Nonferrous Metal Sorter: Final Report

Abstract

The inefficient sorting of nonmagnetic metals can be improved with a nonferrous (nonmagnetic) electromagnet. Scrap metal recyclers sort nonmagnetic metals by hand or by chemical density sorting, both of which are time and labor intensive. The Nonferrous Metal Sorter project aims to reduce manual labor, chemical use, and sorting time by building a testbed that attracts nonmagnetic metals such as copper and aluminum with a special electromagnet. This magnet, a nonferrous magnet, can vary its frequency through a control system. That variable frequency is the point: it should allow specific nonmagnetic metals to be targeted for attraction. As nonmagnetic scrap travels down a chute or conveyor, nonferrous magnets set to different frequencies would attract specific metals and sort the stream.

This report covers the initial magnet design and construction, based on Leonard Crow's research on nonferrous metal attraction. To control the variable frequency needed to attract different metals, an inverter was built. Experiments on the initial design led to an improved second magnet and a customized control system. That control system is the focus of the design and the major component of the testbed: an inverter for frequency control, a capacitor bank for power factor correction, and a sensor network to monitor the system.

§ 01Introduction

In a 1951 paper, Leonard Crow described techniques for attracting aluminum and copper with modified electromagnets. Using his concept, the team set out to build a nonferrous metal sorter that would attract different metals at unique frequencies, so scrap recyclers could sort more efficiently. Currently nonferrous metals are sorted by hand or by chemical density methods, both economically inefficient in time, labor, and material. The team built a functional testbed intended to let future researchers prove that attractive nonferrous sorting can differentiate between metals. That differentiation was not proven by year's end.

The project was split into four major milestones:

1. Develop a specialized electromagnet, based on Crow's design, that attracts aluminum and copper using 60 Hz AC. Achieved; the major winter-quarter deliverable.
2. Modify the magnet's power source to run from a DC source, so the field frequency could be varied with an inverter. Essential to the stretch goal of attracting copper and aluminum at different frequencies.
3. Attach voltage, current, and temperature sensors for data collection, operation, and protection of the magnet.
4. Develop code to use current measurements to track the mass of material picked up. During research this proved impossible, as there was no relationship between current drawn and mass attracted. Code was instead written to run the magnet, the sensor network, and over-temperature protection. PCBs integrating the inverter and sensor controls were built.

The completed design functions as a testbed for future researchers to test whether differentiated, frequency-based nonferrous sorting is possible and economically practical.

The first major block is the H-bridge inverter, which gives control over the frequency of the magnet's field. It uses four insulated gate bipolar transistors (IGBTs) switched by a digital signal processor (DSP) using sine-triangle modulation, programmed in Simulink. The inverter is powered by a 1000 V DC supply and drives the magnet through the capacitor bank.

Because the magnet's impedance changes with frequency, the capacitor bank adds power factor correction between inverter and magnet. It comprises seven series and six parallel 1 µF capacitors and, via relays, changes its capacitance as the magnet's frequency sweeps from 150 to 1000 Hz, controlled by the operator through a ring encoder.

To attract nonmagnetic test pieces, a special electromagnet with a specific geometry was built, following Crow's design: a laminated steel outer cylinder and inner core, a copper ring around the inner core, and 778 turns of 18 AWG copper wire wound around the outer cylinder. A sensor network of voltage and current sensors plus a temperature sensor keeps the magnet from overheating and drives two LCD displays showing voltage, current, temperature, frequency, and capacitor-bank value; it shuts the system down at an upper temperature limit.

§ 02Initial magnet design and construction

The nonferrous magnet works on the principle that diamagnetic materials, placed in an external field, induce eddy currents that produce an opposing field. Two diamagnetic pieces in the field are each repelled by the external field while being attracted to each other. The magnet exploits this by fixing a piece of diamagnetic copper within an alternating field; eddy currents induced in that copper produce attractive forces on other diamagnetic material placed within the copper's region (and repulsive forces outside it). So the attracted material must be smaller than the copper's surface and sit within the copper plane to feel a non-repulsive force.

