Analysis and prediction of air permeability of woven barrier fabrics with respect to material, fabric construction and process parameters
Analysis and prediction of air permeability of woven barrier fabrics with respect to material, fabric construction and process parameters
复制标题
DOI：
10.1007/s12221-017-7241-5
复制
发表时间：
2017-10-01
影响因子：
2.5
通讯作者：
Cherif, Chokri
中科院分区：
文献类型：
--
作者：
Malik, Samander Ali;Kocaman, Recep Tuerkay;Cherif, Chokri
透气性是传统织物以及技术织物（例如防护服、过滤器以及用于气囊和降落伞的织物）的重要性质之一。在外科纺织品的情况下，透气性是热生理舒适性的有效量度。本研究的目的是分析PES阻隔织物和开发之间的相关性和影响材料，结构和工艺参数。不仅纱线、织物和织机参数的单独影响，而且这些影响因素中的少数或几个之间的潜在复杂相互作用对织物的孔隙率和渗透性产生显著影响。人工神经网络（ANN）是合适的工具来映射这种复杂的输入-输出关系，因为直接的解析解是不可能的。采用Levenberg-Marquardt算法和贝叶斯正则化支持相结合的方法训练前馈神经网络模型。根据输入变量的个数，对三种人工神经网络模型进行了优化。据观察，用所有选定的输入训练的模型在测试数据上提供了有希望的结果，即，R-2=0.985，平均绝对误差为1.8%。为了消除过度拟合的任何疑问，还对选定的最终模型进行了10%交叉验证。此外，为了研究优化后的神经网络模型中输入变量的相对重要性，还进行了秩分析。研究结果表明，人工神经网络可以用来调整织机，织物和纱线的参数，快速定制阻隔织物的透气性取决于要求，没有尝试和错误。
Air permeability is one of the important properties of conventional as well as technical fabrics such as protective garments, filters, and fabrics for airbags and parachutes. In case of surgical textiles, air permeability is an effective measure of thermo-physiological comfort. This study is aimed to analyze PES barrier fabrics and to develop correlation between permeability and influential material, construction and process parameters. Not only the individual effects of yarn, fabric and loom parameters but also the underlying complex interactions between fewer or several of these influencing factors exert significant influence on fabric porosity and permeability. Artificial neural network (ANN) is the suitable tool to map such complex input-output relationships, since a direct analytical solution is not possible. Feedforward neural network models were trained with combination of Levenberg-Marquardt algorithm and Bayesian regularization support incorporated in backpropagation. Based on the number of input variables, three ANN models were optimized. It was observed that the model which was trained with all selected inputs delivered promising results on test data, i.e., R-2=0.985 and mean absolute error of 1.8%. To eliminate any doubt of overfitting, 10 % cross-validation was also performed for selected final model. Furthermore, to investigate the relative importance of input variables in the optimized ANN model, the rank analysis was also carried out. Research outcomes reveal that ANN can be used to tailor barrier fabric permeability depending on the requirements quickly without trials and error by adjusting loom, fabric and yarn parameters.