Airwheel A3 broke, error code 7?

[mrelwood]

Airwheel A3 is a seated self balancing two-wheeler, that some handicapped people like me use in place of a wheelchair. Just to sum up its importance to my everyday life.

Today it broke. I was sitting on it at standstill, and gently leaned on a curb to keep me in place. Suddenly the A3 went dead and started beeping. While powered off and pushing it, it did have more resistance than usual, but nothing severe. I'm aware of past Msupers burning mosfets when starting to accelerate from behind an obstacle, but visually all mosfets (and all other components) are fine. Removing the battery for a while didn't help.

The A3 powers on and connects to the app, but it only gives out the continuous beep. Not a hitch from the motors.

App shows "error code 7", but I wasn't able to find what it stands for. I have sent email to the manufacturer for clarification, but I was hoping to get insights from you guys, just in case someone would have a thought on what should I check first. It is already past 2 years of age, so warranty is no longer valid.

[esaj]

On 6/23/2018 at 9:41 PM, mrelwood said:

Today it broke. I was sitting on it at standstill, and gently leaned on a curb to keep me in place. Suddenly the A3 went dead and started beeping. While powered off and pushing it, it did have more resistance than usual, but nothing severe. I'm aware of past Msupers burning mosfets when starting to accelerate from behind an obstacle, but visually all mosfets (and all other components) are fine. Removing the battery for a while didn't help.

If you can disconnect the motor cable, check if the motor resists similarly after that. If not, likely a dead mosfet, it may not show any signs visually from the outside but still be dead (short circuited). 

The issue with mosfet dying when starting from behind an obstacle is caused by very high current. When the motor's not spinning, it has 0V back-EMF, and applying battery voltage to it will cause the highest amount of current possible. That's why electric motors have highest torque at "0-speed" (ie. takeoff) and the mosfets heat up very much if "oscillating" in place (ie. switching between driving backwards and forwards), because when the motor comes to a standstill, the power peak to get it going again is at its highest. The "Zero" electric-motorcycles actually limit the maximum current not only to save the components from burning, but because the torque would get so high that the bike would backflip (they're said to have about as much torque as a stock Hayabusa, of course not even nearly the same max speed and much less weight).

If there's no logic in the firmware to limit the maximum current, and the motor cannot start to turn (due to the obstacle), either the motor or some component in the way (usually a mosfet) will burn.

[mrelwood]

Thanks @esaj. I was finally able to disassemble the vehicle enough to detach the control board for the resisting wheel, and the first MosFet (P75NF75) indeed didn't measure properly!

Now, should I change all 12 (6 per wheel) or just the one that has failed? Is it a fresh fruit or a fresh round of Russian roulette?

Normally these cost barely anything, but I really need them ASAP, so I'm paying 20 times the price. All Finnish distributors out of stock, expecting them in 6 weeks or more, so I'm shopping Ebay, of all places.

[esaj]

12 hours ago, mrelwood said:

Thanks @esaj. I was finally able to disassemble the vehicle enough to detach the control board for the resisting wheel, and the first MosFet (P75NF75) indeed didn't measure properly!

Now, should I change all 12 (6 per wheel) or just the one that has failed? Is it a fresh fruit or a fresh round of Russian roulette?

I'd suggest changing all on the failed wheel, there may be others in addition to that one. Hopefully all the other parts (gate drivers, current measuring resistors or chips and such) have survived without damage.

 

Quote

Normally these cost barely anything, but I really need them ASAP, so I'm paying 20 times the price. All Finnish distributors out of stock, expecting them in 6 weeks or more, so I'm shopping Ebay, of all places.

I think you could probably use some other model that has same or higher maximum current and voltage, as long as the gate charge isn't ridiculously much higher. It's still a gamble though, so you might want to ask for other opinions too.. If changing the model, then I might suggest changing all 12, so both boards behave similarly.

