Introduction

Transparent varnishes play an important role in printing technology. They are applied on top of the printed image in order to protect this layer but also to set its optical appearance, e.g., with regard to high gloss or matting. Whereas protective coatings have to be applied over the full print image, only parts of the image may be lacquered if special design effects should be achieved. In general, clear oil-based varnishes are printed as layers with a thickness of 0.5 to 3 g m−2 (note that the thickness in printing and coating technology is mostly given in form of the coating weight). However, printing is a complex process. Consequently, the thickness of printed layers may be influenced by numerous factors. Besides the chemical and physical properties of the used inks or varnishes, the coating thickness depends on various technical parameters such as the rotation speed of the duct roller, which supplies the varnish to the printing unit, the setting of the ink blades, which meter the amount of varnish on the duct roller, or the contact pressure between the rollers in the printing unit [1]. However, a constant and homogeneous thickness of the printed layers is requested for quality reasons.

Therefore, a powerful method for in-line monitoring of the thickness of the printed layers during the printing process is strongly required in order to ensure the requested quality of the printing products. For colored layers such as printing inks, the thickness can be measured by reflectance densitometry, which relates the intensity of the reflected light to the color density [1]. If the thickness of the layer is low enough, the color density can be regarded as a measure of its thickness. However, this indirect method cannot be applied for clear printed layers.

In principle, the thickness of clear coatings could be determined with white-light interferometry, which has been used, e.g., for the characterization of dielectric coatings [2, 3]. However, this technique requires homogeneous plane and transparent layers and substrates. Moreover, the refractive indices of both materials have to differ significantly in order to be able to separate the thicknesses of coating and substrate. In printing, paper is often used as substrate material. If polymer foils are used, their refractive index usually differs only marginally from that of the printed top coat which is polymer based in most cases as well. The interference pattern of a moving web with a non-uniform coating can change very rapidly, which may impede or even prevent quantitative analysis. Finally, rough surfaces degrade the interferences, which completely prevents measurements by white-light interferometry. Consequently, this method is poorly suited for routine in-line measurements of the thickness of clear printed layers.

Accordingly, there is currently no efficient analytical method, which allows in-line measurements of the coating weight or the thickness of such layers. Consequently, their coating weight can be only determined off-line by gravimetry at present. For a more efficient control of printing processes, it is necessary to develop a powerful method which is fast and sensitive enough for in-line measurements of the thickness of thin layers in the printing press. On the other hand, the instrumentation and the method have to be sufficiently rugged to withstand the specific conditions in a technical printing process, i.e., with regard to vibrations, temperature, dust, etc.

Near-infrared (NIR) reflection spectroscopy is well-known as a rapid, non-destructive and high-performance method, which has great analytical potential for both identification and quantification. Moreover, instruments can be developed according to a modular concept (i.e., with a probe head separate from the spectrometer and connected by fiber optics), which allows easy integration into complex production facilities. Hence, NIR spectroscopy has been developed to one of the most common techniques for process monitoring and control. Its applications range from agriculture and food production via pharmaceutics and chemistry to the separation of plastic waste [4–12]. In the chemical industry, the main application areas comprise the on-line and in-line monitoring of some fundamental chemical processes such as the processing of crude oil [13] but also the control of various polymerization reactions [14–18] as well as the processing of polymer materials [19, 20].

Functional coatings can be also made on the basis of polymerization reactions, e.g., by use of the UV curing technology [21, 22]. The in-line characterization of the resulting polymer layers is a challenging problem due to their low thickness and the high production speeds. However, it was shown in the past, that NIR reflection spectroscopy is able to predict both the coating thickness and the conversion of the varnishes after UV irradiation if qualified chemometric calibration methods are used for quantification [23–26]. Coatings with thicknesses in the range from about 5 to 50 μm were studied. The conversion was predicted for layers based on (meth)acrylates, epoxides and/or vinyl ethers. Moreover, it was shown that conversion and thickness (or coating weight) can be monitored simultaneously either by application of special calibration algorithms [27] or by use of several probe heads [28].

