Viscosity determination of amino resins during synthesis using near-infrared spectroscopy

Introduction

The conventional synthesis process of a urea-formaldehyde (UF) resin is composed of two steps: basic methylolation and acidic condensation. Formation of methylolureas, namely mono-, di-, tri- and tetra-methylolureas, occurs in the first stage (Dunky 1998). In the condensation stage, linear and branched polymer with varied molecular masses is formed due to reactions involving methylol end groups, urea and formaldehyde are still present in the reaction mixture (Mehdiabadi et al. 1998).

In the industrial process, in order to monitor reaction progress, samples are taken periodically from the reactor during the condensation stage and their viscosity is measured after cooling. When the desired viscosity is attained, the reaction is stopped by increasing pH and cooling. However, the method for offline viscosity determination is labour intensive and introduces a significant delay in the measurement. The objective of our work is to develop a calibration method for viscosity estimation, based on measurement of FT-NIR spectra acquired by an optical fibre.

Recently, the capabilities of FT-NIR spectroscopy have been exploited in several industries including: pharmaceutical (Ciurczak 1987), biomedical (Ellis and Goodacre 2006), petrochemical (Macho 2002), food (Paradkar 2002), and polymer synthesis (Minopoulou et al. 2003). Regarding polymerisation systems, published works focus mainly on the characterisation of the final product and not on the control and monitoring of the reaction stages (Santos et al. 2005).

In the particular case of UF resin synthesis, NIR spectroscopy has been used for the analysis of intermediates and to check the conformity of raw materials by Dessipri et al. (2003). Kasprzyk and co-workers (2001) used the technique for qualitative and quantitative analysis of liquid MUF resins, particularly for the determination of melamine content. This study also identified a number of relevant spectrum bands in UF resins.

To implement this technique it is necessary to appeal to chemometrics (e.g. data-based modelling techniques). However, being an indirect method, the NIR needs calibration to allow determination of a set of parameters that minimises the prediction errors of the model (Conzen 2006).

In NIR, the relevant information is contained in a small spectral range. The quality of the spectra in the selected frequency window is usually influenced by several factors such as background noise, baseline drift and light-scattering. These may cause unwanted effects in the development of a calibration model and consequently produce biased results (Pizarro 2004). It is therefore important to take into account not only the selection of the frequency window, but also of the preprocessing method, since they can minimise the influence of those adverse effects and enhance the model's prediction capability. Usual preprocessing methods include use of first and second derivatives, multiplicative scatter correction and vector normalisation (Pizarro 2004).

The main objective of this study is to develop a method for determining the viscosity of UF resins using NIR spectroscopy, and compare it with conventional methods.

Materials and methods

Results and discussion

Kasprzyk and co-workers (2001) identified the regions between 7502–6098 and 5000–4246 cm⁻¹ as being the most important for UF resins, and assigned them to the most significant vibration from important groups, namely amide, amine and hydroxyl.

Calibration models for viscosity were developed taking into account these spectral regions, individually or in combination (7502–6098 and 5000–4246 cm⁻¹) and using different preprocessing methods: first derivative, multiplicative scatter correction (MSC) and standard normal variate transformation (SNV). The results are summarised in Table 1.

OPEN IN VIEWER

Table 1

PLS-R model using wave number 5000–4246 cm⁻¹

Preprocessing method	Calibration model		Validation model
	R 2	SEC	R 2	SEP
First derivative	92·52	27·3	73·31	56
MSC	93·64	25·4	62·89	66
SNV	93·07	26·3	82·42	45·4
First derivative with MSC	93·62	25·2	62·68	66·2
First derivative with SNV	95·02	22·3	80·81	47·5

From the graphical analysis of Fig. 1, and after evaluation of the statistical parameters for the regions considered, it appears that the methods that use only the spectral region 5000–4246 cm⁻¹ (Fig. 1b) yield the lowest errors. This region, as described by Kasprzyk et al. (2001), is associated with vibration bands from starch and amino groups, vibrations from methylol hydroxyl groups (5000–4878 cm⁻¹) and CH vibration bands (4600–4400 cm⁻¹). This is in agreement with the known formation of methylene ether (−CH₂−O−CH₂−) and methylene (−CH₂−) linkages during the condensation stage (Dunky 1998). The best results are obtained for Brookfield viscosities between 50 and 300 cP.

OPEN IN VIEWER

Figure 1

Relationship between error obtained with PLS-R models and true (experimental) viscosity for UF resins, using different spectral regions

Table 1 shows the statistical parameters of all preprocessing methods for this region. The and correlation values were used to quantify the predictability of each method (Büning-Pfaue 2003). The general goal was to obtain low values of standard error of calibration (SEC) and standard error of prediction (SEP) and high values of and (Conzen 2006). The results indicate that the model developed using first derivative with vector normalisation (SNV) as preprocessing yields the lowest values of SEP and SEC and the highest values of and .

Conclusions

For UF resins synthesised by the conventional process, the spectral regions 7502–6098 and 5000–4246 cm⁻¹ were adopted as being the most relevant for implementation of calibration methods. This was in agreement with the known molecular vibration peaks present in these polymeric structures. The models developed took into account each of these regions individually or in combination. It was found that the use of only one spectral region (the region from 5000 to 4246 cm⁻¹) yields the lowest errors.

For each spectral region, different preprocessing methods were tested, resulting in different calibration models. Some methods gave high correlation values, validating their applicability.

This method allows evaluation of the viscosity of the reaction mixture during the condensation step for UF resin using FT-NIR analysis. This simplifies the measurement procedure, avoiding the need to remove samples from the reactor, and reducing operator intervention. Future work will involve application of this approach to online control of a pilot reactor during the synthesis of a UF resin.