Please click here to link back to the first instalment of my CD Welder development.  This page is a continuation of my trials and tribulations.

12 January 2012 and sometimes there is nothing like a clean sheet of paper (or for that matter a blank web page) to reinvigorate ones thought processes.

My welder has been reassembled and I have rewired some stuff to move the ground reference to the negative supply rail to aid oscilloscope measurements, written a few lines of code to exercise the main MOSFET gate drive, removed the existing snubber, incorporated an 18 Ohm resistor in the gate drive, and set about making some measurements.  Everything is working.

Figure 1 shows the new main MOSFET gate drive voltage waveforms.  The pictures are very dark but this was the only way I could capture the transients in a photo for you.  You will see that there are very brief (~20 ns) 2 Volt transients right at the beginning of the transitions.  These are due to the reactance in the fast switching circuit and, although benign, they have been eliminated by ferrite beads on the driving MOSFET drain leads (it is almost always best to remove glitches like these as close to their source as possible).  My snubber design is now redundant.



Figure 1.  New MOSFET Drive.  Horz: 1 us/div, Vert: 5 V/div


The maximum gate dVgs/dt on turn on or off has been reduced from about 350 V/us to just 5 V/us.

I’ve completed about twenty 1,000 A discharges with no problems.  See Figure 2.  The transitions are still fast compared with the weld pulse durations with absolutely no apparent transients.  You might like to compare this with Figure 62 on the previous page where the rise times are not as clean as they are now.  Figure 2 was measured across the bus terminals with no weld cables attached.  For Figure 62 the measurements were made across the probe tips and I suspect that the leads were coiled (a sure way of reducing current rise and fall times - and not good welding practice).



Figure 2.  1000 A Test.  Horz: 5 ms/div, Vert: 5 V/div
CVSet: 12 V, P1: 3 ms, D: 5 ms, P2: 10 ms, Rload: 0.01 Ohms


Nothing untoward is getting hot so it’s time to get back to my weld schedules and see if I can make those stud welds.

Time for another MOSFET (make that several)

Well I got about 40 or so moderate stud welds made at increasing voltages from 10 V to 13 V with P1: 1 ms, D: 1 ms, P2: 10 ms (Figure 3).   After a quick break I thought I’d get right back into it.   When placing the electrodes there was a big spark and I’m faced with at least another dead MOSFET (maybe more).  This seems to be very similar to my last failure experience.  Surely I have fixed di/dt during turn off so that shouldn’t be the problem.



Figure 3.  Some Stud Weld Tests
10.5 V to 13 V Right to Left


Let’s review what has occurred here.

  • My earlier design was switching fast from 20V during Pulse 1 with no failures (other than my heat sink related issue) but the energy was limited because of the crapacitors.
  • While I destroyed a single MOSFET through suspected avalanche with my new capacitors, I only destroyed one MOSFET which undoubtedly protected the others.  However the remainder may have been de-rated.
  • I have measured Vgs threshold of the surviving 12 MOSFETs at 150 uA, 5 and 10 mA.  They are all within 53 mV and have a similar slope.  I have measured Rds at a gate voltage of 5 V and a drain source current, Ids, of 0.5 A.  They are all within 0.0004 Ohm.   It might be that these MOSFETs have been matched through Darwin’s theory of natural selection - but I doubt it.
  • This failure event has destroyed five MOSFETs.  They all failed dead short drain to source but they may not have failed for the same reason and at the same time.
  • The gate drive is completely clean relative to ground (this has been physically measured over thousands of weld pulses).
  • The welder has been tested with numerous 1,000 A welds and the waveforms are clean (Figure 2).  Faults are only occurring during welds at much higher currents.  Maybe I need to use higher currents in my measurements?
  • The charge and discharge circuits are still working faultlessly at about 10.2 A, as are the processor board and power sense boards.

Mark  has suggested that maybe I should be using matched devices and kindly provided an IRF Application Note on paralleling power MOSFETs which supports this.  While my existing MOSFETs come from the same date stamp batch they certainly weren’t matched for Vgs threshold or Rds(on).

There are a few other possibilities.  One is that my previous suspected avalanche experience may have degraded some of my existing MOSFETs without actually destroying them (this phenomena is documented in the literature).  Another is that certain weld conditions are causing an oscillation or some other transient.  A third is a continuation of my avalanche problems.

The bus was extracted from the chassis in under five minutes and within 30 minutes I had found and removed all five dead MOSFETs, all shorted drain to source with two also shorted to the gate.  There is no physical evidence of device failure this time.  This is a moderately expensive failure because I don’t have enough IRF3206’s for replacement.  So I’m up for some new MOSFETs.  Maybe I can find something slightly higher rated for this application?