As indicated, the thickness of printed layers is even lower by about an order of magnitude in comparison with those coatings. Nevertheless, we wanted to ascertain if NIR spectroscopy has sufficient sensitivity to track the thickness (and later possibly also the conversion) of such layers at the high printing speeds, which are typically used at a printing press. Definitely, this ambitious aim could only stand a chance for realization by substantial support of the spectroscopic investigations by powerful chemometric data evaluation methods [29]. The final objective of this work was targeted on the in-line monitoring of the thickness of printed layers made from clear varnishes at an offset printing press.

Usually, printing inks and varnishes are complex formulations, which can be based on UV-curable systems, aqueous dispersions, or linseed oils. In a first investigation, we studied the potential of the NIR method to monitor the thickness of layers made from all of these various varnish systems [30]. Moreover, the influence of the substrate on the precision of the predictions was investigated [31]. In the present study, we will focus on conventional oil-based varnishes.

Linseed oil is one of the oldest raw materials in the printing industry, and it is still a prime constituent of printing inks and varnishes. Conventional oil-based printing varnishes contain up to 50 wt.% of linseed oil (or other vegetable oils) as reactive component. Moreover, they contain a resin (e.g., colophony) for film formation, a diluent (e.g., mineral oil), and further additives such as waxes, plasticizers, catalysts for oxidative drying (mostly cobalt compounds), etc. [1]. Linseed oil mainly consists of linolenic acid and linoleic acid. These high-molecular fatty acids dry according to an oxidative mechanism, which finally leads to chemical crosslinking [32–36]. The oxypolymerization reaction proceeds rather slowly in spite of the use of catalysts. However, very fast drying of the printed layers is required for high-speed printing processes since their surface has to be touch dry for further processing (e.g., stacking). Therefore, the drying induced by the chemical reaction is often combined with an additional drying process resulting from the rapid penetration of the solvent into the paper substrate. Both processes are supported by infrared and hot air dryers at the printing press.

In this study, only transparent varnishes were considered since the thickness of layers of pigmented inks can be determined by established densitometric methods. However, the influence of the surface properties of the printed layers was studied. Lacquers leading to different gloss levels are widely used in printing technology. However, NIR reflection spectroscopy is not only sensitive to the chemistry of a lacquer coat, but also on its surface structure, which leads to different scattering behavior of the printed layers. Therefore, printing lacquers leading to different gloss levels were included in the investigations in order to involve such effects in the calibration models.

Experimental

Materials and sample preparation

Preparation of calibration samples

Investigations were carried out with 4 commercial offset printing varnishes, which are based on linseed oils. They were kindly provided by Epple Druckfarben AG (Neusäss, Germany). Their formulations were set to achieve different gloss levels (see Table 1). Samples for laboratory investigations were made by printing the varnish on paper (115 g m−2) using a laboratory-scale printing machine (Printability Tester C1; IGT Reprotest, Amsterdam, The Netherlands).

Table 1 Gloss of oil-based printing varnishes on 115 g m−2 paper (coating weight 2.5 g m−2)
Full size table

After drying, the gloss of samples made from the various lacquers was determined with a reflectometer (micro-TRI-gloss; BYK-Gardener, Geretsried, Germany). At first, the gloss of all samples was measured at an angle of 60°, which was followed by measurement at different angles according to the different degrees of gloss. Measurements were carried out on samples with a coating weight of 2.5 g m−2. Results are summarized in Table 1.

Printing trials at an offset printing press

In-line monitoring trials were carried out at a sheet fed offset printing press (Speedmaster CD 74, Heidelberger Druckmaschinen AG, Heidelberg, Germany), which comprises five printing units for color offset printing and a special coating unit for the application of varnishes. The press is equipped with UV, infrared, and hot air dryers, which are mounted in the delivery system.

The printing varnishes were printed on the same glossy coated paper, which was also used for calibration. For viscosity reasons, they were applied at a printing unit. Samples with different coating weights were generated by systematic variation of the operating parameters of the printing press, i.e., the setting of the ink blades and the printing speed. The latter was set to 6,000 or 12,000 sheets h−1 corresponding to about 90 or 180 m min−1, respectively.

Analytical methods

NIR reflection spectroscopy

A process spectrometer Kusta 4004 P (LLA Instruments, Berlin, Germany) was used, which is based on a concave holographic grating and an InGaAs photodiode array detector. The detector with 256 elements covers a spectral region from 1,470 to 1,950 nm. A tungsten halogen lamp, which is mounted in the probe head, serves as light source. The probe head is linked to the spectrometer unit by a fiber-optic cable. Spectra were taken in the reflection mode. The spectrometer is described in more detail in ref. [37].

