Neural network based thermal protective performance prediction of three-layered fabrics for firefighter clothing

• Published in: Industria textilă (Bucharest, Romania : 1994) Vol. 70; no. 1; pp. 57 - 64
• Main Authors: MÜGE, DURSUN, YAVUZ, ŞENOL, ENDER YAZGAN, BULGUN, TANER, AKKAN
• Format: Journal Article
• Language: English
• Published: Bucharest The National Research & Development Institute for Textiles and Leather - INCDTP 01.01.2019
• Subjects: Artificial neural networks; Construction; Convective heat transfer; Fabric structures; Fabrics; Firefighters; Fires; Heat transfer; Laboratories; Multilayer perceptrons; Neural networks; Parameters; Performance prediction; Permeability; Protective clothing; Testing laboratories; Textile industry; Textile research; Vapor resistance; Warp; Water resistance; Weft; Weight
• ISSN: 1222-5347
• DOI: 10.35530/IT.070.01.1527

The firefighter protective clothing is comprised of three main layers; an outer shell, a moisture barrier and a thermal liner. This three-layered fabric structure provides protection against the fire and extremely hot environments. Various parameters such as fabric construction, weight, warp/weft count, warp/weft density, thickness, water vapour resistance of the fabric layers have effect on the protective performance as heat transfer through the firefighter clothing. In this study, it is aimed to examine the predictability of the heat transfer index of three-layered fabrics, as function of the fabric parameters using artificial neural networks. Therefore, 64 different three layered-fabric assembly combinations of the firefighter clothing were obtained and the convective heat transfer (HTI) and radiant heat transfer (RHTI) through the fabric combinations were measured in a laboratory. Six multilayer perceptron neural networks (MLPNN) each with a single hidden layer and the same 12 input data were constructed to predict the convective heat transfer performance and the radiant heat transfer performance of three-layered fabrics separately. The networks 1 to 4 were trained to predict HTI12, HTI24, RHTI12, and RHTI24, respectively, while networks 5 and 6 had two outputs, HTI12 and HTI24, and RHTI12 and RHTI24, respectively. Each system indicates a good correlation between the predicted values and the experimental values. The results demonstrate that the proposed MLPNNs are able to predict the convective heat transfer and the radiant heat transfer effectively. However, the neural network with two outputs has slightly better prediction performance