Hybrid sensor for metal grade measurement of a falling stream of solid waste particles

A hybrid sensor system for accurate detection of the metal grade of a stream of falling solid waste particles is investigated and experimentally verified. The system holds an infrared and an electromagnetic unit around a central tube and counts all the particles and only the metal particles, respectively. The count ratio together with the measured average particle mass ratio (k) of non-metal and metal particles is sufficient for calculation of grade. The performance of the system is accurately verified using synthetic mixtures of sand and metal particles. Towards an application a case study is performed using municipal solid waste incineration bottom ash in size fractions 1–6 mm, which presents a major challenge for nonferrous metal recovery. The particle count ratio was inherently accurate for particle feed rates up to 13 per second. The average value and spread of k for bottom ash was determined as 0.49 ± 0.07 and used to calculate grade within 2.4% from the manually analysed grade. At higher feed rates the sensors start missing particles which fall simultaneously through the central tube, but the hybrid system still counted highly repeatable. This allowed for implementation of a count correction ratio to eliminate the stationary error. In combination with averaging in measurement intervals for suppression of stochastic variations the hybrid system regained its accuracy for particle feed rates up to 143 per second. This performance and its special design, intended to render it insensitive to external interference and noise when applied in an eddy current separator, make the hybrid sensor suitable for applications such as quality control and sensor controlled separation.

► A new sensor system is developed for metal grade measurement of falling bottom ash particles.
► The system is hybrid, consisting of an optical and an electromagnetic sensor.
► Grade of ECS concentrated bottom ash in 1–6 mm sieve size accurately measured up to 143 p/s feed rate.
► Accuracy reached was 2.4% with respect to manual analysis.
► Measures for elimination of both stationary and stochastic errors are discussed.