TME is located in Poland, the shipping is something like around 10-12€ (UPS Express to your door, usually arrives in 1-2 business days over air, unless you order chemicals or batteries requiring ground transportation) up to 10kg or something, and they're much cheaper than Digikey or Mouser, but have far less selection:   

N-channel THT-mosfets in TO-220 -packages, Vds >= 75V, Ids >= 75A, Vgs >= 20V, Rds <= 12mOhm :

https://www.tme.eu/fi/katalog/tht-n-kanavaiset-transistorit_112827/#id_category=112827&amp;s_field=niski_prog&amp;s_order=ASC&amp;visible_params=2%2C10%2C35%2C35%2C36%2C262%2C262%2C265%2C265%2C266%2C266%2C270%2C270%2C273%2C635%2C909%2C2362%2C2611%2C2633&amp;used_params=35%3A381960%2C150%2C371917%2C734%2C804%3B262%3A24543%2C24909%2C24545%2C24624%2C24626%2C24707%3B265%3A379757%2C381904%2C381935%3B266%3A24543%2C25986%2C26989%2C26227%2C24909%2C24622%2C25983%2C26375%2C26228%2C25961%2C26260%2C26011%2C26013%2C29792%2C24545%2C28685%2C26589%2C24722%2C24624%2C26231%2C24625%2C26568%2C26353%2C24626%2C24578%2C26618%2C24630%3B270%3A24569%2C26832%2C26539%2C27160%2C26538%2C24557%2C26820%2C25368%2C27510%2C24843%2C25771%2C27560%2C28483%2C24844%2C27459%2C27484%2C26446%2C26831%2C26828%2C24855%2C27404%2C24572%2C25873%2C26453%2C25373%2C24963%2C25626%2C25876%2C24972%2C27107%2C26973%2C24887%2C26822%2C24889%2C27576%2C24859%2C26444%2C26436%2C27113%2C27176%2C26821%2C25090%3B

Check the datasheets of STP75NF75 and compare the gate charges, unless it's multiple times larger, likely the gate drivers can drive the other mosfets just fine... but I can't guarantee it    Also, check the pin out, it's typically GDS (Gate-Drain-Source) when looking from the front of the package, and I don't remember ever seeing any other configuration, but just to be sure.

The eBay/Aliexpress/whatever -ecommerce/auction sites are a bit of a gamble, as you can never be 100% certain that you're getting what it says in the case. There are reports of counterfeits and "gray market" factory rejects (components that failed to meet the datasheet requirements and were supposed to be destroyed but instead end up in "hobbyist" markets). 

[mrelwood]

I don’t have enough understanding to determine which parameters are crucial and what is close enough (or what they even mean), so unless somebody knowledgeable can guarantee a substitute to work, I can’t choose one.

That is sadly a very good point on the Ebay parts being rejected ones. Damn. What’s a man to do?!

[esaj]

50 minutes ago, mrelwood said:

I don’t have enough understanding to determine which parameters are crucial and what is close enough (or what they even mean), so unless somebody knowledgeable can guarantee a substitute to work, I can’t choose one.

I can understand that, personally, knowing (or at least thinking I know) what I know now about electronics, I would take the gamble to use parts that are "near enough" in values, but it would still be a bit nerve-wrecking to ride at start (and could end badly ). And I'm just a hobbyist, so my opinion shouldn't be trusted too much.

Maybe someone like @DaveThomasPilot or  @Christoph Zens could offer their opinion (If I remember correctly, they're both electronics engineers?). Christoph did once write: 

 

 

Quote

That is sadly a very good point on the Ebay parts being rejected ones. Damn. What’s a man to do?!

If you want to take your chances, I can mail you the STP75NF75's (if you didn't order them already), according to my inventory, I still should have ~30 pieces left. All from Aliexpress, so no idea if they're the genuine thing, factory rejects or counterfeits. Never failed on me, but then again, I don't really run any high currents through them.

All the ones I have from reliable sources are either for too low voltage or use wrong case.

[DaveThomasPilot]

I think esaj did a good job of explaining.  Maximum Vds voltage and RDson are the two most important ones.  

I'm not sure I'd agree that gate charge is a non-issue.  It's not the power that's associated with charging and discharging the FET eache cycle, but rather how fast the transitions occur.

Ideally, the FETS don't have non-zero current, and high voltage simultaneously.  While they have both high voltage and high current simultaneously, power dissipation is tremendous.  That happens twice every cycle, when they turn on and off.  So, the transition times must be short relative to the cycle time.

When a FET is on, current will be relatively high, but the Vds will be near zero.  How close to zero will depend on the FET's RDSon (Vds = I * Rdson).  Bigger FET chips have lower RDSon.  But, all things else being equally, more die area means higher Cgs, and more gate charge required to switch the FET.  If the gate driver isn't scaled for the higher gate charge, the transition time will scale (roughly) proportional to the gate charge.