13 January (revised 19 Jan 12).  On reflection the recent failure is probably a consequence of inductive voltage transients during turn off.  Here is why.

During turn-on the current rise is limited by the discharge path inductance which I have estimated to be at least 0.5 uH.  This ensures current sharing between MOSFETs and SOA is assured.

My recent failure occurred at a capacitor voltage of 13 V.  With real capacitors this voltage can be considered constant during the few microseconds of switch transition.  The discharge path resistance in my welder has been measured at about 0.002 Ohms so the maximum possible total discharge current is somewhere around 6,500 A.  The energy stored in the discharge path inductance is about 40 J.  When the discharge circuit turns off this must be dissipated somewhere.  Even with relatively slow 5 us switching the induced EMF during turn-off under these conditions is about 2,600 V so I should have anticipated MOSFET avalanche with sufficient energy for destruction.

I have found some new MOSFETs (IRFP4368) that are significantly more robust than my existing devices although they are bit slower on transitions (not necessarily a bad thing).  They are not too expensive in 50 and up quantities which will also allow me to match them for Vgs and Rds.

If I also incorporate a hefty freewheeling diode across my bus structure and a few Transil Transient Voltage Suppressors (TVS) between the drain and source, physically distributed along the bus.  These should absolutely prevent avalanche.  It appears that other folks have had bad experiences with TVS suppressors so I cannot say that I have not been warned.

15 January 2012 and I have reviewed this page and it’s predecessor.  Aside from a number of typographical and grammatical mistakes I have found a number of invalid numerical assumptions which affect my ”back of the envelope” calculations and their consequential results.  I have endeavoured to correct these today.  If you read something on these pages which you don’t agree with, that defies logic, or understood physics then I have probably made another mistake or poor assumption.  I welcome (indeed I invite) your challenges, questions, criticism and enquiries.  Please contact me with your comments.

More Challenges

17 January and over the past two days I have been attempting to model the transient high current issues that have been cooking my discharge MOSFETs.  While I have made some progress, the use of lumped components and the difficulties of measuring actual circuit parameters makes this a particularly difficult and imprecise exercise.  I suspect that I will need to fit my replacement MOSFETs to allow physical measurements at gradually increased discharge currents to make progress.

I have received some very useful and thought-provoking correspondence from Chris K. which has made me rethink a number of issues (thanks Chris).  Chris has pointed out that the significant discharge path inductance is in the probe leads.  It follows that these should be provided with a fast commutating diode.  I had already come to this conclusion but my computer simulation experiments have resulted in particularly nasty oscillations during turn off.  This needs some more work.

I have also reflected further on others’ experiences with Transient Voltage Suppression (TVS) at high currents.  This is a specialized realm of power electronics where my previous experience is at best limited.  I have never had cause to apply TVS in the form of Transils before so I am on another learning curve.  I can see through my modeling why some folk have literally vaporized their protection but this hasn’t resolved my specific design issues.

Welder Delivered Energy

While I’m waiting for my replacement MOSFETs I thought I might expand on the issue of actual energy delivered to the load.  There are a few web sites and commercial manufactures that advertise welders with ratings of 600 Ws or more.  These ratings are invariably based on the total energy stored in main discharge capacitors.  You may be familiar with the expression for energy stored in a capacitor:


However with both inductance and resistance in the discharge path and the weld junction, and with timed discharge pulses, the full energy stored in the capacitor will not (cannot) be dissipated in the weld junction.  With an ideal power source maximum power transfer occurs to the load when its reactance is the complex conjugate of the source impedance.  If we forget for a moment about discharge path inductance this means that the weld junction resistance must equal the discharge path resistance.  However we don’t have an ideal power source (our capacitors discharge over time) so the maximum possible energy dissipated in the weld has an upper bound:


If you have measured the discharge path resistance of your welder then you can use the attached spreadsheet to estimate optimum pulse width and capacitor voltage to deliver the maximum possible energy into a weld junction of known resistance.  You can also estimate the maximum energy that your welder can reasonably be expected to deliver.  There are a number of techniques for dynamically measuring the weld path resistance and weld current.  I may yet incorporate these in my design, but for now they are future refinements.  Please be aware that this analysis is a theoretical model based on estimated parameters and ignoring the discharge path inductance.  We also know that the resistance of the weld junction is a function of time and a heap of other unknown variables.  Still, you might find it interesting.

September 2012 update.  I have written a new Excel model that incorporates inductance based on a simple LCR discharge circuit.  Click here to navigate to the How Much Current? page.


