Detection of adhesives with Specim hyperspectral cameras

Glues and adhesives are widely used in many industrial applications, from packaging and construction to electronics and aerospace, to mention a few.

Glue is expensive, and using just the right amount is crucial as it can save on costs, increase production, and reduce waste.

Applying too little or too much glue is often the cause of quality failures, such as pop-opens. Applying too much glue can also stain the final product.

Using the right amount of glue in the right place should be monitored carefully. However, detecting glue can be challenging since most glues are transparent, and standard vision systems based on RGB cameras cannot see them. Hyperspectral cameras offer a more appropriate solution.

This application note demonstrates the effectiveness of Specim hyperspectral cameras for detecting different types of adhesives applied on cardboard and rubber (synthetic).

Key concepts:

NIR = NEAR INFRARED (900 – 1700 nm)

SWIR = SHORT WAVE INFRARED (1000 – 2500 nm)

MWIR = MID-WAVE INFRARED (2700 – 5300 nm)

PCA = PRINCIPAL COMPONENT ANALYSIS

PLS DA = PARTIAL LEAST SQUARE DISCRIMINANT ANALYSIS

ARTICLE

Figure 1. Photos of the three types of glue (left) and the support materials (right).

We measured the samples with three Specim hyperspectral cameras that cover different wavelengths:

Specim FX17, a near-infrared (NIR) camera with spectral range from 900 – 1700 nm
Specim SWIR, a shortwave-infrared (SWIR) camera with spectral range from 1000 – 2500 nm
Specim FX50, a middleware-infrared (MWIR) camera with spectral range from 2700 – 5300 nm

NIR ANALYSIS OF GLUES WITH SPECIM FX17 CAMERA

The samples were first measured with the Specim FX17 camera, on the Specim 40 x 20 LabScanner (see Figure 2). The spectral resolution is 8 nm over the full spectral range of 900 – 1700 nm. The pixel size on the image was ca. 0.3 mm.

Figure 2. Specim FX17 hyperspectral camera and 40 x 20 LabScanner

A PCA was caried and a PLS DA model was applied on the data of wet and dry glues. It is noteworthy that on both cardboard (at the top) and rubber (at the bottom), the glues were applied from left to right. The glue 1 on the left, glue 2 in the middle, and glue 3 on the right. Also, for each glue on each support, the “contaminated” spot on the top refers to the dry glue, whereas at the bottom spot, the glue was wet.

Figure 3 shows the results of the analysis. It appears that i) the Specim FX17 camera can detect glues better on cardboard, and ii) it seems to detect the glue 3 (Epoxy) but is less sensitive to the other types of glues. Especially the glue 1, which it did not detect at all on rubber.

An explanation for this could be the quantity of glue applied. Epoxy is much thicker, and we used more of it. Its detection is, therefore, more accessible.

The results also show that the distinction between dry and wet glue is not very accurate with the FX17.

Figure 3. PCA visualization with components 5 (red), 3 (green), and 4 (blue) and PLS-DA predictions accessed with Specim FX17.

SWIR ANALYSIS OF GLUES WITH SPECIM SWIR CAMERA

The samples were then measured with the Specim SWIR camera, on the Specim SisuCHEMA (see Figure 4). The spectral resolution is 12 nm over the full range of 1000 – 2500 nm. The pixel size on the image was ca. 2.3 mm.

Figure 4. Specim SWIR hyperspectral camera and SisuCHEMA

The Specim SWIR camera can be seen as an extension of the Specim FX17. Therefore, the observations made on NIR spectra are valid in addition to the finding made with the SWIR spectra.

A PCA was caried and a PLS DA model was applied on the data of wet and dry glues. Figure 5 shows the results of the analysis. As a reminder, we saw that the FX17 i) detected glues better on cardboard and ii) was most sensitive to glue 3 (Epoxy), medium sensitive to glue 2, and less sensitive to glue 1.

The addition of the spectral range 1700 – 2500 nm provided by the SWIR camera beyond the NIR spectrum of the FX17, allows i) a rather good detection of all three types of glue on cardboard, but also a better performance on rubber (thin and dry layer of glue 1 on rubber can start to be detected); ii) also the separation between dry and wet glue is more accurate but could be better.

Figure 5. PCA visualization with components 5 (red), 3 (green), and 4 (blue) and PLS-DA predictions accessed with Specim SWIR camera.

MWIR ANALYSIS OF GLUES WITH SPECIM FX50 CAMERA

The samples were finally measured with the Specim FX50 camera, on the Specim 100×50 LabScanner (see Figure 6). The spectral resolution is 35 nm over the full spectral range 2700 – 5300 nm. The pixel size on the image was ca. 0.3 mm.

Figure 6. Specim FX50 hyperspectral camera and 100×50 LabScanner

The Specim FX50 camera is very sensitive to glues. The analysis shows that glues strongly absorb light between 3000 and 3500 nm, regardless of the support material.

Again, a PCA was caried and a PLS DA model was applied on the data. Figure 7 shows the results of the analysis. The detection of the glue is much more accurate with FX50. We could detect even very thin layers of glue, even the dry glue 1 on rubber. However, the separation of glue type is challenging.

Figure 7. PCA visualization with components 5 (red), 3 (green), and 4 (blue) and PLS-DA predictions accessed with Specim FX50 camera.

CONCLUSION

This study used three Specim hyperspectral cameras with different spectral ranges to detect three types of glues, wet and dry, on cardboard and rubber. Based on the measurements and analysis, we can conclude as follows:

Regarding the detection of the glues (see Figure 8):

On cardboard
- All three cameras have similar performances.
On synthetic rubber
- The Specim FX17 camera is sufficient for detecting two types of glue out of three.
- The Specim SWIR camera is slightly better since it can start to detect the presence of a thin layer of dry glue 1 which the FX17 could not detect.
- The Specim FX50 is the best sensor in this respect and can reliably detect all three types of glues measured in this study.

Figure 8. Detection of glues by the three Specim hyperspectral cameras. False RGB and PLS DA model detection are presented.

Regarding the separation between the three types of glue (see Figure 9):

On cardboard
- All three cameras can separate the glue from each other.
On synthetic rubber
- The FX17 and SWIR cameras perform the best.

For the FX50, the separation of glues 2 and 3 is not perfect. We saw from the spectra that both glues 2 and 3 absorb the light very significantly without proper dedicated spectral features. It makes their sorting from the support straight forward, but their cross-separation more challenging.

Regarding the separation of the dry glue from the wet glue (see Figure 9):

On cardboard
- The SWIR camera is the most accurate, but only slightly better than the FX50.
On synthetic rubber
- None of the cameras is perfectly reliable.