The Journal of The Textile Institute Vol. 102, No. 8, August 2011, 668–674 ISSN 0040-5000 print/ISSN 1754-2340 online Copyright © 2011 The Textile Institute DOI: 10.1080/00405000.2010.514725 http://www.informaworld.com Ranking fibre and process parameters affecting thermal resistance of needle-punched blankets using neural network model Vinay Kumar Midha* Department of Textile Technology, National Institute of Technology, Jalandhar 144011, India Taylor and Francis (Received 1 September 2009; final version received 4 August 2010) 10.1080/00405000.2010.514725 In this paper, an artificial neural network (ANN) model has been designed to predict the thermal resistance of needle- punched blankets. Web-laying (parallel- and cross-laid), fibre fineness, fibre degree of hollowness, fabric weight, depth of needle penetration and needle punch density are considered as input parameters to predict the thermal resistance of needle-punched nonwoven blankets. In order to reduce the dependency of the results on a specific partition of the data into training and testing sets, three-way cross validation tests were performed, that is, total data were divided into training and testing sets in three different ways. The predicted thermal resistance correlated well with the experimental thermal resistance. The relative contribution of each parameter to the overall prediction of the thermal resistance was studied by carrying out a sensitivity analysis of the test data set. The results of sensitivity analysis show that web-laying is the most important input parameter, followed by depth of needle penetration, fabric weight, degree of fibre hollowness, needle punch density and fibre fineness. Keywords: artificial neural network; coefficient of concordance; needle-punched blanket; thermal resistance Introduction Thermal properties of the fabric are one of the most important aspects of comfort. Nonwoven fabrics are known to have better thermal resistance as compared with woven and knitted fabrics of similar weights. The properties of needled fabrics depend on the nature of component fibres and the manner in which fibres are arranged in the structure. Fibre properties (such as dimension, mechanical and surface) along with various machine and web parameters contribute to the structure that emerges from the needling operation (Midha & Mukhopadhaya, 2005). A proper understanding of the effect of different parameters on the thermal resistance is important for designing the fabric suitable for blan- ket. Midha, Alagirusamy, and Kothari (2004) studied the effect of fibre and process parameters on the thermal resistance of needle-punched blankets using statistical models. But the relative importance of each parameter in influencing the thermal resistance of needle-punched blankets is still unknown. Mathematical models developed by researchers show that thermal conductivity of a porous substance has a nonlinear relation with the thermo-physical parameters (Fayala, Alibi, Benltoufa, & Jemni, 2008). Moreover, the physical properties of needle-punched blanket fabric, such as thickness and weight, which affect thermal resistance are derived from basic fibre and process parameters. Under such conditions, the neural network model is a better alternative to statistical models for predicting the thermal resistance (Chattopadhyay & Guha, 2004). In various studies (Debnath, Madhusootha- nan, & Srinivasamoorthy, 2000; Ureyan & Gurkan, 2008a, 2008b), the statistical models were compared with the artificial neural network (ANN) models and the ANN models were reported to be better in their prediction performance as compared with statistical models. A neural network trained to predict an output from various input parameters also contains the knowledge about the relative importance of the input parameters, which may be very helpful in controlling the process and in design- ing end products. Few attempts have been made to extract such information from the neural network models. Jayadeva, Guha, and Chattopadhyay (2003) determined the most important fibre parameter influencing the yarn properties using the neural network model. Midha, Kothari, Chattopadhyay, and Mukhopadhyay (2010) estimated the relative importance of process and machine parameters affecting thread strength loss during sewing. In this paper, the ANN model is used to predict the ther- mal resistance of needle-punched blanket when the fibre and process parameters are given as input. The relative *Email: midhav@nitj.ac.in