While I would have liked to use some Visual Basic in this spreadsheet to calculate the maximum delivered energy to the weld and the optimum weld path resistance (assumed constant) this might upset your browser security.  I have use a clumsy iterative technique for finding the maximum deliverable energy.   If you are familiar with ’Goal Seek’ or ’Solver’ in the Excel environment then these plug-ins will simply confirm my clumsy iterative method.

18 January and the first two consignments of my replacement MOSFETs have arrived (different manufacture date codes, but my measurements show that this is not important).  I labelled each device and set about measuring Vgs and Rds(on).  With relatively fast switching of the gate Vgs matching is not going to bring about a significant performance improvement, but with my linear bus structure matching Rds(on) and positioning devices with the highest Rds(on) resistance closest to the weld electrode connections is expected to improve current sharing between devices in the fully on state.  While Rds(on) is defined for your particular MOSFET you may want to measure this at a much lower gate voltage for the purposes of ensuring equal conduction between devices at turn on, particularly for slow gate drives.

I am still waiting on a few additional bits for my transient protection but I am confident that I can now reassemble the main bus and compete a range of further tests at moderately high currents without crashing and burning.

Estimating Lead Inductance

I have alluded to the inductive voltage spike issues that are caused by lead inductance during turn off above. A question you might ask is what is the lead inductance on your welder?  While you can measure this using an LCR Bridge it will vary depending on the relative lay of the two conductors.  There are three configurations that are of interest.  The first is two parallel conductors, the second is a single circular loop, and the third is a three turn coil (I can just manage to get about one turn in each of my electrode cables and the weld junction makes for the third turn).  The relevant approximations are:


If you study these equations you will see that you don’t want any loops in your cables.  You want to keep your cables together (minimize the area between the cables), use large diameter cable with lengths as short as practicable, and you don’t want any other metallic objects in the cable loop.  Ferrous metals will increase the effective inductance and non-ferrous will have induced eddy currents that will waste weld energy.  Doing the numbers for my welder with 1.3 m lengths of 0G cables I figure that:

  • parallel leads (separated by a double thickness of insulation):  0.6 uH
  • single turn (close to my usual configuration):  2.3 uH
  • three turns (definitely not recommended):  5.3 uH

I have estimated the inductance of my parallel capacitors as a little over 0.5 uH and the bus structure is a negligible 50 nH (this is another advantage of my bus structure).  So you can see that the discharge cables are the dominant contribution to the discharge path inductance.

While some inductance and resistance in the discharge path is useful because these slow the rate of current rise contributing to current sharing between your MOSFETs, when you turn the discharge current off, the energy stored in the inductor needs to be dissipated somewhere.

20 January 2012 and I have again reviewed this page and it’s predecessor resulting in a number of further revisions - largely due to my failure to incorporate discharge inductance in my earlier calculations and estimates.  Dudes, please let me know if you think I have made a stuff up - I would really appreciate your comments and criticisms.  My new MOSFETs are fitted but I cannot complete the bus assembly until my Transils and free wheeling high pulse-current Schottky diodes arrive.

2 February and my transient protection has arrived at last.  Placing these components exactly where I want them requires rework of the bus structure which I have been working on today.  The Transil axial leads are too thick so these have been peened to about same dimension as the MOSFET lead thickness.  The diode needs a lead connection to the Anode.  I have settled on using a stranded cable with a sectional area of 5.5 mm square mm which has a resistance of less than approximately 0.0001 Ohms and a 32 ms fusing current of about 10,000 A.



Figure 4.  Diode and Transil Protection on Bus


3 February and MOSFET bus has been reassembled but in my haste to be making welds I forgot to install the gate and source connections.  I disassembled and remade the bus, did not complete my usual continuity checks, and on power up I got an over-temperature warning on the charge board.  On investigation there was a drain to source short on the main discharge MOSFETs.  On disassembly I found that the MOSFETs and protection were all OK but there were three issues: I had failed to install one insulating sleeve in the bus, the insulation between the mounting brackets and the drain was not ideal, and one of the Transil leads appeared to have contacted with a fastener.  This is a consequence of me being in a rush and also squeezing additional components into a design in which they were not intended.

While the over-temperature alarm worked, the charge board was not designed to drive a short circuit indefinitely and by the time the alarm sounded the charge board MOSFETs were already toast.  The charge MOSFETs will dissipate 300 W at 25°C but de-rate at 1 W/°C and, under short circuit conditions, are already dissipating 295 W.  As a consequence the discharge board MOSFETs have also gone West.  This is essentially the same cascade failure that occurred on the 6 December  and just like last time it is not a design fault with these boards - they are being forced to operate well outside their design specification.

