Integrated hyperspectral and Raman sensors for fast characterization of plastics in e-waste recycling streams

Electronic waste (e-waste) is a fast-growing and complex material stream, and its adequate management requires robust identification tools. Plastics are major components of e-waste (~25% of its total) and polymer compositions vary depending on its initial function in the electronic device, including the presence of additives to compensate for function loss and polymer blends (e.g ABS/PS). Recycling of these increasingly complex e-waste polymers calls for innovative, fast, and spatially-resolved identification tools compatible with conveyor belt operations.
We propose fast identification of polymers in e-waste using hyperspectral imagery (HSI) and Raman spectroscopy. We evaluated the potential of such methods for resolving diagnostic spectral features (fingerprints) in 23 polymer types, including black plastics, and determined the minimum acquisition requirements for robust identification of polymers considering the recycling industry demands. We employed short-wave infrared (SWIR-HSI, SPECIM AisaFENIX
We identified fingerprints for each material based on previous literature reports: a positive identification was assigned for fingerprints with signal-to-noise-ratios > 4 (hull-corrected reflectance - HSI) or > 3 (Raman). We identified 60% of the transparent and light-colored plastic types using SWIR-HSI, and two out of five black plastic types using MWIR-HSI information. In total, HSI-reflectance methods were suitable for characterization of 16 plastic types (70%), acquiring simultaneous spectral and spatial information at fast rates. Still, the need for characterizing black plastics was not fully met using HSI-reflectance sensors.
We propose to employ point Raman spectroscopy as the ultimate tool for polymer identification and evaluated its potential for integration with HSI data. We performed short-time point Raman measurements (≤ 2 seconds), investigating if the method is compatible with the fast-paced data acquisition rates required by the recycling industry. We identified fingerprints for all polymers using signals from short-time point Raman measurements (transparent plastics: 0.5 seconds, black plastics: 1 second).
We elaborated sensor-specific libraries for each plastic type for future cross-validation and correlation in automated polymer detection during recycling operations (dictionary learning). This reference data represents maps which capture the spectral variability within the reference materials as a training dataset. To evaluate the added value of the spatial variability recorded by HSI sensors, we showcase the application of our spectral library for polymer identification and mapping in artificially mixed samples of plastic debris.
Our RAMSES network, a project financed by the EU via the KIC Raw Materials, develops the tools for combining different sensor types: HSI-reflectance sensors and RGB cameras at the beginning of the line record information needed to determine representative spectral domains. This signal is then used, along with the polymer’s spectral library, to identify transparent and light-colored plastics. The spatial and spectral domain information are used to direct Raman point acquisitions for identification of black plastics and polymer blends. This intertwined multi- sensor network relies on fast data acquisition and processing supported by spectral libraries and machine learning algorithms targeting object detection and spectral data fusion. This investigation contributes to enabling highly accurate identification of polymers, specifying the conditions for rapid smart multi-sensor data acquisition and processing.