During process control at the printing press, NIR spectra of the printed sheets are recorded, while sheets are lying on the impression cylinder (see below). Therefore, a special metal reflector was developed for recording the calibration spectra in the laboratory, which exactly mimics a segment of this cylinder with respect to curvature and material. This setup ensures identical measurement conditions during calibration and process control, which is a crucial condition for achieving a high precision of the predictions during in-line monitoring. The probe head was mounted in a distance of 30 mm to the reflector and at an angle of 15° to the normal on its surface in order to avoid specular reflection of the probe light into the probe head.

NIR spectra were recorded immediately after printing, because the varnishes start to dry by an oxypolymerization reaction just after printing. This process leads to changes in the NIR spectra. For this reason, a well-defined time regime was required for the measurements in the laboratory, which had to correspond to the time flow at the press as close as possible. For each measurement, ten spectra (each of them averaged from 1,500 accumulations) were recorded, while the printed paper strip was moved across the metal reflector. In this way, the recorded spectrum is averaged across the surface of the sample, which contributes to compensate possible inhomogeneities of the thickness induced by the printability tester.

Figure 1 shows NIR spectra of layers of three printing varnishes with different gloss levels printed on paper. It is obvious from Fig. 1a (left) that the absolute reflectance of the glossy layers is significantly higher than those of the two matte coatings which is due to the lower portion of diffusely reflected light. Furthermore, the derivatives of the spectra plotted in Fig. 1b (right) clearly show significant differences between the spectra of the various varnishes in the region of the 1st overtone of the CH stretching vibrations.

Fig. 1
figure 1

NIR reflection spectra (mean spectra averaged from ten single spectra; left) and their 1st derivatives (right) of oil-based varnishes with different gloss levels printed on 115 g m−2 paper with a coating weight of 5 g m−2

Full size image

The extinction coefficients of fatty acids in the NIR region of the spectrum are rather low like those of most other organic compounds too. They were determined from transmission spectra of neat varnishes (using a 1-mm cuvette) and were found to be about 1 l mol−1 cm−1 at 1,720 nm.

In order to relate the spectral data to the coating weight of the printed layers, they were analyzed by multivariate calibration methods based on the partial least squares (PLS) algorithm [29]. Chemometric analysis was carried out with the KustaSpec software package supplied with the NIR spectrometer. Details about the calibration processes will be given in the “Results and discussion.”

For in-line monitoring, the NIR probe head was installed above the impression cylinder of the coating station of the printing press because this position was rather easily accessible and provided enough space for mounting. Great care was directed towards the precise alignment of the distance between probe head and cylinder as well as the tilt angle in order to ensure identical measurement conditions to those during calibration in the laboratory. As mentioned, the spectra were taken against the impression cylinder. The monitoring started after a forerun of 30 printed sheets in order to allow the press to achieve the full operation speed. NIR spectra were recorded continuously at a rate of 30 spectra/s. This resulted in three spectra per sheet at a printing speed of 12,000 sheets h−1.

Gravimetry

Reference data for the calibration procedures were obtained by gravimetry using a laboratory balance (resolution, 0.1 mg). The determination of the coating weight was preferred to the direct measurement of the thickness due to the higher precision and reproducibility of the former method. Moreover, the indication of the coating weight is common practice in coating and printing technology.

Furthermore, the coating weights of printed sheets obtained from the printing press were crosschecked by gravimetry as well. At first, the weight of blank sheets was determined, and the sheets were labeled. After printing, the weight of the printed layers on these labeled sheets was determined. This procedure also allowed a clear identification of the respective average value of the predicted coating weight obtained from the NIR spectra of the same sheet.