So what next?

  • I can redesign the charge board to sustain an indefinite short circuit very easily (the problem is not the power rating of the MOSFETs but getting rid of the dissipated heat).  The existing boards use parallel MOSFETs in T0 220 cases.  However the case to heat sink thermal resistance (Rcs) of this package is typically 1.1 °C/W with greased mica washers and this is the limiting factor for heat dissipation.  So I need to go to a thermally more efficient case like a TO 247 or TO-3P with a typical thermal resistance, Rcs, of 0.25 °C/W.
  • The cooling fan direction is going to be blowing all of that warm air around in the case.  Before I put a lid on case I’ll need to go to induced flow, incorporating some ducting and blow the hot air outside the case.
  • The temperature sensor on the charge heat sink should be moved as close as possible to the charge MOSFETs.   At the moment it is about 40 mm (1.5”€) away.  This is a significant delay in the thermal fold-back limiting that I have implemented in the software.
  • The existing charge board design has an additional thermal feedback mechanism that can be used to reduce the power dissipation by about 1 W/°C through the negative Si junction voltage coefficient of the current sense BJT (about -2 mV/°C).  At the moment this transistor isn’t thermally coupled to anything.  While I could mount it beside one of the current sense resistors their thermal response is too slow to be useful for short term response.  The best place for this BJT is right on the MOSFET package but this means mounting off board so I won’t be implementing this.
  • If the charge board MOSFETs are changed then there should be no reason to alter the discharge board.  The maximum discharge power dissipation will be just over 200 W and will reduce at almost 40 W/s from fully charged 3 Farad capacitors.

In order to get welding again I found two TO 220 package 300 W P Channel MOSFETs left over from an earlier project.  I set about installing these on the charge board followed by some bench testing.  Something really odd was happening - the charge board was leaking 53 mA at -12 V Vds, 0 V Vgs, and almost an Amp at -24 V Vds.  This just wasn’t possible under the test conditions!  It turns out that one of MOSFETs wasn’t actually a MOSFET at all but a three terminal regular (maybe I should have actually read the part number before installing the device).  While my supplier has almost certainly given me the wrong part (and made a tidy profit in the process) I am somewhat embarrassed by my own failing to check the part number.  Worst of all I have no other P channel MOSFETs that are suitable for this application and tomorrow is a public holiday.  It will be another few days before I am welding again.

6 February (revised 8 February due to a silly mistake).  Here is the thermal design for the charge board MOSFETs.  This is a different design problem from the transient (pulse) thermal considerations for the main discharge MOSFETs because we are now interested in steady state performance.  I have selected two Fairchild FQA36P15 devices operating in parallel.  They come in a nice big TO-3P package that will fit with my existing circuit board and they are not too expensive.

Assuming current sharing (devices will be matched for Vgs/Ids), under short circuit conditions each device will dissipate somewhat less than 25.6 V * 11.7 A / 2 devices = 150 W. This is worst case because the charge current is typically 10.2 A and the charge voltage drops under continuous load so we have 20% conservatism right at the start of the design.

We need some thermal information which comes straight from the heat sink and MOSFET data sheets:

    Tj(max) = 175°C
    P(max) = 298 W (Tcase = 25°C, derated at 1.96 W/°C ~2 W/°C)
    Rjc        = 0.51 °C/W (max) junction to case
    Rcs       = 0.24 °C/W (typical) case to heat sink
    Rsa       = 0.55 °C/W heat sink to ambient (still air)

Note that Rcs is not specified as with or without an insulating washer or thermal paste and it is not clear to me from the data sheet whether the tab is fully insulated or connected to the drain (turns out it is not insulated so electrical insulation is required).  The heat sink is fan-cooled at a linear air flow of approximately 150 m/min.  This allows a reduction in Rsa by a factor of about 0.33.  The temperature calculations are straight forward.  While I have worked from right to left starting at ambient temperature you could also work from left to right starting with Tj(max).


So Tj(max) is theoretically exceeded by about a 18°C.

We also need to check the MOSFET power de-rating.  The maximum power dissipation is reduced by (116°C - 25°C) * 2 W/°C = 182 W so the the maximum power dissipation reduces from 298 W at a case temperature of 25°C to 116 W at 116°C.  This is below the desired 150 W power dissipation in the steady state.  At first glance it looks like my heat sink just isn’t big enough (or it needs greater air flow, or a more exotic cooling mechanism such as fluid or Peltier).

