Sensory Waste PCB Sorting

In this project we investigated how can sensory sorting of waste PCBs improve recycling yields of the valuable metals contained therein. We defined “sensory” as registration of the individual waste PCBs in an image database and subsequent waste-bin classification by a Machine Learning algorithm. For this purpose Saxion cooperated with a metals recycling company and an innovative waste processing company. We focussed on answering two questions: a) quantitatively evaluate the sorting accuracy and costs of sensing methods based on a database and b) describe how the definition of the wastebin contents can in practice be made compatible with high-grade recovery processes.
For the first question we created a database of waste PCBs that were previously manually labelled according to their wastebin classification (A, B or C). Then we trained a number of different ML models tasked to automatically classify the images. A subset of the images was sent to manually re-classifyy, without the original label i.e. “blind”. The same subset was also classified by our best ML model. The results of the comparison showed that for 90% of the instances, the ML model either classified correctly or made the same mistake as the human specialist.
For the second question we decided to perform experiments with the high-grade recovery process of one of our partners. A number of identical waste PCBs, collected from elsewhere as production fails, were subjected to the process of DCI – direct carbon immobilisation. Here, the organic waste content is transformed to CO and H2 gas (syngas) as well as pure carbon residue. The metal and ceramic content of the waste PCBs remains as an output useful for further metallurgical processing. We concluded that PCB sorting before DCI brings the same value as without DCI but that sorting the metal/ceramic mix post-DCI further improves metal recovery.

About half of the e-waste generated in The Netherlands is properly documented and collected (184kT in 2018). The amount of PCBs in this waste is projected to be about 7kT in 2018 with a growth rate of 3-4%. Studies indicate that a third of the weight of a PCB is made or recoverable and critical metals which we need as resources for the various societal challenges facing us in the future.

Recycling a waste PCB today means first shredding it and then processing it for material recovery mostly via non-selective pyrometallurgical methods. Sorting the PCBs in quality grades (wastebins) before shredding would however lead to more flexibility in selecting when and which recovery metallurgy is to be used. The yield and diversity of the recovered metals increases as a result, especially when high-grade recycling techniques are used.

Unfortunately, the sorting of waste PCBs is not easily automated as an experienced operator eye is needed to classify the very inhomogeneous waste-PCB stream in wastebins. In this project, a knowledge institution partners with an e-waste processor, a high-grade recycling technology startup and a developer of waste sorting systems to investigate the efficiency of methods for sensory sorting of waste PCBs. The knowledge gained in this project will lead towards a waste PCB sorting demonstrator as a follow-up project.