Physically, the magnet is a cylinder with a rod through its center. A copper washer, inner diameter matching the rod and outer diameter matching the cylinder's inner wall, is bonded to the magnet face. The whole assembly is wrapped with wire and driven by an alternating source. Current through the coils creates a field in the core:

This field passes through the outer core while inducing current in the copper washer, whose opposing field creates a shaded-pole effect: the flux through the center of the face lags the flux through the outer core. The shifting field exerts horizontal forces that pull the object toward, and center it on, the magnet face.

The force depends on the eddy currents in the copper disc and in the material being attracted. Those currents arise from Faraday's law of induction:

So the eddy currents grow with more coil current (stronger field), a higher rate of change of flux, or more windings. A higher rate of change of flux comes from raising the AC frequency with an inverter, which increases force without dramatically increasing resistive loss (more current just raises ohmic heating).

Using this framework and Crow's prior work, the magnet design began. An electromagnetics simulator was needed to verify geometries and visualize forces, because the nonferrous magnet has flux traveling through air for several centimeters, far larger than the millimeter air gaps of a normal magnetic circuit. Two programs were surveyed in early January: Computer Simulation Technology (whose free student version was too limited) and ANSYS Maxwell Electromagnetics (Maxwell 3D), available in full in the lab and ultimately used. Early simulations used helixes for the coils, causing 12 to 24 hour solve times; switching to the built-in coil solver cut this to about ten minutes, and simulations confirmed the Crow-based magnet would attract nonferrous metals. An E-core design with a shorted secondary was also simulated but abandoned, since it only began attracting above 1 kHz. Simulation showed geometry, along with primary current, is the key to attraction.

The first magnet built was the E-core design; it worked as a normal electromagnet but did not attract nonferrous metals. With 12-gauge wire it drew enough current to blow the Variac's 15 A fuse repeatedly, so 18-gauge wire was adopted for later designs. It also overheated, exceeding 200 °C after twenty seconds, past the rating of the wire and Kapton tape. A copper secondary in 12-gauge wire and copper bar, with a 67-turn 18-gauge primary, still did not attract nonferrous metals, matching the simulated null result at 60 Hz.

The first working nonferrous magnet followed Crow's original design: a laminated steel outer and inner core, a shorted copper secondary around the inner core, and a primary wound on the outer core, driven at 60 Hz AC.

Several core materials were weighed, powdered iron, ferrite, Mu Metal, and laminated steel, against geometric availability, saturation, and ohmic heating. Laminated steel was chosen for reduced core losses at the first test frequency of 60 Hz and across the planned 100 to 1000 Hz range. Since laminated steel is not sold in many shapes, laminated steel toroids were sourced on eBay from a maker in New York: inner diameter 2.9", outer 4.66", height 2", at $46.60 each. A scrapped laminated-steel transformer core was machined into a rough oval for the inner core. Three 3" by 1" copper discs were bought on eBay at $19.99 each and machined down to 2.9" to fit inside the outer core; one was bored out to sit around the inner core, one left solid, to test whether the inner core mattered. Crow had hinted the inner core acted only as a centering device; Maxwell 3D confirmed it was not necessary for attraction. Machining the copper took nearly five hours of careful milling and lathing to avoid overheating the cutters; the bored disc got hot enough to change color.

Two toroids were stacked and taped with Kapton tape (DuPont-rated to 200 °C), since ohmic heating beyond ordinary tape's tolerance was expected. In practice the wires heated faster than the core, so thermal protection with a 190 °C cutoff was added in spring. The primary was hand-wound with 135 turns of 18-gauge wire (rated 16 A at 90 °C).

