Classification of amino resins and formaldehyde near infrared spectra using K-nearest neighbors. Issue 5 (October 2019)

• Record Type: Journal Article
• Authors: Gonçalves, M; Paiva, NT; Ferra, JM; Martins, J; Magalhães, F; Carvalho, L
• Is Part Of: Journal of near infrared spectroscopy. Volume 27:Issue 5(2019)
• Page Start: 345
• Page End: 353
• Publication Date: 2019-10
• Subjects: Near infrared -- amino resins -- near infrared -- K-nearest neighbors -- formaldehyde; Near infrared spectroscopy -- Periodicals; 543.5
• DOI: 10.1177/0967033519861008
• Languages: English
• ISSNs: 0967-0335

Abstract

Amino resins are synthetic adhesives that can be divided into three major types: urea–formaldehyde (UF), melamine–urea–formaldehyde (MUF), or melamine–formaldehyde (MF). When less than 5% of melamine is added to a UF resin, the resin is called a melamine-fortified UF (mUF) resin. The extensive application of these resins in wood-based products is due to their many advantages: ease of use, strong bonding, resistance to wear and abrasion, heat resistance, and relatively low price. Several near infrared (NIR) models have been developed for this type of adhesives and have been used in industrial plants. However, the NIR spectroscopy is sensitive to the type of resin (UF, MUF, MF, or mUF) and even to the synthesis process, therefore different NIR models must be constructed per resin basis. This work presents two methods: (a) a method to distinguish the NIR spectra of formaldehyde from the NIR spectra of amino resins, and (b) a method to classify the NIR spectra of amino resins by class of resin. The method for the separation of formaldehyde from amino resins achieved 100% correct classification for the dataset used. This method was based on defining a baseline cutoff for the NIR spectra at which there were no amino resins bonds overlapping formaldehyde bonds. For the classification of amino resins, this work used the methodology of K-nearest neighbors, up to 91 neighbors, and principal component analysis, up to 10 principal components. The best classification method obtained an accuracy of 96.1% and can be used industrially to automatically select the most suitable NIR model for amino resins, helping to reduce the time taken for an NIR analysis and automatically preventing the wrong selection of NIR models by an operator.