Weber and Coulomb's MX-5

Post up a thread for your EV. Progress pics, description and assorted alliteration
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

4Springs wrote:If I were going to buy or build a unit, I'd like to install it where I can see/hear a warning from under the bonnet. When working on HAZV I'm constantly looking up at my isolation switch, and I imagine I'd like to have the same reassurance from a device like this.

One thing to be aware of is that if you're touching a battery terminal when the device does its test ("tick tick tick tick" as both relays click on and off over a couple of seconds) then you will get zapped. A far from fatal zap, but also far from pleasurable. Then again, maybe this is a feature. It'll learn ya not to touch them battery terminals!

I was only going to make it do its thing when I turn the key to ON. When the key is OFF, each 360 V half-pack is broken up into 4 or more segments of 90 volts or less, by EV200 vacuum contactors. The insulation monitor can't do its thing when the pack is broken up like that. You'd need a separate insulation monitor for each segment. But it's much safer to work on when broken up like that anyway.

But it does go against the grain for the IMU not to have its own piezo beeper so it can still give a warning even when it can't communicate via the optic fibre. After all, every one of our 218 cells has its own piezo beeper on its BMU (CMU).

I was thinking that one way to get the piezo back on pin 1 would be to combine the two touch current measurements into one analog input. Both opto transistors could feed the same pulldown. For when they are turned on separately they can be calibrated separately, for their different CTRs (current transfer ratios of the optos). But when they are both on, for self-test mode, they will not sum correctly unless the CTRs happen to be the same. I guess this doesn't matter. The main thing in self test mode is that there should be some current so we know it's not broken.
Last edited by weber on Tue, 17 Dec 2013, 17:50, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
4Springs
Site Admin
Posts: 926
Joined: Thu, 23 Dec 2010, 01:14
Real Name: Christopher Walkden
Location: Selbourne, TAS

Weber and Coulomb's MX-5

Post by 4Springs »

weber wrote:One thing to be aware of is that if you're touching a battery terminal when the device does its test ("tick tick tick tick" as both relays click on and off over a couple of seconds) then you will get zapped. A far from fatal zap, but also far from pleasurable.
You only get zapped if you are leaning on the chassis, but that is fairly likely, so point taken.
weber wrote: But it does go against the grain for the IMU not to have its own piezo beeper so it can still give a warning even when it can't communicate via the optic fibre. After all, every one of our 218 cells has its own piezo beeper on its BMU (CMU).
You could use a relay as a buzzer, but then as you said, there is a possibility of a kick from a battery terminal. (actually if there is a fault then I suppose there is anyway!)
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

Wow, IMU changes galore.

Here are some more to-and-fro thoughts, sorry. ..

Could the 2.5 V regulator leak back to the input side ?
With a diode feeding the 1F cap on the as an input to Regulator at 5V might be better?

Would using a resistive divide of 2 and 5V rail on the opto
Allow better accuracy or repeatability for Temp variation?
VOL628A data sheet looks problematic with If -LED near 20mA with CTR drop off. If your setpoint is 10mA for setting fault state this would appear to be more in the Normalized range. Or even better 1mA.
Main thing is test max current is 20mA.


Would say a interlock to the relay vdd supply stop software glitch from activating "touch" test?
Say using a third opto with input from ELV side.
I'm just HV shy!

The ELV side could sense test currents with 100R between opto out and chassis. Why? Just an option.

How will the pack A and B IMU boards interact with each other during Touch tests?
Can a pack be enabled with no loads then tested?

Will the charger and ACC 12V DC-DC be connected to battery sets for Touch tests?

And/Or motor controller?

Will the IMU's need to change when you go to Stage 2 with 900V Tritrium controller?

Also is the input impedance to ADC for Bat voltage okay?
And why have scaling for max at 1.5 V?

I had trouble with your contents list in first post to find the full battery diagram, electrical diagram section for battery links go to CAD drawing and location grouping list.

On MX5 cars I came a cross the Bullet-SS with supercharged V8 built in QLD for $100k.
Woonder what they'd think if you asked to use their dyno?

