Written by Dieter Bingemann, Ph.D.
At some point during the ’90s, the world officially entered the “plastic age,” with plastics passing steel as the number one material used by volume. These man-made organic macromolecules, also called polymers, are ubiquitous in our daily lives — from reducing airplane weight to preserving food freshness. As plastics are most often only intended for a single use, these products are also responsible for a dramatic trash problem worldwide, as exemplified by the “Great Pacific Garbage Patch” in the Northern Pacific Ocean.
As polymer products are commonly manufactured through injection molding at fairly low temperatures, recycling of these materials into the production process should be simple. However, very few types of plastics are miscible, which mandates a clean separation of the different polymers in the recycling stream.
Currently, plastic sorting is performed manually using the recycling codes imprinted on most plastics – a high cost solution, which renders recycling unprofitable. However, as the near-infrared spectrum of each type of polymer is unique, spectroscopy can be used to identify and sort a large variety of plastics with high specificity. In this application note we will demonstrate the principle of plastic sorting using near-infrared spectroscopy.
We recorded the spectrum of about 40 plastic items of known identity taken from household garbage using a diffuse reflectance setup based on a Vivo illumination stage. The samples were placed on top of the Vivo’s sapphire window, and a 400 µm VIS-NIR fiber mounted in the center bore of the stage collected the light reflected from the samples. A Flame-NIR spectrometer measured the spectrum from 950-1650 nm, with a white standard used as the reference on the stage. With a cost about one-fourth that of a benchtop NIR spectrometer, Flame-NIR uses proprietary electronics to achieve surprisingly low noise levels from an uncooled InGaAs array detector in a small optical bench.
OceanView software was used to acquire data in absorbance mode so as to plot log(1/R) of the reflectance spectra, with R being the reflectance of the sample. This can be a more intuitive way to show reflectance spectra for chemical analysis, as taking log(1/R) of the reflectance spectrum allows us to see concentration as scaling with signal intensity. We averaged 10 spectra with an integration time of 75 ms each for each spectrum.
Depending on the sample thickness, roughness, and density, different amounts of light are reflected into the detection fiber. To focus on the actual chemical differences between samples, rather than these physical attributes, we normalized the spectra such that each spectrum’s mean is zero and the total variance across all wavelengths is one (the so-called Standard Normal Variate method). The resulting normalized spectra are shown in Figure 1, colored by type of plastic.
Principal Component Analysis (PCA), a chemometric analysis method, was applied to the data collected in an attempt to group and classify the plastics by type using their spectral signature.
Even though each spectrum was recorded for a different item from the trash collection, each corresponding to different origins and manufacturers, the spectra are very similar for each plastic type, with only the polypropylene spectra (PP) showing some variation from item to item (recycling code 5).
We can visualize the grouping more clearly through principal component analysis of the spectra, which captures similarities and differences between spectra in a simple diagram, as shown in Figure 2 using the first two principal components.
In addition to the chemical identity of the monomer, the reflection spectra are also sensitive to the branching in the polymer chain. For example, it is possible to distinguish between high-density polyethylene (HDPE) and low-density polyethylene (LDPE), recycling codes 2 and 4, respectively. A similar grouping is also visible for the polypropylenes (PP), with yogurt cups forming the lower PP group, and the more flexible thin plastic films the upper PP group.
Considering the third principal component as well (Figure 3), we can visually identify the different plastic types easily with the location of the spectrum in the principal component space. Indeed, a simple linear discriminant analysis (LDA) classifies all samples correctly as either PE, PS, PP, or PET (combining the small subgroups of identical chemical makeup). Note the much closer distribution of PP using these principal components.
NIR spectroscopy makes plastic sorting as simple as 1-2-3, and the Flame-NIR spectrometer has the right performance, compact footprint, and low noise to become a cost-effective tool to improve the efficiency of sorting facilities.
Using automatic sorting through NIR spectroscopy, plastic recycling can become economical, turn trash into a renewable resource, remove large amounts of material from the garbage stream, and make the earth a greener place.