However, we are designing for a condition that only occurs on main discharge MOSFET or capacitor failure, we have our initial 20% design conservatism and:

  • fold-back limiting when the heat sink reaches 50°C,
  • the reduction in constant charge current due to thermal coupling of the BJT to the current sense resistors (estimated to be about 80%),
  • forced air flow over the MOSFET case, and 
  • and the charging duty cycle (about 95%).

So I can be reasonably confident that the charge board will actually be operating at about 2/3 maximum ratings and that it should now shut down before it over heats.

8 February and you may have noticed an error in my thermal analysis from 6 February (I forgot that the heat sink is dissipating heat from two MOSFETs, not one). Despite my open invitation for your criticism no one seems to have noticed, and in any case I have corrected it now.

I have about five other projects on the-go at the moment and one of them needs a prototype board yesterday.  I took the opportunity and remade both of the CD Welder charge and discharge PCB’s in order to get the current sensing BJT thermally coupled to the source resistors.  Only two minor issues of note: my enchant has past its use-by-date resulting in forced etching and some undercutting on the PCB.  I also managed to miss-route a very short low current track on the discharge board so I won’t be making a new one of these.

9 February and the everything is back in the case and making 520 A pulses.  Figure 5 shows the new charge board with the big MOSFETs and the current sense BJT thermally coupled to the source resistor.  Although not visible the MOSFETs are now about 12 mm (1/2” €) from the heat sink thermal sensor.   You’ll see a few extra holes in the heat sink which is not ideal but shouldn’t be critical.  Remember to get appropriate mounting hardware for your MOSFETs.  These mica washers were cut down from a TO 3 package.  The circuit has been tested now over a number of rapid repetitive moderate current pulses and there is no leakage and nothing is getting unduly warm.



Figure 5.  New Charge Board with Thermal Limiting and Fold-back


Figure 6 shows a moderate 520 A current dual pulse (CV: 18 V, P1: 1 ms, D: 1 ms, P2: 12 ms) pulse.  The pulses are clean with fast transitions, no overshoot or oscillation.  Note that the welder leads have yet to be connected.  Using the change in capacitor voltage over the 13 ms discharge time the internal discharge resistance is estimated to be under 0.002 Ohms.



Figure 6.  520 A Pulses.  Horz: 2 ms/div, Vert: 5 V/div
CV: 18 V, P1: 1 ms, D: 1 ms, P2: 12 ms


Tomorrow I’ll be putting the welder leads back on and gradually increasing the weld current in increments of about 500 A to see what damage I can do.

10 February and I have progressively increased my weld current to 2,400 A through the welder cables without mishap today.  Figure 7 shows the discharge.  You can see the brief inductive voltage transient as each pulse turns off, clamped to about 7 Volts.  Interestingly my circuit model is performing remarkably close to actual performance so far (this is good news because I have run the model right up to 7,000 A discharges with no component overload issues).



Figure 7.  2,400 A Pulses.  Horz: 2 ms/div, Vert: 5 V/div
CV: 17 V, P1: 1 ms, D: 1 ms, P2: 12 ms


During these tests the first pulse is necessarily short (optimally about 0.7 ms) to ensure that the main MOSFETs are switching off the maximum possible current for a given set of weld parameters.  The second pulse is somewhat longer to ensure a reasonable capacitor discharge which allows me to calculate the discharge path resistance and therefore the weld current.  Interestingly the internal discharge path resistance has decreased during these higher current welds to about 0.0006 Ohms.  I expect that this is a consequence of higher currents improving the physical bus connections.

While the only welds that are occurring just now are between the test load and the weld probes I’ll continue with this progressive approach for now.

12 February and I haven’t managed to find the time to complete my weld tests this weekend.  In my last set of tests the discharge currents should have been significantly higher. I have concluded that the problem is with how I am attaching my test load to the weld probes - these connections formed really good welds which is indicative of their relatively high resistance.  Hopefully I can make some progress with my tests tomorrow.

13 February and I’m starting to get some serious weld currents at last.  Figure 8 shows the test load with earth clamps to ensure a good physical test load connection.  Figure 9 shows a 3,400 A discharge from 18 V.  The total discharge path resistance is slightly over 0.005 Ohms.  The resistance internal to the welder remains at 0.0006 Ohms.  The test load has a calculated resistance of 0.001 Ohms (at 25°C) and the weld leads account for another 0.001 Ohms.  I suspect that I want to remake the some of my cable terminations, cleaning the copper before reassembly, otherwise the maximum current will be limited to about 4,200 A.