Results and discussion

Calibration procedures

Effect of the oxypolymerization reaction on the NIR spectra

The fatty acids, which constitute the main components of oil-based varnishes, dry by an oxypolymerization mechanism [32–36] in combination with the penetration of some ink components (e.g., mineral oils) into the paper in order to support drying. Nevertheless, the printed layers are still far from an equilibrium state upon leaving the printing unit. The chemical reaction further proceeds in the bulk, which is reflected in the NIR spectra. Figure 2 shows the 1st overtone region of the CH stretching vibrations in the spectra of the glossy varnish (after subtraction of the spectrum of the paper substrate and conversion to absorption spectra). It is obvious that the ratio of the bands at 1,760 nm (overtone of ν as CH2) and 1,720 nm (a combination band of ν as CH2 and ν s CH2) [38] changes with time which comes from the oxypolymerization reaction. Finally, these changes have only a moderate extent, but the strongest decline occurs during the very initial phase. Therefore, the spectral changes resulting from the beginning polymerization can have a destructive effect on the prediction performance of chemometric calibration models (regardless of the parameter to be predicted; see below), if this effect is not considered during calibration.

Fig. 2
figure 2

First overtone region of the CH stretching vibrations in the absorption spectra of the glossy lacquer (Epple 1667) after 0, 15, and 180 min, respectively. Dotted lines represent the results of the band separation (top). Time dependence of the ratio A 1,760 nm/A 1,720 nm (bottom)

Full size image

In order to overcome this problem, the calibration process has to be carried out in a time slot which corresponds as close as possible to the running time of the sheet in the printing press between printing unit and sampling point. Since this time period is in the order of a few seconds only (or less) if the printing press is operated at the usual printing speeds, an exact match cannot be achieved. However, NIR spectra to be used for the calibration procedures have to be taken immediately after sample preparation. In the present study, they were recorded just after application of the varnish with the printability tester (i.e., after about 15 sek).

Calibration models for individual varnishes

At first, calibrations models were developed for individual varnishes, i.e., for varnishes which lead to layers with a certain gloss. In order to develop a stable and powerful PLS calibration for a matte varnish (Epple 688), 35 samples with different coating weights (from 0.5 to 5 g m−2) were printed. The range was chosen to be broader than the usual thickness of printed layers in order to make the calibration models more stable. Subsequently, the NIR spectra were taken and the coating weight was determined. For the build-up of the calibration model, the data were split into a calibration (18 samples) and a validation (17 samples) set, and the PLS algorithm was applied to them. The optimum number of factors of the model was determined by use of the test set validation method. In order to improve the calibration model, different kinds of pre-processing were applied to the spectra, e.g., normalization, the first and the second derivative as well as a multiple scattering correction. These methods may help to reduce or to remove negative effects resulting from scattering, which is discussed in more detail in ref. [39]. Moreover, the spectral range was limited to 1,476 to 1,854 nm in order to eliminate the disturbing influence of humidity in the air, which strongly absorbs above about 1,860 nm.

For each version of the model, the root mean square error of prediction (RMSEP) and the coefficient of determination R² were calculated, which were finally used as criteria for the selection of the optimal model. The best model was selected with respect to the lowest RMSEP and the highest R². In case of the matte varnish, a pre-treatment of the spectra by application of the first derivative was found to result in the best model. The calibration curve based on this model is shown in Fig. 3.

Fig. 3
figure 3

PLS calibration curve for the coating weight of a matte oil-based varnish (Epple 688) printed on 115 g m−² paper

Full size image

Calibration models for the other varnishes were built up in a similar manner. Table 2 summarizes the data of the optimum model for each varnish. Although models with excellent parameters were obtained for all varnishes, data suggest that printed layers with matte appearance tend to result in models with slightly lower prediction errors and higher R². This effect might be attributed to the higher scattering power of such layers, which is assumed to have a positive impact on measurements carried out in diffuse reflection.

Table 2 Parameters of the optimum calibration models for the coating weight of various oil-based printing varnishes on 115 g m−2 paper
Full size table

In order to evaluate the predicting power of the created calibration models, they were used to determine the coating weight of independent test samples. The preparation of these samples and the chemometric evaluation of the measured data were carried out at a later date than the calibration process in order to ensure the long term stability of the measurement process and the calibration model.

At first, the effect of the progress of the oxypolymerization on the precision of the prediction of the coating weight was checked. Layers of the glossy varnish were printed on paper. Their coating weights were predicted from the NIR spectra using the specific calibration model for this varnish (see Table 2), which had been build-up from spectra, which were recorded just after printing. Spectra of the validation samples were taken immediately after printing (i.e., after about 15 sek) and after various time intervals. Figure 4 compares the predictions obtained from the spectra of the fresh samples with those recorded 30 min after printing. It is obvious that a close correlation of the predicted data with the reference data from gravimetry (i.e., a correct prediction) was only obtained after 15 sek, whereas an offset was observed at longer times. For example, the coating weight is underestimated by about 0.15 to 0.2 g m−2 after 30 min. This offset is due to the changes in the spectra induced by the polymerization reaction. Figure 2 reveals that the main part of the changes in the spectra occurs during the first 30 min. Consequently, Fig. 4 ones again clearly emphasizes the necessity of a well-defined time regime during the calibration process which has to correspond to the time flow at the printing press.

