DC-Link Capacitor Precharge

Background

This investigation is the result of two independent projects: retrofitting a siliXcon SX controller to the EM 5.7 and also reverse-engineering a dead 5.7 BMS.  

It was not at all clear to me how the 5.7 battery would precharge the large DC-link capacitors inside the controller, so I conducted some experiments.  Spoiler alert, it does not perform precharge!

Capacitor precharge is important because, otherwise:

• The only limit to inrush current is due to the battery's internal resistance and the wiring resistance.  The total resistance may be on the order of a hundred milliohms.  This implies a very high inrush current. 

• Rapidly charging/disharging a capacitor can shorten its life.  This is part of the design criteria in, for example, the capacitors for pumping a laser flashlamp. 

• High inrush current through the main contactor can degrade its contacts, and eventually they can weld themselves closed (resulting in an inability to turn off the system).

Kelly's Instructions

The photo below comes from the Kelly User Guide for the controller on the EM 5.7.  We see that Kelly specified a 1000 ohm / 10 watt precharge resistor across the main contactor.  They say, “All contactors or circuit breakers in the B+ line must have precharge resistors across their contacts.  Lack of even one of these precharge resistors may severely damage the controller at switch-on.”

I was always puzzled as to how precharge was being accomplished by the 5.7's battery.  When the battery is switched off, there's no voltage at the Anderson connector.  With a precharge resistor across the main contactor, we would see battery voltage that was current-limited by the 1k resistor.  According to I = E/R, the current would be a maximum of about 50 milliamps (54V / 1000 ohms).

5.7 Battery, SX Controller Power-Up

While retrofitting the SX controller to the EM 5.7, I decided to see what the capacitor precharge waveform looked like.  It confirmed my worst fear.

In all the following oscilloscope captures, the 0-volt reference line is denoted by the “2->” near the bottom-left of each photo.  All plots use 10 volts per division for the vertical axis.   

This first plot was taken with a timebase setting of  100 uS per division.

We can see that it only took around 200 microseconds to charge the DC-link capacitor from nearly zero to 50 volts.  siliXcon specifies the DC link capacitor in the SX controller at 1950 uF.

Capacitors experiencing R-C charging follow an exponential curve.  The curve below is so steep, it looks almost linear. 

Working backwards using the exponential equations for R-C charging, the total resistance would be about 100 milliohms.  This implies an instantaneous surge current of about 500 amps! 

5.7 Battery, Kelly Controller Power-Up

I then decided to see what the charging curve looked like for the original Kelly controller.  No specifications are given for its DC-link capacitance.  I measured capacitance using two different instruments which resulted in ~7200 and ~9000 uF.  (These values are at the extreme range of both instrument's capability.)

It took over 1000 uS to charge the Kelly controller's capacitors.  This is approximately 5 times longer than the siliXcon SX controller using the same battery.  So it makes sense that the Kelly's capacitance is roughly 5 times greater than the siliXcon SX's capacitance.

I have noticed that sometimes the 5.7's battery will switch itself off immediately after it has been switched on.  I'm assuming this is because the BMS detects a large current surge.  Usually, after cycling power again everything is fine, but sometimes it takes three attempts.  The worst case occurs with a fully charged battery and fully discharged capacitors.   Theory tells us that at t=0, a capacitor's instantaneous surge current is defined by I = E/R (and decreases exponentially).  

EM 5.7 BMS

As part of my reverse-engineering effort, I was provided with the information shown below.  It's from a dealer document that describes an addition to the 5.7's BMS.  Apparently, a component was to be inserted in the wiring for the discharge contractor coil and charge contractor coils.  (Note the black blob between white connectors in the rightmost photo below.)  These components were describe as being resistors.  I don't have these components and I don't know what they were intended to accomplish. 

But after seeing the lack of any precharge capability on the 5.7's battery, I wondered if they could somehow affect precharge?  It does not look like the extra components are present in the photos I took of my battery when it was apart.  If they were intended to affect precharge, I figured EM might have learned something and their later Race model would be different.

EM Race Battery, SC Controller Power-Up

So I decided to examine the battery switch-on behavior of my EM Race, which uses the siliXcon SC controller.  Here the DC-link capacitors are specified as being 1980 uF.  It took only about 400 uS to charge.