Tested at 60 Hz, this magnet successfully attracted copper and aluminum, with and without the inner core, at roughly 45 VAC and 19 A, though this damaged the wire. Attractive force rose with voltage and current, verifying the simulated link between force and primary current. At the highest test of 115 VAC the wire reached 190 °C after 30 seconds, the steel cores about 80 °C, and the copper secondary about 50 °C, making thermal protection a primary concern for any industrial version. The attracted test pieces slid down the face under their own weight, meaning the attractive force did not exceed gravity acting orthogonally, and building-supplied AC did not allow the field frequency to be changed.

§ 03Inverter design and construction

Since frequency was the key to differentiating metals, the current through the primary had to be frequency-controlled. An H-bridge inverter using sine-triangle PWM was chosen in the fall for its reliability and ease of construction, giving a sinusoidal output to drive the magnet. The switches were Infineon FF150R12ME3G IGBTs ($92.57 each), available in the lab, with Power Concepts 2SP0115T2A evaluation boards as gate drivers ($90.05), avoiding building gate drivers from scratch.

An H-bridge (single-phase voltage source inverter) uses four switches, two high and two low; alternating them generates square-wave PWM. A plain square wave would not give the continuously changing current the magnet needs, so sinusoidal PWM (SPWM) was used, comparing a triangle carrier to a sinusoid at the desired output frequency. The IGBTs are rated 200 A and switch up to 20 kHz. A TI-2833 Delfino microcontroller DSP on a Piccolo experimenter board (TMDSDOCK28335, $154.26 total) implemented the sine-triangle modulation, programmed graphically in Simulink. A lab coworker, Atif Maqsood, experienced with these DSPs, helped set up the control in Simulink and debug the inverter. The fundamental frequency, originally 250 Hz, was later made adjustable via a potentiometer on an A/D pin so it could be changed without reprogramming.

The inverter was fed from a 1 kV, 30 A DC supply through a filter capacitor (originally 450 V, 2200 µF; increased in spring to 1000 V, 970 µF) to reduce stray inductance. Current then enters two IGBT packages on heat sinks, driven by the DSP through the gate drivers, with outputs to the magnet primary.

With the inverter at 200 V, 5 A, 250 Hz, the magnet attracted better than at 60 Hz, though power was still about 1 kW. The lighter aluminum piece stuck without help; the heavier copper piece did not, a reversal from 60 Hz, where even the aluminum would not stay. The inverter would not accept more than 300 V: the gate drivers auto-shut-down when stray inductance rose to damaging levels. The advising Ph.D. student suggested more input capacitance on the IGBTs, a busbar between IGBTs instead of wire, or moving the 2200 µF capacitor closer to the gate drivers.

The input filter capacitor was replaced with a Cornell Dubilier 947C971K102DLHS metal-film capacitor, 970 µF at 1000 V DC (to use the full 1 kV supply), roughly 6.5" tall by 4.5" in diameter, $138.40. The inverter was housed in a Bud Industries RM-14213 aluminum 19" rack case ($133.70) for space and ventilation. A PCB powered the gate drivers (15 V), supplied the DSP (5 V), and jumped signals from DSP to gate drivers.

Testing the 135-turn magnet with the inverter confirmed theory: as field frequency rose, eddy currents in the test metal rose and attractive force rose while less primary current was needed. So a well-designed inverter is essential to making this commercially deployable.

§ 04Second magnet and power factor correction

After success at 60 and 250 Hz, a second magnet of the same dimensions was built with more primary turns to reduce power consumption: 778 turns, henceforth "Magnet-2". An enclosure was considered (a 120 V, 6" cooling fan was bought; a steel pipe was rejected for interfering with the field; an ABS tube 12" in diameter with aluminum suspension rods was proposed but not fabricated), but since Magnet-2 was a testbed needing easy access, it stayed in open air with a rear fan. Any industrial version would need a NEMA- and NEC-compliant enclosure, a substantial design task of its own.