Fig. 4
figure 4

Prediction of the coating weight of independent samples of the glossy varnish (Epple 1667) on paper just after printing (15 sek) and after 30 min, respectively, using a calibration model based on freshly printed samples (15 sek; see Table 2)

Full size image

Next, the effect of gloss on the prediction of the coating weight was investigated. In case of the matte varnishes, layers from the two different lacquer formulations (Epple 688 and Epple 1131) were printed on paper, and it was tested if their coating weights could be predicted with the same calibration model, i.e., with the calibration of Epple 688 shown in Fig. 3. In Fig. 5, the prediction results are plotted vs. the corresponding coating weight of each sample, which had been obtained by gravimetry.

Fig. 5
figure 5

Prediction of the coating weight of independent samples of the two matte varnishes (Epple 688 und Epple 1131) on paper using the calibration model generated with samples of Epple 688 only (see Fig. 3)

Full size image

It can be clearly seen that the coating weight of the layers of both matte lacquers is predicted with high precision from the NIR spectra (RMSEP688 = 0.16 g m−2, RMSEP1131 = 0.17 g m−2), even though samples of only one varnish (Epple 688) had been used for calibration. However, the gloss of the layers made from both lacquers is very similar (see Table 1), which indicates similar surface properties (e.g., with respect to roughness).

The prediction of the coating weights of the two other varnishes provided results with similar precision, when the corresponding calibration for the specific gloss level was used for analysis of the data. However, it would be of considerable practical interest, if calibration models still predict correct results, in case they are used to determine the coating weight of layers with different surface structures. Therefore, it was investigated if calibration models, which were developed for layers with a specific gloss level, could be used for the quantitative analysis of the thickness of layers with different gloss. An example is given in Fig. 6: it compares the predictions of the coating weights of matte printed layers made from Epple 1131 using either the specific calibration established for this varnish or the calibration, which was developed on the basis of samples of the high-gloss varnish (Epple 1667).

Fig. 6
figure 6

Prediction of the coating weight of samples of a matte varnish (Epple 1131) on paper using the calibration model for a glossy lacquer (Epple 1667). For comparison, values predicted with the appropriate calibration for the matte lacquer are shown

Full size image

Whereas the calibration model developed for the matte lacquer predicts the coating weight with high precision (RMSEP = 0.16 g m−2), the values obtained with the model created for the high-gloss varnish show a considerable offset of the predictions from the actual coating weights (RMSEP = 0.78 g m−2), i.e., this model strongly underestimates the coating weight. Evidently, the offset is related to the different roughness of the layers of both varnishes. The roughness of a surface influences its reflection behavior. In particular, the ratio of diffuse and specular reflection strongly depends on the surface structure. Rough lacquer surfaces lead to a high percentage of diffusely reflected radiation, whereas the diffuse part of the signal of glossy layers presumably mainly originates from the surface of the subjacent paper.

In the past, numerous works dealt with theoretical modelling of the effects of surface roughness on the reflection properties of these surfaces [40–44]. Generally, scalar Kirchhoff theory was used to describe scattering on structures with scales in the wavelength region (microroughness), whereas a facet model was used for roughness features much greater than the wavelength (macroroughness). Many industrial important materials such as coated and printed papers exhibit such two-scale rough surfaces. In order to derive analytical expressions for the gloss, Gaussian height distributions of the surface irregularities were assumed. The decrease of the specular reflectance due to the increasing surface roughness was found to be a function of the root mean square height σ and the surface slope distribution of the pattern. In particular, the gloss decreases with (σ/λ) cos θ, but increases with λ/L c, where λ is the wavelength, θ is the angle of incidence and L c is the height–height correlation length. A detailed discussion on the quantitative relationship between specular and diffuse reflection from rough surfaces is given in [42].