This waveform looks pretty funky.  Some of that may be due to inductive ringing in the measurement wiring as I was not able to easily probe directly across the controller as I did for the 5.7's measurements. 

Dragonfly, Power-Up

Just for comparison, this is how it should be done.  Charging the Dragonfly's capacitors takes three orders of magnitude longer than than the EMs.  Note that the oscilloscope timebase is 50 milliseconds per division.  The Dragonfly uses the same siliXcon SC controller as ePure Race, so both DC-link capacitors are the same at 1980 uF.  At the risk of belaboring the point:

EM Race charging duration: 500 microsconds. 

Dragonfly charging duration: 400 millisconds.

Controller Capacitor Self-Discharge

The rate of controller capacitor self-discharge varies with capacitance and anything drawing current inside the controller (even when it's not enabled).  I tested the controllers on the bench with no external connections. 

The Kelly KEB discharged fairly rapidly, dropping from battery voltage (49.5 volts) down to 6.2 volts in 20 minutes. 

The siliXcon SX took much longer.  I charged it to 50 volts with a bench power supply in order to make a direct comparison with the Kelly.  After 4 hours, the SX was only down to ~36 volts.  After 24 hours, the SX was down to ~32 volts.  After 48 hours, the SX was down to about half voltage (25 volts). 

This clearly presents a hazard when working around the controller's power terminals.  The procedure siliXcon outlines for removing a controller for any maintenance is as follows:

It is imperative to discharge the inverter's DC bus capacitors following the instructions below.  Do not short the +/B+ to B- terminals, as it may lead to an arc.  After disconnecting the battery, allow a 5-minute waiting period for the self-discharge of the DC bus capacitors, or alternatively, apply a 100Ω 10W resistor between the +/B+ and B- terminals for 15 seconds to facilitate capacitor discharge.  Confirm that the voltage between +/B+ and B- is below 10V DC.

Although in some applications the self-discharge may occur rapidly, my bench experiments indicate the SX caps could still exhibit half of battery voltage many hours later.  In fact, the capacitors still exhibited 1.8V after the controller sat idle for many day.

Inrush Current Waveforms

Both oscilloscope plots below have a vertical scaling of 50 amps per division and a timebase of 50 microseconds per division.  In the left photo, the current probe's bandwidth was limited to 5 MHz.  The right photo is using the instrument's full bandwidth of 15 MHz.

We can see the peak current is in excess of 300 amps in the bandwidth-limited photo, and in excess of 350 A when using full bandwidth.  Both plots show anomalies that may be due to “switch bounce” in the contactor and/or effects of inductance in the wiring.

The takeaway is that a very high inrush current is observable even when starting with a charge of 25 volts on the capacitors.

Incidentally, the waveform that can be seen in the instrument setup photo is 1 amp per division with the battery switched-on when the capacitors were already fully charged.

Mitigation Strategies

Even though Electric Motion's method does not result in immediate catastrophic failure, it's still not correct. 

If the 5.7 controller were to fail, it fairly cheap/easy to purchase a new one directly from Kelly.  Replacing the contactor inside the battery would be more difficult (assuming the part is even available).  However, the retail price of the ePure's SC controller is around $1800.

As for what I can do to fix this, it would be possible to install a precharge resistor and momentary contact push button inside the 5.7's battery case.  This might be much more difficult inside the ePure's battery due to a lack of space. 

For now, I think power-cycling the battery should be kept to a minimum.  Previously, I thought nothing of turning off the 5.7's battery if the bike was going to sit for 20 minutes.  Bad idea.  Similarly, I'd sometimes power-up the battery just to make it easier to load the bike onto the trailer.  No more.

Of course, all of this testing was with the controller capacitors close to fully depleted (near 0 volts).  Which is the worst case.  If the capacitors have some residual charge, it's much easier on everything to go from, say, 50 volts to 54 volts.

One mitigating factor in siliXcon's favor is that the  controller draws almost no current when not enabled.  The caps are still at half voltage a day later.  So they are not charging up from zero, like the Kelly.  The Kelly controller is brutal to the contactor.

I also think pushing the bike could charge the caps a bit because of  back-EMF generated by the motor.  I notice the ePure's tether LED lights up when I roll the bike off the trailer.