Great to read that your out enjoying your hard toil on the road!

7c
Bullet SS
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

7circle wrote:Could the 2.5 V regulator leak back to the input side ?
With a diode feeding the 1F cap on the as an input to Regulator at 5V might be better?
Yes the 1F cap would discharge back through the reg if the reg's input was shorted, however I'm pretty sure the 12V to 5V isolated DC-DC will have a rectifier in its output stage that will not allow reverse current, even if its 12V supply goes away. The 1F cap only has a 2.7 V rating so it cannot go on the 5V side. You would need two in series with balancing resistors.
Would using a resistive divide of 2 and 5V rail on the opto
Allow better accuracy or repeatability for Temp variation?
Not sure what you're saying here. The micro has an analog reference voltage of 1.5 V so that's full scale. The micro has a built in temperature sensor so we can compensate for the effect of temp on CTR with a straight line approximation since we don't expect to experience temps below -20^C. We can also compensate for the effect of current on CTR. But even if we didn't compensate, if we allow for the worst case and that means we sometimes raise the alarm when it's only 10 mA touch current, not 20 mA, that really doesn't matter.
VOL628A data sheet looks problematic with If -LED near 20mA with CTR drop off. If your setpoint is 10mA for setting fault state this would appear to be more in the Normalized range. Or even better 1mA.
Main thing is test max current is 20mA.
Yes. Good points about changes of CTR with temp and current. We will make sure we err in the safe direction.
Would say a interlock to the relay vdd supply stop software glitch from activating "touch" test?
Say using a third opto with input from ELV side.
I'm just HV shy!
Hmm. That sounds a little extreme. The touch test is not itself dangerous. Can just disconnect the 12V supply to the iso DC-DC module and disable the micro too.
The ELV side could sense test currents with 100R between opto out and chassis. Why? Just an option.
You mean for say calibrating the tester with a multimeter. Not a bad idea. I will add that. Thanks.
How will the pack A and B IMU boards interact with each other during Touch tests?
Not at all. But they will both be controlled and read over their optic fibres by other micros that will be in communication with each other over CAN bus. One Tritium Driver Controls unit per half-pack.
Can a pack be enabled with no loads then tested?
Yes.
Will the charger and ACC 12V DC-DC be connected to battery sets for Touch tests?
The HazV to 12V DC-DCs would normally be connected and the chargers normally would not. But we can arrange any combination. We have battery contactors and we have a separate contactor for each HazV load or charge source.
And/Or motor controller?
Can choose to do test with or without that too.
Will the IMU's need to change when you go to Stage 2 with 900V Tritrium controller?
No. But we must keep in mind that in the 720 V series case, if we tested with the motor controller connected, then there could be up to 720 V relative to chassis in a fault situation, not merely 360 V. We are designing for 2 kV isolation, but yeah maybe we should make the 5W probe resistors twice the resistance shown on the last schematic.
Also is the input impedance to ADC for Bat voltage okay?
Yeah. No problem.
And why have scaling for max at 1.5 V?
That's the micro's maximum analog reference voltage when given a 2.5 V supply.
Last edited by weber on Sat, 21 Dec 2013, 20:10, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

The IMU PCB artwork has been sent to OSH Park for fabrication. 3 for US$27.30 inc postage. These guys have taken over providing the BatchPCB service that used to be run by SparkFun.

Here's the final schematic. Thank you CometBoy, Renard and 4Springs, and a really big thankyou to 7Circle for many valuable ideas we have incorporated.