Since the NIR measurements are carried out in diffuse reflection, it is obvious that the intensity and the spatial distribution of the scattered light, which is received by the detector, depend on the reflection properties of the surfaces of the lacquer coats. Consequently, the roughness of the surface is reflected in the NIR spectra. However, the results shown in Fig. 6 clearly illustrate that the prediction potential of a PLS calibration model, which is based on coatings with a certain gloss level, is seriously overstrained by the extent of the variation in the spectra, which results from samples with a quite different gloss. Hence, this constellation leads to mispredictions of the coating weight.

In commercial printing facilities, frequent changes of printing varnishes with different degrees of gloss may occur. If NIR spectroscopy should be used for process control, this requires powerful all-purpose calibration models. In order to overcome the detrimental effect of the surface roughness on the prediction performance, the variation of the gloss has to be included into the calibration model. Accordingly, PLS calibration models were developed, which contain spectra of samples with different gloss levels.

Universal calibration model for various gloss levels

In order to create an all-purpose calibration model for varnishes with different gloss levels, the spectra, which were collected for the individual calibrations of the glossy (Epple 1667), the semi-matte (Epple 1277), and one of the matte varnishes (Epple 1131), were merged in combined calibration and validation sets (each with 45 samples). For each degree of gloss, the samples were evenly distributed to both sets. The calibration procedure was carried out in the same manner like for the single calibrations: the spectral data beyond 1,854 nm were eliminated to exclude the effect of atmospheric humidity, the spectra were pre-treated by various methods, and finally the PLS algorithm was applied to them. The calibration model with the best paramters (RMSEP, R²) was obtained by normalization of the spectra followed by application of the first derivative. Figure 7 shows the calibration curve of this model.

Fig. 7
figure 7

Universal PLS calibration function for the determination of the coating weight of oil-based varnishes with different degrees of gloss (printed layers on 115 g m−2 paper)

Full size image

In spite of the variety of the gloss of the printed layers, the parameters of the universal calibration model are very similar to those of the single calibration models specific to a certain gloss level (see Table 2). In particular, the error of prediction is in the order of 0.15 g m−2, which corresponds to the precision of the model for the glossy lacquer.

Similar to the procedure with the single calibration models, the performance of the combined chemometric model was evaluated with independent test samples of the three varnishes. Additionally, the coating weights of samples of the second matte varnish (Epple 688), which was not included in the calibration model, were predicted with this model as well. Results are summarized in Fig. 8.

Fig. 8
figure 8

Prediction of the coating weight of independent samples of printed layers with different degrees of gloss using the universal calibration model shown in Fig. 7. No samples of Epple 688 had been included in this calibration model

Full size image

The data clearly show, that the coating weight of samples made from different varnishes—and accordingly with different degrees of gloss—can be predicted with high precision by use of the universal calibration function (RMSEP = 0.16 g m−2). No offset is observed between the results for different varnishes (compare with Fig. 6). Even the coating weight of printed layers made from Epple 688 could be predicted with similar precision, i.e., with an error of only 0.165 g m−2. The results prove that the interfering effect of the gloss on the prediction performance of a PLS calibration model can be completely suppressed, when the variation of the gloss is included in the design of the model. For the practical application of NIR spectroscopy for in-line monitoring of printing processes, this result is of great relevance since it considerably reduces time and effort required for calibration. This approach could be also used to compensate the effect of other parameters (e.g., the variation resulting from the use of different paper substrates [31]), which have an impact on the correct prediction of properties of the coatings to be measured.

In-line monitoring of printing processes

The capability of the universal calibration model to predict the correct thickness of printed layers with different gloss under in-line conditions was tested at a sheet fed offset printing press. Printing trials were carried out at two different line speeds. The speed of the ink duct roller was kept constant at 50 % of its maximum speed. In order to get printed layers with a different coating weights, the relative opening of the ink keys, which control the amount of varnish on the duct roller, was varied (labeled as Opening in Figs. 9, 10, and 11). In this way, layers with coating weights between 1.2 and about 4.0 g m−2 were obtained.

