Kelly KDH14651B post-mortem

There were quite a few things happening since last time and I was so busy working on them that I could not spare time to update the blog timely. My excuse finishes here :)

OK, the main thing was that my Kelly controller KDH14651B blew up in smoke. The good thing about it that after all reading all comments from around the net I was kind of expecting it. In a way I felt a relief that I don't have a doubt about its reliability.
Actually it happened at the end of September when I was drive-testing my HR-EV. I was driving it around for a day to test the power consumption and battery capacity. After fresh charge it scored 53km and 203Wh/km consumption in Vilnius city-like driving profile (quite a few ups and downs) which was quite good comparing to standard 250Wh/km as a benchmark. The main parameters during this event were following: speed ~70km/h, slight up-slope driving for around 2-4 previous minutes, battery voltage ~124V, battery current ~180A, controller heatsink temperature ~+29C. So it blew up while continuously delivering around 22kW power for last few minutes. Just to note that the highest power possible I recorded with Kelly was 43kW.
Kelly were not willing to provide me some money compensation. Only offer was a notable discount on new controller which would be basically the same one as just blew up. This was not an acceptable option. Here I'll put some photos what I found investigating its failure.

Disclaimer: Of course it is not proper engineering failure analysis so I reserve my right to be wrong at some of my conclusions. I am basing my findings on my own electronics knowledge and experience.

First I took out my control box from the car. Looking closely you may see the Kelly's cover is slightly popped open from inside explosion.

And here is the photo to see it closer. The cover was just glued to aluminum chassis.

When taken the controller out of my control box I could open the cover more widely. A lot of black ash inside.

Some ash left on inside of control box where the controller was sitting:

Of course the controller like many products of China was built in such way that it would not be possible to disassemble normally for repair. That is a reason that I have not heard of any repaired Kelly Controller. You just scrap the old one and pay Kelly for the new one. That did not stop me :). I sawed off the two sides of aluminum chassis to split top and bottom parts apart to have access to the whole board. And below is the photo how it looks with the top off.

Hmmm, first impression is bad. Not just because of black ash all over. The layout of components is really bad if you are dealing with multi-hundred A currents switching. The electrolytic capacitors sit on one side and power connectors on the other with many switching transistors in between. This layout is a recipe for uneven current and voltage spikes, temperature distribution and stress on the components. There are 42 IRFB4227 transistors in this controller connected in a half-bridge configuration: 21 on lower arm and 21 on upper. The basic rule of paralleling is to keep the conditions as equal as possible for every paralleled part. This rule was clearly taken carelessly.
In addition to that there is really "interesting" solution to increase the power of the controller taken closely in the picture below.

As you can see there are two 10x1.5mm copper bars connecting the bigger and longer row of power transistors on the left to smaller row on the right. This gives me a clear hint that originally the controller was designed to handle maximum currents of up to "400A" and needing a more powerful model they added this section to extended the current by ~200A-ish to total "650A". That would be OK if design approach was taken properly and wide connector plates were used to ensure small inductance and, again, even distribution. Having a long bar with 15mm2 cross-section would certainly not help to achieve that. In hundreds Amp currents every 0.1uH counts.

Closer look at possible failure cause indicated that the epicenter of explosion was first transistor which was closest to bar connectors. Based on previous observations this surely looked like a most probable place to fail: closest to the load and furthest to filter circuits.

It was clear that this transistor was producing majority of bang and smoke.

 The PCB track beneath it was also severely damaged. I reckon that after it's failure all other transistors failed in avalanche fashion which is quite normal for paralleled designs. Out of 42 I found possibly 3 that were showing good initial parameters.

Another point is that it is very important to place good filter capacitors on power lines of such switching devices which could help to reduce voltage spikes and even the conditions for paralleled designs. For that you would need good film capacitors. I didn't find them. I only found ceramic SMD capacitors stacked in several places. These usually not good enough for high current spikes filtering.

The next thing was looking at the attachments of these transistor for heat transfer. For that I was taking off bus bars from the transistors. The bus bars are 3x16mm giving 48mm2 cross-section. This is usually good enough for transferring currents from A to B of 600A from which could potentially be drawn from battery. But on controller itself this should be by a margin bigger to allow more even conditions distribution (again) due to smaller resistance and inductance. Anyway I found that each 7 H-lower arm transistors were attached on L shape 3mm thick aluminum plate which is in turn bolted to the chassis base through PCB. Notice anything wrong? Yes, the PCB was used as a heatsink transfer element! OK, the PCB has perforation with metallisation to build many heat transfer channels but still it is not sufficient to drain the excessive heat from transistors down to the base plate at good rate. Below is the photo of PCB what I found under plate and bus bar.

This makes the excess temperature be the main contributor to failure. Remember, I was driving for few minutes in slight uphill. This allowed the transistor temperature to rise but transfer to baseplate and heatsink was not sufficient to keep transistor in normal operating temperature range.

Here is how the PCB looks from below.

Taking into account MOSFET's increased heat dissipation at increasing temperatures this heat transfer certainly does not look enough to me.

Here is how the aluminum case base plate from below PCB looks.

After these findings I certainly would not buy another one from Kelly even at half the cost because I know it would blow up again. It might work well in golf-kart or similar but it is not suited for high currents.

After that I left without controller. As there are no acceptable options (reliable, not overpriced, flexible, powerful enough, with easy integration, available within 1-2 week lead time, etc) I decided to build my own which would be based on IGBTs. But that is the topic for different post...