The main changes since rev 3 are:
1. We have increased the touch-test resistors back to 100k, but now with a 750 V rating (7W), so we can, if we choose, run the test with the two half-packs connected in series, and thereby detect insulation failure in the motor controller, and possibly even the motor. 7Circle and Coulomb.
2. We have reinstated the piezo beeper, by using only one analog input to read touch-test current from either HazV+ or HazV- (summing them in self-test mode). 4Springs.
3. Rather than finding somewhere to put another yellow LED, we used the otherwise unused red LED of the dual red/blue to indicate when the HazV+ relay was on. The existing yellow LED indicates when the HazV- relay is on. (Parts used in our Cell Monitoring Units (CMUs/BMUs) are used in the IMUs where possible, to reduce inventory.) Weber.
4. We have added R26 (100R) to allow measuring the touch-test current with a multimeter for testing and calibration. 7Circle.
5. We have added a 6th resistor (R3f) to the ladder on the high side of the battery voltage divider, so we can maintain 6 mm creepage everywhere. 0805 resistors are typically rated at 150 V limiting element voltage (max working voltage) and these will see no more than 67 V each continuously (and 30 mW). However they only have 1 mm creepage, hence the need for 6 of them. Coulomb.

Image

Here's OSH Park's rendering of what the two sides of the board should look like. Yes, they will actually be purple. All thru-hole parts will go on the top side (shown on the left below). Only low-profile surface-mount parts go on the bottom side (shown on the right below). You can see the 6 mm creepages across the optos, reed-relays and isolated DC-DC and from the test resistors to everything else (except that the microprocessor related circuitry is all within 5 volts of HazV- and so doesn't need creepage from that resistor end).

Image

[Edit: Changed images from .png to .gif.]
Last edited by weber on Mon, 13 Jan 2014, 09:42, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

Some time ago, offgridQLD made what I think is an excellent suggestion: "Perhaps voice alarm based on whats triggering the stress. That would keep the stock dash looking clean."

I found this low cost text-to-speech device from SparkFun (US$60).
https://www.sparkfun.com/products/11711

Image

It uses the Epson speech chip, which implements the DECtalk algorithm developed in the 1980s by the now-defunct Digital Equipment Corporation. Its default voice, called "Perfect Paul", is used by a certain famous wheelchair-bound physicist. So we could have Stephen Hawking telling us about potential problems with the MX-5 as we're driving along. For example
Voice Alarm -- Perfect Paul

Of the female voices, I like "Rough Rita". She sounds like she's smoked a few too many Pall Malls in her long and weary life.
Voice Alarm -- Rough Rita

[Edit: I made these audio examples by typing the relevant phrases into the Windows software version of DECtalk that you can download here http://www.theflameofhope.co/dectalk/dectalk440.zip]

[Edit2: Some people have way too much time on their hands. DECtalk sings Bohemian Rhapsody. From this thread.]
Last edited by weber on Tue, 07 Jan 2014, 05:31, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

(edit: png replaced by gif clean up... Thanks weber)

(edit: was android 2.3.6 on my phone.)
Perhaps your browser has a cached higher res image but my view is to bitsie.... can you replace or upload a clearer image?

If you add voice alarm messaging you'll need to put scanning leds on the front bumper..... Might make hassle of f fitting them worth while if they show SOC....
Sorry getting silly.

A friend from uni who collected Albert quotes got to meet Stephen Hawking. Was so suprised when she posted a selfie with him. (from a birth in roughing it in philipines to elec eng in UK via Aus... makes me smile )
Last edited by 7circle on Mon, 13 Jan 2014, 10:25, edited 1 time in total.
User avatar
coulomb
Site Admin
Posts: 6357
Joined: Thu, 22 Jan 2009, 20:32
Real Name: Mike Van Emmerik
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by coulomb »

7circle wrote: Your teasing me with pixelation! Perhaps your browser has a cached higher res image but my view is to bitsie.... can you replace or upload a clearer image?

The short answer is no, we can't upload a clearer image, but we don't need to, and it wouldn't look any better even if we did. The image you quoted is over 1300 x nearly 1800 pixels, and is very clear. But the forum software displays it on the page as scaled to 750 pixels maximum. This is actually a good thing, as really wide images cause problems with scrollbars... trust me, this is not a bad compromise.

To see the full image, my suggestions involve something that is browser specific. [ Edit: Weber suggests a browser independent work around in the next post. ] For most browsers, you need to use the context menu (usually the right mouse button, but some computers may have left and right buttons swapped, it may be different if you are using a digitiser instead of a mouse, etc.)

