One of the steps in the manufacturing of synthetic fibres involves using finishing oils to ensure proper lubricity and adherence between fibres, and also the absence of static electricity. Choosing an appropriate oil and dosage are essential with a view to ensuring effective subsequent processing and use. The aim of this work was to develop a fast method for determining the different finishing oil content in acrylic fibres by use of near infrared spectroscopy (NIRS) in conjunction with partial least-squares regression (PLSR). The high similarity between the NIR spectra of finishing oils led us to assume that a single calibration model might allow determine the oil content. However, the inability to quantify accurately different finishing oils by using a sole calibration model, constrain to the prior classification of the fibres coated with the different finishing oils. Two different pattern recognition methods were used: supervised independent modeling of class analogy (SIMCA) and artificial neural networks (ANNs). However, the low contribution of the finishing oil to the NIR spectrum for the fibre sample, the high similarity between the NIR spectra for the different oils and the substantial contribution of the linear density of the acrylic fibre to the spectrum precluded correct classification by SIMCA; on the other hand, ANNs provided good results. By constructing appropriate PLSR models for the different types of finishing oils, these can be accurately determined in acrylic fibres. © 2002 Elsevier Science B.V. All rights reserved.
|Journal||Analytica Chimica Acta|
|Publication status||Published - 22 Jul 2002|
- Acrylic fibres
- Artificial neural networks
- Multivariate calibration
- NIR Spectroscopy