@@ -16,10 +16,35 @@ It is based on a discrete electrostatics solver (shown below), and also document
## Magnetic
Another proximity measurement uses the magnetic field instead of the electric field. Integrated circuits are available with extremely sensitive magnetostrictive, magnetorestive, and hall effect sensors. Using differential pairs of these elements, very low-cost, non-contact rotary and linear encoders can be made. <ahref='http://web.cba.mit.edu/sam/kiri/doc/loadcell.html'>This page</a> documents experiments with a 1 DOF magnetic load cell, based on the Austrian microsystems AS5510 IC, which boasts 500nm resolution for $1.41 (Q: 100).
Another proximity measurement uses the magnetic field instead of the electric field. Integrated circuits are available with extremely sensitive magnetostrictive, magnetorestive, and hall effect sensors. Using differential pairs of these elements, very low-cost, non-contact rotary and linear encoders can be made. Below, we document experiments with a 1 DOF magnetic load cell, based on the Austrian microsystems AS5510 10 bit linear encoder IC, which boasts 500nm resolution for $1.41 (Q: 100).
<imgsrc='prior/as5011-loadcell.jpg'width=40%>
### AS5510 1 DOF Loadcell
<figure>
<imgsrc='prior/as5011-loadcell.jpg'height=300px>
<imgsrc='img/loadcell-hall/pcb.png'height=300px>
</figure>
The PCB is mounted to a piece of waterjet aluminum with a flexure cut into it. A magnet sits in a hole on the moving part of the flexure. I designed the flexure to be roughly 5 um/Newton. The encoder wants a air gap of less than 1mm, so I milled away the board inside the SOIC-8 footprint and flipped the part over to bring it closer to the magnet underneath.
I'm using an 8mm circular diametrically magnetized magnet. We could swap in a smaller magnet to push the resolution of the prototype, but I started with the datasheet's recommended magnet form factor. I'm running the chip at its highest sensitivity (+-12.5 mT full scale) and in slow mode (12.5 KHz as opposed to 50 KHz, but .5 mT peak-to-peak noise as opposed to .8 mTpp).
The graphs below show instron characterization of the prototype. First, the flexure had a stiffness of 4.6 um/N, close to the designed 5 um/N. Second, the time series comparison is very boring because the two series are so identical. We see that a displacement of 400 um (corresponding to 100N) takes the encoder nearly over its full range (512 is the full range, as it measures positive and negative displacement to 10 bits (1024)).
Finally, we can plot the relation between the two variables (force and hall reading). Running at the full 12.5 kHz (i.e., without any summing of samples), we get about 5x LSB per Newton. This isn't amazing resolution, but it is very much good enough for my application. Further, we can easily take a hit in speed to get an increase in resolution by summing (i.e., averaging) samples, depending on the applciation.