So, it's very important that the switch time be very short relative to the switching frequency.  Are the FET drivers in EUC control boards over-designed so the change in switching times is negligible?

Parasitic gate wiring inductance can resonate with the gate capacitance and cause voltage spikes to occur on the FET which can quickly destroy them.  Has anyone looked at the FET Vgs while it's switching?  I've wondered if that's not the root cause of many of the FET failures reported.

If switching times are so fast as to not contribute significantly to power dissipation, why isn't the switching frequency higher so as to avoid any annoying audio?

Something is causing the FETs to die in the first place.  Until the root cause of the failure is know, it's not clear what would constitute a "stronger" FET.

 

 

 

 

 

 

 

 

[mrelwood]

2 minutes ago, DaveThomasPilot said:

Until the root cause of the failure is know, it's not clear what would constitute a "stronger" FET.

Thank you very much @DaveThomasPilot! Seems it is best I order the exact same MosFets, hoping that I get units that aren't disqualified for parameters that would affect the function or durability in this usage.

[esaj]

20 minutes ago, DaveThomasPilot said:

I think esaj did a good job of explaining.  Maximum Vds voltage and RDson are the two most important ones.  

I'm not sure I'd agree that gate charge is a non-issue.  It's not the power that's associated with charging and discharging the FET eache cycle, but rather how fast the transitions occur.

Ideally, the FETS don't have non-zero current, and high voltage simultaneously.  While they have both high voltage and high current simultaneously, power dissipation is tremendous.  That happens twice every cycle, when they turn on and off.  So, the transition times must be short relative to the cycle time.

When a FET is on, current will be relatively high, but the Vds will be near zero.  How close to zero will depend on the FET's RDSon (Vds = I * Rdson).  Bigger FET chips have lower RDSon.  But, all things else being equally, more die area means higher Cgs, and more gate charge required to switch the FET.  If the gate driver isn't scaled for the higher gate charge, the transition time will scale (roughly) proportional to the gate charge.

So, it's very important that the switch time be very short relative to the switching frequency.  Are the FET drivers in EUC control boards over-designed so the change in switching times is negligible?

Parasitic gate wiring inductance can resonate with the gate capacitance and cause voltage spikes to occur on the FET which can quickly destroy them.  Has anyone looked at the FET Vgs while it's switching?  I've wondered if that's not the root cause of many of the FET failures reported.

If switching times are so fast as to not contribute significantly to power dissipation, why isn't the switching frequency higher so as to avoid any annoying audio?

Something is causing the FETs to die in the first place.  Until the root cause of the failure is know, it's not clear what would constitute a "stronger" FET.

I've been wondering about the gate circuitry often (by myself and here in the forums). At least older Ninebot One's used "bootstrapping circuit" (I think that's what it's called?) with transistors where there was a capacitor that was "pumped" to a higher voltage between the high-side gate and the motor phase, many others use actual gate driver-chips. The only design I've seen the "proper" ferrite-beads for killing resonation + gate-resistor and reversed diode for fast turn-off was Firewheel. The new Ninebot Z's seem to have reversed diode near the gate, so I assume that's for fast turnoff, but the picture I saw didn't show the gate drivers (likely behind the boards, they use SMD-mosfets now). The dead ACM-board Rehab1 sent me had the paralleled mosfet gates tied together (no resistors in-between), a design Chris Hettenhausen of 1Radwerkstatt called "electronic trash".

I based my "multiple times larger" -gate charge limitation to what Christoph said, since usually even the high voltage mosfets seem to have gate charges somewhere around 150nC or so... Probably I should just shut up and leave it to the professionals?  

 

[DaveThomasPilot]

I probably missed the discussion about 150 nc.  If you could approximate the gate drive current (or series impedance), it's easy to calculate switching time.  Then, see what percentage of the cycle it's switching.

My experience was with high speed (> 1 Mhz), non-resonant, DC/DC convertors where the transition speed was everything.  I have no experience with the control board FET drives, so I can't say that it's an important spec.

But,  if the selection of the switching frequency was not limited by gate slew rate (or cost/size of gate driver), why isn't it higher?

Radiated EMC is a possible answer.  But, when I'm riding my 16s, FM radio reception is obliterated.  Even strong, local stations are replaced with a whine from the wheel.  So, also speculate they worry little about FCC compliance.

