The extraction of purified higher-value alloy streams from non-ferrous metal waste remains a ‘challenging task’ owing to the diversity of alloys and the resemblance in their physical properties, observes Lorraine Braibant of the University of Liège in Belgium. She reports that a combination of visible and near-infrared (VNIR) and X-ray transmission (XRT) hyperspectral imaging customised for a multi-sensor system have proven to be an efficient technique to identify valuable metal fractions.
‘It is virtually impossible to discriminate metals like zinc, copper and brass based on X-ray transmission imaging alone,’ ventures Lorraine Braibant, who is part of the University of Liège’s PICKIT Project. She notes that materials with coating or dirt can significantly reduce the accuracty of X-ray transmission imaging. This problems results in a generous amount of false detections thus making it difficult to produce a pure stream of non-ferrous alloys.
Insensitive to surface conditions
Therefore, Liège University’s research crew turned instead to multi-energy X-ray transmission (XRT). Braibant described this a ‘non-destructive’ method to probe the composition of alloys based on the relationship between the atomic number of the absorbing element and the shape of its attenuation curve. ‘XRT characterisation is advantageously insensitive to the surface conditions of the fragments,’ she points out.
The PICKIT process
The PICKIT Project ran for two years (2015-2017) with the aim to optimise the classification of the post-shredded non-ferrous metal content present in the end-of-life electronics waste stream. Specifically, the researchers were determined to efficiently sort alloy scrap metals from end-of-life electronics into five ‘alloy classes of interest’. These are: aluminum alloys (light grey metal); copper alloys (dense red metal); zinc alloys (dense grey metal), brass alloys (dense yellow metal); and pieces of brass coated with nickel (shiny aspect).
The waste stream underwent a screening step to yield fractions of 20-100 mm in size, followed by a magnetic separation step and eddy currents separation step.
Speaking about classification based on VNIR reflectance curves, the researcher observes that ‘nickeled brass is identified with limited success’. This might be due to specular reflection that is more likely to occur with nickeled brass and saturates the VNIR sensor. Except for the nickeled brass, other alloy fragments are successfully identified at a rate above 85% when the decision is computed at the object level.
Also, it has become clear that the integration of XRT data improves the classification of aluminum both at the pixel and object level, Braibant reports. However, it ‘slightly deteriorates’ the identification of copper, increasing the confusion with brass.