Tested with the inverter and a 60 Hz supply, Magnet-2 worked better at 60 Hz than at 250 Hz, because the added turns raised inductance by roughly 3221%, pushing impedance to 77 Ω and drawing only about 50 mA. Impedance rises with frequency:

and inductance rises quadratically with turns:

So more turns give a small linear rise in resistance and a large quadratic rise in inductance; higher frequency adds inductive reactance, worsening the parasitic effect. This increased impedance cut the current at a given voltage, weakening the field and the attractive force. To compensate, series capacitance was added to correct the power factor:

After spring break, the priority was power factor correction for Magnet-2. Early "back of envelope" calculations proved wrong and were redone; the correct value at 60 Hz was 50 µF. Two 100 µF, 250 V metal-film capacitors were placed in series with Magnet-2. This drew about 5 A at maximum voltage and attracted both aluminum and copper, holding the aluminum but not the heavier copper. Connected to a single phase of a three-phase outlet (208 V at 60 Hz), it drew 10.8 A and firmly held both the aluminum and copper discs, at about 2.2 kW.

To test power factor correction with the inverter at 250 Hz, 1 µF, 600 VAC capacitors ($6.70 each) were bought; the needed value was 2 µF. Current rose to 1.90 A at 550 VDC (measured at the inverter input), but neither disc stuck, and the capacitor's low voltage rating caused nonlinear impedance above 550 V. It was also realized that measurements should be taken at the inverter output (alternating current at a non-rated frequency), not the input. New 1 µF, 1.5 kVAC capacitors ($7.20 each) were used; two in parallel underperformed, so a third was added in parallel as an experiment, letting the magnet draw nearly 4 A at 600 VDC and hold both the copper and aluminum discs. The 600 VDC test matters because 600 VDC (with 480 VAC three-phase) is a common industrial voltage.

Because the magnet's impedance rises with frequency while capacitive reactance falls with frequency, the power factor cannot stay constant across a sweep. A switching capacitor bank was built to add capacitors in series or parallel:

§ 05Sensor network and capacitor-bank control

Alongside the capacitor bank, a sensor network was built with voltage and current sensors at the inverter input and output and at the capacitor-bank output (just before the magnet), plus over-temperature protection to keep the coils below their 200 °C rating, killing power at 190 °C. An Arduino Mega 2560 ($38.50) was chosen as the controller for its available libraries and its 25+ usable pins (at least twelve digital pins were needed for the capacitor switching network). An industrial system would use a PLC; for the testbed the Mega sufficed.

The voltage sensor was a LEM LV 20-P ($45.54) for accuracy, linearity, low thermal drift, fast response, and interference immunity. The current sensor was a LEM LA 55-P Hall-effect sensor ($26.54). For over-temperature, a thermocouple ($25.69) with a MAX6675 amplifier ($10.79) was chosen over thermistors and RTDs, though this caused issues later.

The LV 20-P needs ±12 V rails, supplied by a RAC10-12DK/277 board-mount supply ($11.87), and 10 mA through the device; with an average magnet voltage near 500 V, a 50 kΩ series resistor (dissipating 5 W) set the input current. The sensor's output current mirror was measured over a 100 Ω resistor and scaled to voltage in code. The LA 55-P current sensor, on the same ±12 V rails, needs about 50 A for best accuracy, so wire was looped through it (loops ≈ 50 A / I_measure); its output current mirror was read over a 100 Ω resistor and converted by the 1:2000 ratio.

Near the operating magnet, the thermocouple's wire picked up large voltage spikes from the alternating field. Analog low-pass filters could not remove them, so noise was filtered digitally: a program discarded improbable readings, then passed the rest through a moving-average filter. The thermocouple's 200 ms sample period limited it to five readings per second; a thermistor (sampling near 9.6 kHz) would have filtered noise faster, but the magnet heats slowly enough that the thermocouple was acceptable.

