Neural Network for Predicting Thermal Conductivity of Knit Materials

The major aim of comfort research is to find the comfort temperature for an individual or group. This subjective property can be evaluated by means of thermal conductivity as a physical characteristic of fabric. This phenomenon depends on many fabric parameters and it is difficult to study the effect of ones without changing the others. In addition, the non-linear relationship of fabric parameters and thermal conductivity handicap mathematical modelling. So a neural network approach was used to predict the thermal conductivity of knitting structure as a function of porosity, air permeability, weight and fiber conductivity. Data on thermal conductivity are measured by experiments carried out on jersey knitted structure. INTRODUCTION Garment comfort has become an important purchasing criterion sought by consumers. Thus, textile companies try to find a compromise between materials and styles in order to produce comfortable clothing. Comfort is one of the most important attributes of textiles used in clothing. It is influenced by fabric factors, fiber conductivity, environment and human factors etc. One major aim of comfort oriented research is to find the optimal heat exchange between wearers and clothing systems adapted to an individual or a group of individuals. This property is related to the thermal conductivity of fabric. The thermal conductivity study of textile material is important to characterize the phenomenon of energy transfer. Fourier, in 1822 studied the heat transfer and found proportionality between heat flow and the temperature gradient of the surrounding layers: . ( ) grad T φ λ = − (1) Where, λ is the thermal conductivity [W.m.K]. The mathematical models developed by several researchers (Maxwell 1904; Vary 1952; Kunii and Smith 1960; Woodside and Messmer 1961; Zenner and Schlûder 1970; Bauer et al. 1991; Bogaty and Collar 1987; Fricke 1993) show that the relation between the thermal conductivity of porous surrounding and its thermo-physical properties are non-linear. Many studies have been conducted to analyze the relationship between various fabric parameters and comfort properties by using statistical methods (Chen 2003; Farnworth 1983; Hoge 1964; Zhang et al. 2002). The most common problems faced in statistical modelling are the non linear relationship of different fabric parameters with the thermal property. In addition, most of the fabric parameters (thickness, weight, porosity, permeability, etc.), derived from basic fabric specifications such as yarn and fabric characteristics, are closely related to each other. Hence, it is difficult to study the effect of one parameter without changing the other. So, to predict the thermal conductivity by considering the influence of all fabric parameters at the same time, a new system is required. In this case, artificial neural networks can be successfully used. The neural networks have been used to predict various comfort related properties such as human sensory perceptions and overall comfort index (Park et al. 2000; Wong et al. 2003 2004; Hui 2004). In this work, a neural network approach was used to predict the thermal conductivity of knitting structure as function of porosity, air permeability, yarn conductivity and weight per unit area. Data on thermal conductivity were obtained by experiments realized in laboratory. In developing the ANN model several configurations were evaluated. Optimal neural network was selected with one hidden layer and one output: thermal conductivity. The networks were trained with training data set and then tested with untrained values. Thermal conductivity values obtained from network were compared to actual values obtained from instruments. MATERIALS AND METHODS A total of 81 samples of knits were taken for the study, of which 80% were used for training and the others for testing of the network. These samples are composed with different: Matter: Cotton, Cotton/PES, Wool/Acrylic, Wool/Polyamide. Yarn Count: 18 to 306 Tex. Number of threads: 2 to 6 threads. Gauge: E5,E7, E12, E20, E24. Table I presents the Statistical Analysis of input parameters (Porosity, Air permeability, Weight per unit area and yarn conductivity). TABLE I. Statistical values of input parameters Input parameters Mean value Standard deviation Minimum value Maximum value Yarn thermal conductivity (W/K.m) 0.0899 0.053 0.03 0.289 Surface weight (g/m) 268.9 91.06 134.6 530 Porosity (%) 55.01 21.67 6.48 86.44 Air Permeability (L/m/s) 1802 973.41 387 5586 The output parameter: thermal conductivity of these samples is calculated by the apparatus of adiathermic property illustrated by Figure 1. FIGURE 1. Experimental device of the adiathermic property determination This property called K is calculated according to Eq. 2 2