I figure that the energy delivered to the test load from these pulses is about 37 J.  This is enough to make some serious steel welds but can be increased significantly by reducing the discharge path resistance, increasing the pulse width and, if necessary, increasing the charge voltage to 20 V.



Figure 8.  Test Load (seven strands of 0.63 mm diameter copper wire, 130 mm long)



Figure 9.  3,400 A Pulses.  Horz: 2 ms/div, Vert: 5 V/div
CV: 18 V, P1: 1 ms, D: 1 ms, P2: 10 ms


If you look carefully at Figure 9 you will see that after the first 1 ms pulse the free wheeling diode is clamping the inductive voltage spike to about 9 V (this pulse is very brief - just a few us).  You will also notice that the capacitor voltage at the beginning of the second pulse is about 1 V higher than at the end of the second pulse.  The free wheeling diode is returning the energy stored in the lead inductance during the first pulse back to the capacitors, causing their voltage to rise.  This is good.  There is no oscillatory behaviour apparent on any of the discharge pulses.

After each set of measurements I turn everything off and set about analysing the video, looking at each pulse waveform for consistency and any oscillatory behaviour.  After reviewing the footage for Figure 9 I adjusted the test load and turned the welder back on.  Oops - the main capacitors were only charging up to 0.9 V?  This is strange.  The last thing the welder did before I turned it off was to charge up to 18 V.  I can’t have a bus short because the capacitor voltage would be zero into the test load.  The charge board is getting warm but it’s working fine, delivering just over 10 Amps.  None of the protection devices can have failed.

Actually the last thing my welder does when turned off is discharge the main capacitors followed by the power supply filter capacitor.  It looks like the discharge board MOSFETs have shorted drain to source during the discharge connecting the current sense resistors across the main discharge capacitors.  On power up when the charge board tries to charge these capacitors it supplies a little over 10 A which gives us 0.9 V.  It took five minutes to remove the discharge board and confirm that the MOSFETs are shorted and, as a consequence, the current sensing BJT has an open circuit base emitter junction.

While this is technically another failure it proves that the charge board will work into a sustained dead short and there is nothing wrong with the main discharge circuit or its transient protection, the power supply, the discharge control board or the MCU board.

So I’ll be remaking the discharge board after all, but why did the failure happen?  The charge board has been working hard during these tests as a consequence of numerous repetitive high current pulses from 18 V.  As a consequence of the recent redesign the heat sink (common to both the charge and discharge boards) is heating to about 42°C.  The discharge MOSFETs have an excellent power rating but they aren’t matched, they struggle to get rid of the heat fast enough because of their small TO 220 packages, and they de-rate rapidly at 2.2 W/°C (case temperature).  As a consequence I suspect that one junction got too hot and failed followed rapidly by the other.

As it happens I have a bunch of spare N channel MOSFETs in TO 247 cases left over from my earlier main bus disasters.  A couple of these should ensure that the discharge board will withstand a sustained rail short (just like the charge board).  Once the new board is made I will continue on from where I left off today, with further increases in discharge current until I get to a dead short load or until something else fails.  Unfortunately making the new board will have to wait until tomorrow.

14 February and the new discharge board is made (Figure 10) and mounted on the heat sink.  Unfortunately during assembly I managed to break one of the leads for the LM335 temperature sensor right at its TO92 case and I don’t have any spares.  While I could swap out the temperature sensor from the discharge MOSFET heat sink this would require a whole heap of eventual rework.    Hopeful the replacement sensor will be here tomorrow.



Figure 10.  Discharge Board (left) with Hefty MOSFETs


I’ve just got time for one more constructors’ note this evening.  When applying thermal grease you want to ‘wet’ all thermally bonding surfaces and exclude any air using as little thermal grease as possible.  With mica washers I usually put a dab on the washer and use a finger to make a thin translucent and even coating.  This goes onto the clean and dead flat heat sink and holds the washer in place.  Then I use the same procedure to coat the mounted side of the device.  Too much thermal paste (or old thermal paste that has become very viscous) is asking for trouble.  Using this procedure eliminates thermal grease all over the place but, just like using mustard, you will always use more than you need and the excess will end up in the trash.

15 February and temperature sensors haven’t arrived so I have made the swap and started where I left off.  One of the transils went short circuit at just below 5,000 A (Mark had warned me that this was likely).  Rather than pull the bus to pieces I simply cut the leads of the two devices and carried on making test pulses.  While the transil failure mechanism is clear I’ll be thinking about whether this will have any significant consequences for the discharge MOSFETs.