 

 

[mrelwood]

By the way, I think I know why this MosFet died. Looking at the back the MosFet surface that attaches to the cooler is very rough, like a bad solder job. First I thought that the finish had boiled due to it crapping out, but it is screwed tight to the cooler. So this must’ve been how it came from the factory. No wonder it died, it didn’t get proper cooling.

[esaj]

11 minutes ago, DaveThomasPilot said:

I probably missed the discussion about 150 nc.  If you could approximate the gate drive current (or series impedance), it's easy to calculate switching time.  Then, see what percentage of the cycle it's switching.

It wasn't mentioned here, just my general take reading the datasheets of the mosfets commonly used in the wheels. The IRFP4368's used in KS16S actually have gate charges closer to 400nC (370nC typical, 570nC max), so they likely use a more beefy gate driver (never checked the gate drivers on the board on mine though).

 

Quote

My experience was with high speed (> 1 Mhz), non-resonant, DC/DC convertors where the transition speed was everything.  I have no experience with the control board FET drives, so I can't say that it's an important spec.

But,  if the selection of the switching frequency was not limited by gate slew rate (or cost/size of gate driver), why isn't it higher?

I've thought it's due to minimizing switching losses (less switching over time, ie less transitions with switching losses), but could be wrong. 

 

Quote

Radiated EMC is a possible answer.  But, when I'm riding my 16s, FM radio reception is obliterated.  Even strong, local stations are replaced with a whine from the wheel.  So, also speculate they worry little about FCC compliance.

The RF-stuff just goes way over my head, but I doubt they do much testing on the radiation emitting from the wheels.

 

10 minutes ago, mrelwood said:

By the way, I think I know why this MosFet died. Looking at the back the MosFet surface that attaches to the cooler is very rough, like a bad solder job. First I thought that the finish had boiled due to it crapping out, but it is screwed tight to the cooler. So this must’ve been how it came from the factory. No wonder it died, it didn’t get proper cooling.

That's odd, if all the mosfets are directly connected to the same heatsink without electrically insulating pads/mat/whatever the stuff is called (kinda like tape/soft um... cloth?), I'd expect them to short circuit the high- and low-side drains together, basically shorting over the high-side mosfets. On the other hand, if that's the failed mosfet, I guess a sudden a huge current surge with the mosfet shorting could actually melt the metal (I've seen picture of a mosfet from EUC Extreme's board that had actually melted the legs off and exploded).

[mrelwood]

And in case you are interested, here’s the control board itself. I decided on ordering just a few replacement MosFets from Ireland, so I should have them already next week.

[mrelwood]

2 minutes ago, esaj said:

That's odd, if all the mosfets are directly connected to the same heatsink without electrically insulating pads/mat/whatever the stuff is called (kinda like tape/soft um... cloth?), I'd expect them to short circuit the high- and low-side drains together, basically shorting over the high-side mosfets.

There is an thin, flexible opaque orange plastic film between the MosFets and the cooling bar, which then has a silicone bar between it and the metal vehicle frame. The MosFet screws are individually insulated with hard plastic bushings as well, and there is thermal paste between the orange film and the cooling bar, so the MosFets themselves are free of paste.

 

[mrelwood]

Photos vs words...

[esaj]

27 minutes ago, mrelwood said:

And in case you are interested, here’s the control board itself. I decided on ordering just a few replacement MosFets from Ireland, so I should have them already next week.

Based on my (uneducated) best guess, the transistor/resistor/capacitor/diode-combos near every other mosfet could be the high-side charge pumps, don't know if the "SD06"-marked chips farther to the left on the first picture are gate drivers, but I doubt it (only two and further away... current sensors?). I'm way out of my depth here, but I guess charging the capacitor to get high enough voltage to the gates does slow down charging the gate, so using mosfets with higher (like "multiples" ) gate charge could be detrimental here. Can anyone say if there's any benefit to use a charge pump made from discrete components vs. a gate driver IC or if it's just cost cutting? If it's cost cutting, it's still weird, it's not like the gate drivers push up the price that much.

12 minutes ago, mrelwood said:

Photos vs words...