Fig. 9
figure 9

In-line monitoring of the coating weight of layers of the high-gloss varnish printed on 115 g m−2 paper. The prediction of the coating weight from the NIR spectra (grey circles) was based on the universal calibration model. Off-line data from gravimetry (black rectangles) are given for comparison

Full size image
Fig. 10
figure 10

In-line monitoring of the coating weight of layers of the semi-matte varnish printed on 115 g m−2 paper. Coating weights from gravimetry are given for comparison

Full size image
Fig. 11
figure 11

In-line monitoring of the coating weight of layers of a matte varnish (Epple 1131) printed on 115 g m−2 paper. For comparison, coating weights determined by gravimetry are shown

Full size image

Since the NIR probe head was mounted above the impression cylinder of the coating unit, in-line spectra were recorded immediately after printing. All spectra recorded from the same sheet were averaged before further processing. After each change of the ink keys or the printing speed, the spectra of the next 30 printed sheets were discarded in order to allow the press to achieve steady state conditions. Then, a series of numbered sheets with defined weights was printed. After the end of the printing trial, the coating weights on these sheets were determined by gravimetry. The numbering of the sheets allows for an exact assignment of the spectral data recorded in-line to specific sheets and their coating weights determined off-line, which enables a qualified comparison of both data sets. Results of a typical printing trial recorded during the application of a high-gloss varnish are given in Fig. 9. In-line and off-line data are shown together. Each change of the settings of the ink keys is marked by an arrow.

Figure 9 shows a close correlation between the coating weights determined off-line and the predictions from NIR spectroscopy. The error was found to be only slightly higher than that obtained during the investigations carried out in the laboratory (standard deviation, 0.17 g m−2). Obviously, the increase of the printing speed does not have a negative impact on the measurement process. Moreover, the findings clearly prove that the universal calibration model is able to precisely predict the coating weight of varnish layers under the conditions of a technical printing process.

In order to substantiate this promising result and to prove the potential of the universal calibration model for varnish layers with different gloss levels, similar investigations were also carried out with the semi-matte and one of the matte varnishes. The results are given in Figs. 10 and 11.

The results in the two figures demonstrate that the developed measuring method is also able to quantitatively monitor the coating weight of layers with various matte surface structures. The comparison with the reference values shows that the precision of the predicted data is rather high, i.e., the mean errors in the two figures are 0.15 and 0.16 g m−2, respectively. The scattering of the predicted data is apparently somewhat lower than in Fig. 9, which is in accordance with observations made during the calibration process in the laboratory (see Table 2). Probably, the stronger diffuse reflection of the various matte varnish layers leads to a slightly improved quality of the spectra.

Nevertheless, the in-line monitoring experiments in Figs. 9, 10, and 11 have clearly shown that the universal calibration model is a powerful tool for the prediction of the coating weight with a broad range of different gloss levels, which does not only work in the laboratory, but also under process control conditions. This finding is an important result because it makes the in-line monitoring method more versatilely usable and helps to strongly reduce the efforts required for calibration, since otherwise separate calibrations would have been to be developed for each specific degree of gloss. Even with this kind of all-purpose calibration, an accuracy of the measurements in the order of 150 mg m−2 was achieved, which is roughly the same precision, which is achieved with specific calibration models. However, although this is an imposing accuracy with respect to the spectroscopic measurement process, it is on the other hand the minimum level of performance, which is required for the monitoring of printing processes.

Conclusions

In this paper it was demonstrated, that the coating weight of printed layers of conventional oil-based offset printing varnishes can be determined with NIR spectroscopy. Investigations were carried out in a range from 0.5 to 5 g m−2. The spectral data were analyzed by means of multivariate chemometric methods. However, the studies have also shown that surface properties such as the gloss strongly affect the prediction of the coating weight. In order to minimize or even to exclude this effect, universal calibration models were developed, which contain spectra of layers with a broad range of different degrees of gloss. The test of these calibration models resulted in rather low prediction errors of about 0.16 g m−2, which is only marginally higher than the errors obtained with the specific calibration models developed for each of the varnishes with a certain gloss level (0.12 to 0.14 g m−2).

The predicting performance of the universal calibration models was tested under realistic process control conditions at a commercial sheet fed offset printing press. The coating weight of printed layers with different gloss levels was determined. Regardless of the specific degree of gloss, an excellent correlation between the predictions from the NIR spectra recorded in-line and the reference data determined off-line by gravimetry was observed. Thus, the results clearly show that the thickness or rather the coating weight of transparent varnish layers can be determined in-line during the printing process with high precision by the use of NIR reflection spectroscopy.