Figure 11 shows a nice clean 5,400 A pulse (the maximum current is right at the beginning of pulse 1) and Figure 12 shows the test load heating up after repeated 3,500 A pulses.  Note that the test load resistance increases appreciably as it gets hot and starts to oxidize hence the reduced current.  The weld leads are starting to jump a bit at these currents.  A few more pulses and the test load wires burnt out.

I am working the welder really hard in these tests with the next pulse firing just as soon as the capacitors have recharged.  If the welder is going to break then I want it to break during these tests.



Figure 11.  5,400 A Pulses.  Horz: 2 ms/div, Vert: 5 V/div
CV: 18 V, P1: 1 ms, D: 1 ms, P2: 10 ms



Figure 12.  Repeated 3,500 A Pulses into Test Load


My final tests for today were three consecutive pulses, each measured at just over 9,300 A into a calculated 0.0005 Ohm load (14 strands of 0.63 mm diameter copper wire, 130 mm long).  Woop! 

(September 2012 Update:  the actual discharge current in this test was actually closer to 8,700 Amps.)

Note that the waveform at Figure 13 is a composite of two successive frames video frames.  The pulses are still clean (they have clamped transients but I can’t see any oscillation) and nothing untoward is getting hot.  At these currents almost all of the capacitor energy is being discharged with a residual 2.5 V following the second pulse.  The current at the end of Pulse 1 is also down on the peak by about 25% due to the preceding 1 ms discharge.



Figure 13.  9,300 A Pulse.  Horz: 2 ms/div, Vert: 5 V/div
CV: 18 V, P1: 1 ms, D: 1 ms, P2: 10 ms
(Note:  composite image from successive frames)


It’s time to get back on with some welding practice at last!

16 February and my still is back in action after I completed the stud welds.  The negative electrode has left some cosmetic marks on the stainless steel but nothing that wont polish out and no leaks. The welds are strong and require leverage with a spanner to break them.  When they eventually do break bits of the stud head rip and remain attached to the stainless steel.

I made a whole lot of trial welds before fixing the still.  While adjusting the welder controls I had two instances of brief over temperature warnings that immediately cleared.  Nothing was actually hot.  I suspect that there is a software issue with changing ADC sources too rapidly compared with the ADC cycle time.  I will have to go back to the Atmel data sheet on this one.

I also attempted some welds to the heating element in a proportionally controlled hot plate.  I don’t have a spare hot plate to practice on so I thought I’d use some of my earlier schedules that were so successful with paper clips.  Wrong!  The energy levels of the welder are now way too high for the old schedules.  When I got the pressure right the welder melted more than half the diameter of the weld element so there was nothing left to impart mechanical strength.  When I got the pressure wrong then was a huge spark and the wire broke clean off.  Clearly I need to practice my technique before I undertake a real weld on something useful, and all of my earlier weld schedules are no longer of any use.

With an unlimited supply of distilled water I can get get back on track with my anodizing and wrap up the Buddy Locator refinements.  I still have a lot of work to do on the welder starting with replacing a temperature transducer, securing the main processor and discharge control boards with some thing slightly more secure than double sided tape, making a cowling for the fan and reversing it, and making the front and top panels.

I’ve been getting braver with my welder.  See the movie file below.  For the bridge wire experiments the resistance of each strand was 0.0028 Ohms.  Note that the four stand bridge wire is still connected for the can experiment limiting the current to somewhat less than 7,500 A.  After my attempted coin weld it is apparent that this was an unnecessary precaution.



Click Image to View Movie


Figure 14 shows successive video frames at 30 fps from two of the clips.  These were with a single 25.5 ms pulse from 18 V.  The first is an exploding 1 mm (0.04”) diameter copper bridge wire.  The second is my first attempt to weld two copper alloy coins together.




Figure 14.   Exploding Bridge Wire and Attempted Coin Weld


While the copper wire certainly exploded the coin weld wasn’t very successful.  The negative electrode melted half way though the top coin and a fair chunk of the probe tip had disappeared.  Although there is a reasonable amount of melting apparent between the coins it did not form a strong solid weld.  However if I can get the energy that went into vaporizing the probe tip into the weld junction then all will be good.  The welder continues to work without fault.


On the Death of Transils

28 February and I’ve been thinking hard about the death of that Transil back on 15 February.  It clearly failed because of exceeded energy dissipation limits (Amps * Volts * Time) with a weld pulse of just under 5,000 A.

My free wheeling diode is a Schottky device and therefore not subject to recovery time.  It should be switching on and off in a few 10’s of nanoseconds.  The forward voltage drop is relative small compared to the Transil’s turn on voltage, even with a very high current pulse.  How can the Transil ever end up with above its breakdown voltage, yet alone being destroyed?

