Assessment · 2026-07-25
Where Magnetocs stands
Magnetocs was our senior design project back in 2017 and 2018. This page is my 2026 update: an honest read of what it is, and what it is not. There are two ways to score it, and they land very differently.
I am treating our testbed as what it is: rigorous engineering and a real open question, not a product. Saying so myself is the point.
§ 01Two scorecards
We integrated electromagnetics, power electronics, control, thermal, sensors, and PCB design into one working testbed in a few quarters, and documented what failed as carefully as what worked.
Our device attracts nonferrous metal. Industry already recovers it faster and cheaper by doing the opposite, and the differentiation problem is being won by sensors and AI, not magnets.
§ 02What we are proud of
The breadth was real. We built a Crow-geometry nonferrous magnet, an H-bridge IGBT inverter under DSP and Simulink control, a switching capacitor bank doing live power factor correction across 150 to 1000 Hz, a sensor network with over-temperature protection, and two magnet iterations. Most senior projects attempt a fraction of that.
What we are prouder of is that we told the truth in the report. We documented the E-core that did not work, the power factor calculation that came out wrong, the thermocouple noise, and, most tellingly, that our fourth milestone was physically impossible because there is no relationship between coil current and attracted mass. We killed our own feature on the evidence. That candor is why we are publishing this platform in this form, and why we are willing to grade our own work honestly below.
§ 03The premise problem
Our magnet attracts nonferrous metal toward its face. But the industry recovers nonferrous metal by the opposite mechanism: an eddy-current separator spins a magnetic rotor that repels conductive metal and throws it off the belt, at tons per hour. That technology is mature, cheap, and everywhere. Crow's attraction method was underdeveloped for a concrete reason, and our own numbers confirm it: roughly 2.2 kW to hold a single disc is not in the same universe as industrial throughput.
Meanwhile the genuinely valuable problem, telling copper from aluminum from a specific alloy, is being solved by sensing composition directly, not by tuning a magnet. In hindsight, the project chased a solved problem with a harder method, while the valuable problem sat on a different technology road.
§ 04The field today
The market has split into two fronts, and Magnetocs straddles both without leading either.
Getting nonferrous metal out of the stream
Owned by eddy-current separators that repel and eject conductive metal. This is mature, mass-deployed hardware sold by the thousands, with continuous gains in belt speed, fines recovery, and energy use. Our attraction approach competes here and loses on power and throughput.
Telling the metals apart, the goal we reached for
Being taken by sensor-based sorting and AI, not magnets. Laser spectroscopy (LIBS), X-ray fluorescence and transmission (XRF, XRT), and deep-learning object recognition now separate aluminum by alloy series and split copper, brass, zinc and stainless in real time, at production speeds. This is exactly the frequency-selective sorting we aimed for, delivered by optics and compute instead.
The market context is genuinely hot. Copper demand, tens of millions of tonnes of e-waste generated a year, critical-mineral supply pressure, and the drive for low-carbon "green" aluminum all push money into better sorting. That tailwind is real. It just blows hardest on the sensor-and-AI road, not the electromagnet one we took.
§ 05SWOT
- A working, documented, reproducible testbed with a full public report.
- Real power-electronics, controls, and systems-integration depth.
- Honest scope: proven claims and named open problems, not hype.
- A distinctive, ownable name and a live domain with history.
- An open physics question (frequency-selective attraction) no commercial player is chasing.
- Attraction is a power hog: about 2.2 kW to hold one disc.
- We never demonstrated differentiation between metals.
- The mechanism is orthogonal to how industry actually recovers nonferrous metal.
- A single low-throughput bench unit with no shown path to tons per hour.
- A hot market: copper demand, e-waste volume, critical minerals, green aluminum.
- Strong credibility platform and entry ticket into recycling-tech conversations.
- The frequency-selective physics could yield a genuine research contribution, even a paper, without being a product.
- Entrenched incumbents with deployed fleets and active AI roadmaps.
- Sensor-and-AI economics improving fast, closing the window further.
- The valuable capability (alloy differentiation) is already commercialized.
- The power and throughput ceiling may be fundamental to attraction itself.
§ 06My verdict
I am keeping Magnetocs as exactly what it is: rigorous engineering and a real open physics question. I am not going to inflate it into a live venture just because it has a good name and a place to sit. Its true value is not the magnet. It is that the three of us could build a complete electromechanical system and reason honestly about it, a skill that outlasts the device.
If the recycling and critical-metals itch is real, and it is a well-funded space right now, the frontier is sensor-based and AI sorting and the economics of recovery. Magnetocs is an entry ticket to that conversation, not the product to bring to it.
The best thing this project produced was not a sorter. It was three engineers who could build one.