. Temperature and the heating rate are the two input variables applied in the current ANN model. For both sets, 10-10 neurons in logsig-tansig transfer functions two hidden layers was concluded as the best architecture, with almost (R > 0.99999). Results approved a good coincidence between the actual with the predicted values. The model foresees very efficiently when it is simulated with new data.
聚苯乙烯（PS）、低密度聚乙烯（LDPE）和聚丙烯（PP）的纯聚合物是塑料废物的主要代表。由PS、LDPE和PP组成的混合聚合物的热裂解通过热分析技术“热重分析仪（TGA）”实现，加热速率范围为（5−40 K/min），分为两组：PS和PP的（1:1比例）混合物，以及PS、LDPE和PP的（1:1:1比例）混合物。利用TGA数据实现机器学习方法之一“人工神经网络（ANN）”。在两组实验中均采用了反向传播模型中以Levenberg-Marquardt（LM）为学习算法的前馈ANN，以预测混合聚合物的重量分数。温度和加热速率是当前ANN模型中应用的两个输入变量。对于两组实验，logsig-tansig传递函数的两个隐藏层中10 - 10个神经元被确定为最佳架构，相关系数几乎（R > 0.99999）。结果表明实际值与预测值之间具有良好的一致性。当用新数据进行模拟时，该模型能够非常有效地进行预测。