I suspect the problem is the series inductance of the free wheeling diode and associated wiring which I estimate to be about 10 nH (the data sheet states a series inductance of typically 7 nH for the diode).  When I modelled the switch with Spice using lumped representations of distributed components I now have a   4,600 A peak, 1.3 us (to half peak current) negative exponential current  pulse as shown in Figure 15.  The voltage across the Transil under these conditions is just under 100 V so I am operating right at the Transils’ peak power limit for 1.3 us pulses, and the welder is operating well below 50% of deliverable weld energy.  The model is almost identical to the weld conditions when the Transil actually failed.



Figure 15.  Transil Killer


But why didn’t the other Transil share the current and also die?  The reason that the Transil did not current-share is because they were not matched.  Protek have provided this link on matching Transils (and LittleFuse have a similar Application Note).  They need to be matched to 1% of breakdown voltage at modest to high current.  The second device didn’t die because the first one had shorted - the expected failure mode.

The next question is what am I going to do about this.  While I have found some more robust Transils (30 kW at 10/1000 us) they will still be individually challenged even with a moderate 5,000 A weld pulse without matching.  I could also reduce the Transil voltage specification from 22 V to 18 V to reduce the peak pulse power.  However I will still need at least four matched Transils and they just don’t fit well on the bus.

So my answer is actually to nothing for now.  While I don’t like the concept of using the switching MOSFET avalanche characteristic deliberately,  the total energy in this inductive spike is around 1 J at 10,000 A.  The switching MOSFETs are staying at room temperature, and under these conditions, their individual avalanche energy is better than 2 J.

As an aside I seem to have had a 5 GB bandwidth excursion over the past 24 hours.  My ISP is looking into the problem.  Although this is beyond my control I know just how frustrating broken links and downed sites can be.  I apologize for this outage.

29 February and after some research I have found a Transil that provides 110 Volt  fold-back  clamping that will prevent MOSFET avalanche at  15,000 A (8 x 20 us negative exponential pulse).  The device will fit nicely into my existing physical bus structure with just one extra hole to drill.  I figure this device has a safety factor of about 5 for my current design peak discharge and they apparently parallel up without matching!  On the down side these devices are relatively expensive at about  US$50 each, they are not available off-the-shelf from the manufacturer, and the minimum order quantity is 56?  I have managed to track a few down from secondary sources.  Please Let me know if you are interested in a couple of these to prevent avalanche in your main MOSFET switch.  If there is enough interest I’ll be happy to make an OEM purchase and share the MOQ savings with you.  But maybe you you want to hold off until I have confirmed that they work first?  First in, first served.

Tomorrow I will be starting on my welder front panel.  I have about two weeks up my sleeve between consulting jobs, while my Buddy Locator corrosion trials with anodized transducers are finishing.  If you have never tried anodizing and have need of making some hardened, corrosion inhibited, pretty-coloured aluminium (US colored aluminum) then you want to give this a go.  The best possible link I can provide you with is by Ron Newman.  If you follow Ron’s instructions then your first results will probably be even better than mine.  What’s more your CD welder can also double up as a power supply for your anodizing line.



Figure 16.  Buddy Locator Transducers Anodized for Corrosion Tests


After 250 hours of salt water immersion these two transducers remain completely free of corrosion or discolouration despite deliberate attempts to abrade their surfaces using chrome plated stainless steel and periodic ultrasonic shock impingement.


More on Welding Coins

Goodness knows why you might want to weld coins together but they are a great visual demonstration of welder capability and produce big sparks and noise.  My coin welds haven’t been too successful.  As noted above the copper weld electrodes melt about half way through the coin before sticking, with a significant portion of each tip remaining embedded in the coins.  I’m using a step weld configuration with 18 V and 25 ms pulses.  See Figure 17.



Figure 17.  Close - but no Cigar


Between the coins I get a thin slurry of molten copper across the contact area.  I suspect that my manually applied weld pressure is too high but I also suspect that a part of the problem is the makeup of the coinage.  The 10 cent piece in Figure 17 is actually a copper coated steel blank (I can lift it with a magnet).  The weld electrodes are ripping right through the copper layer and becoming firmly embedded in the steel.  The conduction path through the steel is a relatively high resistance compared with the copper/copper interface between the two coins, hence the limited melting at the weld junction and deep electrode pits.  New Zealand’s other coinage isn’t going to help.  The silver coins are nickel on steel and the gold coins are some form of aluminium bronze.

Just like its predecessor, This page is getting a bit long for ease of editing and uploading.  If you have read this far and want to find out how the project is developing then please click here to continue or navigate to Pulse Arc.