After my work on oneTesla, I immediately became interested in pushing even bigger sparks out of those magical little FGA60N65SMD TO247 IGBTs by moving up to an H-bridge and a bigger secondary resonator. I slowly iterated over various configurations during my Junior year at MIT (2014-2015) until I came up with a robust, portable, and (I think) relatively elegant design that runs off an ordinary mains outlet. I now use it as my go-to small-sized demonstration coil.
0. Page Contents
- Electrical Design and Schematics
- Mechanical Design and CAD Drawings
- Future Work
- H-Bridge of FGA60N65SMD TO247-Package IGBTs with 300A OCD (Over-Current Detection) Setpoint
- UD2.8A (my own slight revision of the UD2.7) Universal DRSSTC Driver with Primary Current Feedback, Tunable-Inductor Phase Lead, OCD, and UVLO
- 4s2p Primary MMC of CDE 942C20P15K 150nF, 1.5kVDC Film/Foil Capacitors for a total rating of 75nF at 6kVDC
- ??? Primary Coil
- 3.5″ x 12″ Lathe-Wound 36AWG Secondary Coil
- 4″ x 13″ Spun Aluminum Toroid
The electrical system is based on my general purpose H-bridge inverter PCB plus a universal DRSSTC driver. The key specifications of the H-bridge board are repeated here from my original post:
- H-Bridge with four spots for TO247 IGBTs or MOSFETs
- Gate protection network on each gate
- 0805 and 2010 dual package landing pattern for gate resistors
- 1.5KExxx series TVS across each TO247 device
- CDE940 series snubber capacitor
- Mounting space for two 1″-diameter GDTs
- SMA connectors for GDT drive signals
- Voltage doubler input with two TO220 diodes with heatsinks, two 10.16mm grid, 35mm diameter bus capacitors, and bleeder resistors
- 5mm x 20mm fuseholder
- 10AWG holes for external DC bus input
- 4oz copper on both sides with low inductance laminated bus structure
Digikey Bill of Materials (Excel file hosted in Dropbox)
The main modifications in my UD2.8A driver are an 0.1″ jumper for switching between the two fiber optic inputs, SMA input and output connectors instead of molex headers, an extra DC barrel jack for power, SMT LEDs instead a molex header, and a new layout with better DFM that un-tents all vias and moves all components to the top side with 150 mil board-edge clearance for all SMT components. This last part should make assembly much easier by allowing the board to be clamped by any two sides without covering any pads. Strictly speaking, tenting vias (covering them with solder mask) without filling them (which can be super-expensive!) makes no difference for hand soldering, but is poor practice for reflow oven soldering. Who knows, maybe someday someone will want to send a batch of these to an assembly house. Here’s what the first revision (UD2.8A Rev. A) looks like. It normally goes in a grounded metal box (not pictured):
I also have a second revision that reshuffles things to add back the molex LED and GDT output headers. I’ve called this one the UD2.8A Rev. B (sorry for the confusion, the UD2.8B is my FPGA-based controller). In this picture, it’s installed in a metal chassis (with the lid popped off) in my CM600 DRSSTC assembly:
The MMC is built on its own PCB with convenient #10 machine screw connectors for tapping it in various places:
I use a pair of 1:300 triad CTs (same as in oneTesla) for feedback and OCD, I may need to move up in size if I’m going to run much more than 300A primary currentt. I also have a fan for cooling the bridge and use a small TDK-Lambda 24V, 1A switching supply to power the Universal Driver. Here’s a shot of the contents of the entire chassis (minus fuseholders, which I hadn’t yet installed at this point):
4. Mechanical Design and CAD Drawings
I decided from the beginning that I wanted a laser-cut acrylic enclosure for this DRSSTC because wood would make the assembly too heavy and hinder quick design iteration. At the same time, however, I also wanted to avoid the pains of “puzzle piece” construction that made the original oneTesla chassis such a pain to service. A mini 80/20 frame, such as the one I used for chassis in my 6.111 oscilloscope project, might have worked, but I also wanted to reduce costs (that stuff is surprisingly expensive for how little material you get!) and avoid losses from the metal in the chassis forming a closed loop that can experience an induction heating effect.
GGY’s big twin DRSSTC (“Model 82”) use a chassis made of 80/20 extrusion connected together in exactly the way that might cause the induction heating effect, but according to him, the performance did not significantly change when he insulated the extrusions in a way that broke the loops. The suspicion here is that the anodizing layer is thick enough that an actual conductive loop is not forming. My own experience with my big CM600 DRSSTC is that even if this is the case, the right set of conditions might someday cause the insulating layer to suddenly arc over and produce a spectacular display of light and smoke!
I had parallel layers of aluminum tape on the bottom of my 8.5″ secondary coil that I thought were electrically connected, but instead they were insulated by their adhesive layer. One night, I applied just enough power to the coil to cause the adhesive to break down and suddenly there was smoke and light coming from the bottom of my secondary coil and the streamer length was cut down. The most perplexing thing was that every time I bled the bus and examined the whole resonator assembly, nothing appeared to be damaged. After realizing what was going on, I removed all of the aluminum tape from the bottom of the coil and observed scorch marks under the areas where there had been overlap. In my opinion it can only hurt to have places where loops can form if thin layers of insulation break down…
The solution I came up with is to use four 4.5″-tall 80/20 extrusions as edge supports and rely on the sheets of acrylic bolted to the top, bottom, and sides to make the whole structure rigid:
This breaks all of the metal loops in the chassis and allows any single piece of acrylic to be easily detached for quick servicing and assembly. The downside is that the chassis is less rigid than if it was built around and 80/20 box, but by picking the thickness of the acrylic correctly relative to the edge length, good rigidity can be maintained. In my case, I designed the chassis for 6mm (0.236″) acrylic, which worked fine. The one exception is the fan/fuse/power connector panel, which I cut from 1/8″ acrylic so that the panel mount fuses I picked could be mounted without milling pockets in the 6mm sheet.
So far, I have experienced zero IGBT failures with the current configuration and 300A OCD setpoint. Maybe it’s time to see how far I can push the primary current before these IGBTs become unreliable…
Demo at MIT CPW 2015:
Higher-performance run with better tuning and longer bang time using modified firmware on an original oneTesla interrupter:
Even though big IGBT bricks are getting pretty fast and pretty cheap these days, I really like the idea of a medium-sized DRSSTC powered by TO247 IGBTs because you can get the same results with good design as you would otherwise get through brute force of having a gigantic, invincible transistor die. To push the performance even further, I’m planning to add a second bridge to drive the resonator through a split primary-capacitor arrangement so I can get twice the primary current.
Since I want to keep this DRSSTC running off of an ordinary mains outlet, I’m planning to add an active PFC rectifier to the H-bridge board so that I can eliminate the variac in my setup, push the bus voltage a little higher to 400V (right now it’s nominally 340V with a voltage doubler, 400V is basically ideal for 650V-rated parts), and keep the current draw at the limits of a typical residential breaker (15A or 20A, right now a single bridge draws an average of roughly 10A, with an active PFC I’ll be able to actively limit the average power draw).
Stay tuned for more updates!