Classification of plastics containing brominated flame retardant through hyperspectral imaging and chemometrics

Almost all kind of plastic used nowadays need to incorporate flame retardant to meet the fire safety standards required from the consumers and comply with the regulations. Depending on the specific application of the plastics, the amount and the type of flame retardants can differ, so that in waste plastics, a large variety of polymers and flame retardants can be found.
According to the current European regulations, recycling of polymers from all categories is increasing, also in the case of plastics containing flame retardant. The most commonly used brominated flame retardants are often fully compatible with integrated waste management systems. However, only plastics of the same polymer type and with a close match in additive content can be conjointly recycled. Therefore, a fast and reliable way to identify and distinguish both the polymer and the contained substances, is required.
With the ultimate goal of an automatic implementation in sorting and recycling plants, in this work the possibility of using a spectroscopic imaging technique has been investigated for classifying different plastics containing different brominated flame retardants, according to both these categories.

Two different kind of polymers (acrylonitrile-butadiene-styrene, ABS and polystyrene, PS) have been prepared in a disk shape, with the use of three different common flame retardants (pentabromophenyl ether; 1,2,5,6,9,10-Hexabromo-cyclododecane and 3,3',5,5'-Tetrabromobisphenol A), both in natural color and in black (adding carbon black). Reference samples of each plastic (without flame retardant) have been prepared as well, repeating the preparation of each sample twice so that a total number of 32 plastic samples have been analyzed.
They have been analyzed all together with a hyperspectral camera operating in the NIR region between 1000 and 2500 nm (Umbio Inspector).

After having cut the noisy regions of the spectra, binned the image, and preprocessed the signals (SNV + Smoothing), a suitable number of pixels for each class has been selected and used for building the classification model.
Since the simultaneous identification of the specific polymer and flame retardant is a hard challenge when both natural and black plastic are analyzed, a hierarchical approach has been chosen. The natural and black plastics have been divided at first, then each pixel have been classified accordingly to the polymer of which the plastic is composed, and as a final step, the presence of flame retardant in the plastic and its exact type have been identified.
The results of this model allowed to classify all disks according to the polymer of which they are constituted and almost all of them according to the flame retardant that was used in the preparation. In particular, all the plastics where no carbon black was added were correctly classified according both to the polymer and the flame retardant; while, among the black ones, more than half of the samples have been assigned to the correct class.
These results constitute a promising starting point for the implementation of NIR sensors on the recycling line of waste natural and black plastic and represent one of the very first examples in this field.