Yeah, that's the stuff I'm talking about... the middle leg (drain) of each mosfet is directly connected to the metallic back plate of the part, and since the high-side mosfet drains are connected to the battery voltage and the low-side drains are connected to separate phases of the motor, they need to be electrically insulated from each other (well, you could leave out the insulation on the high-side, but then the heat sink will sit at the battery voltage, which may not be a good thing, depending where else it touches and whether someone pokes their fingers there).

[Keith]

1 hour ago, DaveThomasPilot said:

Radiated EMC is a possible answer.  But, when I'm riding my 16s, FM radio reception is obliterated.  Even strong, local stations are replaced with a whine from the wheel.  So, also speculate they worry little about FCC compliance.

Oh God! Don’t let the U.K. police know that or they’ll be using Television detector vans to find us next!

[mrelwood]

Update: I replaced all Mosfets on both control boards, and put it back together just enough so I can try if it works. And seems it does! It balances, steers and all in all behaves normally.

While screwing the Mosfets to the cooling bar, I noticed that the Mosfet at the slot where the burnt Mosfet used to be was slightly loose even after I turned the screw tight. Turns out the thread was slightly too short for that one Mosfet! One more possible explanation for it to burn. I extended the M3 threads a bit to get it well secured.

I just received the Msuper X 1600Wh, so my enthustiasticism is quite torn between the two... I think I'll put the A3 together tomorrow.

...-ish...

[jaytee]

On 7/10/2018 at 3:00 PM, mrelwood said:

Update: I replaced all Mosfets on both control boards, and put it back together just enough so I can try if it works. And seems it does! It balances, steers and all in all behaves normally.

Very cool story. I'm also disabled but young so not a fan of any of the senior citizen options. I'm looking into the Airwheel A3 and the Ecorider Mark3 (from singapore). I am worried that both these seated options are not suited for tallish guys (6'3') and will be uncomfortable

 

[mrelwood]

2 hours ago, jaytee said:

I am worried that both these seated options are not suited for tallish guys (6'3') and will be uncomfortable

I did make an additional saddle for the A3 to get more height and comfort. Will not ride without it, the original saddle is quite narrow. I’m 193cm tall (6’ 5”?).

It’s hard to compare since the A3 was out of order for several weeks, but it really does seem like it has more torque. I blamed the old battery for the notable lack of power, but it seems the lack is not that notable after all. Could be my mind doing tricks though.

[mrelwood]

A late update to my A3 saga.

I did indeed get the burnt motor drive board fixed by replacing the Mosfets. On the first proper test ride I rode up a short but steep path in order to see if there was indeed more torque. The A3 stopped balancing and only sang me the familiar one note funeral song. Reboot, no change. I think the error code was 6 this time.

Another Mosfet had blown. No biggie, I thought. Replaced the Mosfet, only to find out that the issue didn’t go away this time. Took me several long sessions to find out that a tiny SMC transistor labeled ”Y2” next to the burnt Mosfet was burnt as well. Ordered a few online, and was able to get it running, sort of.

The other motor was running jerky. I was able to ride it carefully, but the motor felt like it skipped a third of the coils. Didn’t feel safe at all, so I continued trying to fix the thing. Replaced caps, transistors and Mosfets here and there, and measured every single component on the board, although I didn’t know how to read the results for most of them.

Finally gave up. Airwheel didn’t reply to my email, but I found the board for around 350€ shipped from Canada. Waited for over a month before I finally got an email from Airwheel Europe quoting 22€ for the board, 25€ for shipping! I ordered three right away! Which is when they stopped replying, again.

At that point I had found a bargain for a Segway/Ninebot MiniPro, which ended up serving me pretty well through the winter, so I had just about forgotten about the A3.

Then I finally got a reply from Airwheel apologizing for ”email problems”. We finalized the purchase, and a week later I had replaced both drive boards and the A3 was running fine!

...until I went out to visit the Post office. The A3 battery died after just a few km, and I again had to call a cab.

I rebuilt the 16S4P battery block from the two Lhotz batteries I had waiting for a future project. That was one bell of a job, but since I have previous experience with building these batteries, all went fine and the punch from a healthy battery pack is lovely! Now I can finally say that the A3 is indeed finally in a full working condition. The range is even slightly better than new, as the original cells totalled only at 520Wh. 2x Lhotz = 680Wh, although the cells are well used as well.

Lesson learned? 3000km is past the lifetime of the Airwheel A3. I did ride it a LOT for two years, so under regular use I’d set the expected lifetime at 5 years.