The aim of the paper is to predict fabric properties from specified machine settings
and yarn parameters using ANNs and comparing the results obtained using the
traditional Regression Analysis techniques. In the weaving field, the loom settings
used as the input parameters for the network are the machine speed, shed height, shed
depth, backrest position, spring position and damping constant for different yarn
counts. Wool yarns were used on the projectile and rapier machines and cotton yarns
on the projectile and air-jet machines. Six trials were used for training the NN to
predict the woven fabric properties: e/in, p/in, gm/mt 2 , strength(kg) and
elongation(mm) for warp and weft and three trials for testing. For knitting, the rib and
interlock machine settings are the input tension, take down tension and stitch length
(mm) for different yarn counts and designs. Twelve trials were used for training the
NN to predict the knitted fabric properties: courses/10cm, wales/10cm, gm/mt 2 , stitch
length, lengthwise, widthwise shrinkage, bursting strength and abraison resistance and
four trials for testing. The same input trials were used to predict fabric properties
using regression analysis techniques. A comparison between the values of the
predicted woven and knitted fabric properties using both techniques is made. Results
show that the performance of NNs is better than Regression in both weaving and
knitting fields. Regression failed to predict fabric properties for air-jet and rapier
machines and most of fabric properties for projectile machine. NNs predicted all
fabric properties with a percentage relative error range of 0% and 8.14%. For knitting,
Regression failed to predict abraison resistance for rib and interlock fabrics,
wales/10cm and widthwise shrinkage for interlock fabric. NNs predicted all fabric
properties with a percentage relative error range of 0% and 32% except for the
widthwise shrinkage of interlock fabric which it failed to pr
By: A.Moussa,I.Ali,G.Sobhy, H.El Markaby, H.El Helw,D.El Khamy& A.El Geiheiny
Submit Date: 11/11/2002 20:00