For the capacitor-bank switching, SCRs were considered but could not be reset as needed, so relays were used: TE Connectivity Potter & Brumfield relays rated 40 A at 12 V ($2.83 each), driven by RFP30N06LE MOSFETs ($1.69 each, 30 A, 47 mΩ on-resistance) that easily handle the 130 mA relay coils, with 1N4001 diodes clamping inductive kicks to protect the MOSFETs.

Sensors, relay switching, and the Arduino Mega were combined onto a single PCB nearly a square foot in size, with twenty MOSFET relay switches (eight spare for future expansion), three connections each for voltage and current sensors, and a ring-encoder jumper for capacitor-bank control. From EasyEDA ($70 for five boards), the populated board had minor errors: Arduino Mega mounting holes were sized for a smaller Arduino (fixed with hot glue), and a reversed pinout on the +12 V supply (a Meanwell IRM-45-12, $13.50) was corrected by standing the supply on standoffs and rerouting 18-gauge wire. The capacitor bank was wired separately from a PCB because of its high currents, a tedious soldering job in 22-gauge control wire and 12-gauge power wire.

The Arduino code switched the relays reliably and sent sensor readings, frequency (from the DSP's A/D value), and a coded capacitor-bank value to two LCD displays. Finally the inverter, capacitor bank, sensor network, and magnet were integrated and tested. The capacitor bank still needed refinement to switch smoothly as the inverter frequency changed.

§ 06Conclusion

Over six months the team achieved all four major milestones: a specialized electromagnet that attracts nonferrous metals, an inverter to control field frequency, a sensor network for data and control, and integrating code. Two stretch goals were not met: proving that controlling frequency lets the magnet differentiate between copper and aluminum, and drafting a journal paper on those results. Even so, the team delivered a working testbed for future research by the customer's team.

The two winter-quarter deliverables, a modified electromagnet attracting nonferrous metals at 60 Hz and an inverter supply for testing other frequencies, were both completed and tested to the customer's satisfaction, along with every Gantt-chart task (simulating geometries, fabricating two magnets, and building the inverter). A customer request to research other nonferrous attraction methods was unsuccessful.

In spring a major overlooked issue forced design changes: the 778-turn magnet's high inductance (183 mH) created 77 Ω of impedance. Since attraction depends on primary current, frequency-swept power factor correction was needed, so an adjustable capacitor bank was built alongside the sensor network. Spring work was productive: a working sensor network and capacitor bank, integrated with the winter's magnet and inverter, with code to run the sensor network, the inverter's DSP, and the capacitor bank. The two unforeseen problems, power factor correction and over-temperature protection, were both resolved, each conceptually simple but time-consuming to implement well.

Across voltages and frequencies, the power needed to retain the aluminum disc was similar: a "retention point" of 2.2 kW for Magnet-2 at 208 V, 60 Hz, and 1.8 kW at 600 VDC, 250 Hz. The 600 VDC and 208 VAC points matter because they are common industrial values; a facility on 480 V three-phase could run three magnets from one line. But this alone does not let the magnet differentiate between metals, so it does not yet solve the sorting problem.

The testbed cost under $2100 (under $2800 total spend, excluding lab equipment and magnet casing). The costliest part was the inverter, whose IGBTs and gate drivers alone were about $700; the magnet came in under $400. The major milestones were achieved; the stretch goal of differentiating between metals was not. That differentiation is the next step in developing this into a nonferrous sorting method, and the hope is that a following team will carry it out.

Appendix ISimulink inverter control

The DSP inverter control was implemented in Simulink: an ADC path scaling the A/D reading to a modulation index, an electrical-angle block generating the sinusoidal reference from the commanded frequency, and ePWM blocks producing the sine-triangle gate signals for the H-bridge, clocked at 150 MHz. The full block diagram appears in the PDF.

References

1. L.R. Crow. Design, Construction and Operating Principles of Electromagnets for Attracting Copper, Aluminum, and Other Non-Ferrous Materials. Universal Scientific Company, 1951.
2. Stoll, Richard L. "The analysis of eddy currents." 1974.