For Firefox, the context menu has an option View Image (first option), and this will display the image by itself as a new page (so use the Back button to get back to the page you were viewing afterwards). NOTE: this will by default show the page scaled so that all of both the height and width are not larger than your screen, less overhead for menus and so on; F11 (full screen) may help here. So show even higher resolution (so you need a scroll bar or bars), click on the image to zoom in. Another click will zoom scaled to fit the page again. This is using Firefox 26.0 with Windows XP.

In Google Chrome, you can use the context menu item "Open image in new tab", and use the same click to zoom method and even F11 for full screen that Firefox has. Well copied, Google!

Until I tried it, I was sure that Internet Explorer had similar options. But IE 8 seems to think of the image as being the scaled size, and doesn't seem to give you any useful options for viewing the image full size in-place like Firefox does. (Show Picture would seem to be the appropriate option, but it is greyed.) Hopefully this is fixed in IE 9 and later. [ Edit: not fixed in IE11; thanks for the update, Weber. ] For IE 8, the only way to see the image full size seems to be to use Save As, save it as a BitMap format (BMP file) (this will be some 9 MB for this image!), find it in My Pictures or Documents/Pictures (you can use the Go To My Pictures context menu, at least on the Windows XP version), and double click on that to view it. Smooth work, Microsoft!

Edit: there are also Weber's zoom techniques as given in the following post, but these (for me at least) zoom the other text as well as the image of interest. The main problem is that if you need the horizontal scrollbar (unlikely for this tall image, but possible for others), it is at the bottom of the post, not the bottom of the screen, so you need the vertical scrollbar to get to the horizontal scrollbar. Despite this, it's probably the best technique if you are using Internet Explorer.

[ Edit: I tried not to sound so snooty after the "short answer is no" quip. ]
Last edited by coulomb on Mon, 13 Jan 2014, 10:05, edited 1 time in total.
MG ZS EV 2021 April 2021. Nissan Leaf 2012 with new battery May 2019.
5650 W solar, 2xPIP-4048MS inverters, 16 kWh battery.
Patching PIP-4048/5048 inverter-chargers.
If you appreciate my work, you can buy me a coffee.
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

Above the previous version of that diagram I wrote:
"Zoom to see more detail. e.g. In Firefox, Ctrl and plus-sign or two-finger spread on trackpad or touchscreen."
Clearly I need to add this note every time an image may need zooming to be legible. But I can drop the "In Firefox" in future.

It turns out that the above zoom methods work in all the common browsers and operating systems. But perhaps that note is a bit too terse. I'll spell them out in detail, and add a third method:

To zoom in, hold down the keyboard key marked "Ctrl" (on a Mac use the key with the apple and clover-leaf instead) while bouncing another finger down and up 5 or 6 times on the key with the plus sign (and equals sign) on it.

To reverse the effect of the above, use the key with the minus sign (and underscore) instead. Using the key with the digit zero (and the right parenthesis) will return you to the default zoom level.

If you have a trackpad or a touch screen, to zoom in you can place your thumb and finger together on the device and spread them apart. Do the reverse to zoom out.

If you have a mouse with a scroll wheel, you can hold down the keyboard key marked "Ctrl" (on a Mac use the key with the apple and clover-leaf instead) and turn the wheel.

[Edit: Attempted to reduce snootiness level. Image]
Last edited by weber on Mon, 13 Jan 2014, 09:34, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

Thanks for the replies W & C

Short answer my end faulty.

Using a 4" android 2.3.6 and a search reveals they have png image issues. Doue! Just replace my scratced up and dodgey mini-usb 3.5" Android 2.3.4 $50 phone with a supermarket $70 4" Android 2.3.6 phone. It is a bit out of date but i'll hunt for a fix.

So I used an online image converter to jpg highest quality at 680kB and to gif 82kB and both look much clearer than this browsers stupid pixellie view of the png.

for anyone else on old 2.3.6 here is the converted gif

(edit img to url) uploads/1268/converted_fe4c167c.gif

The ver1 image you posted was gif.

Well I'm the wiser now.

Now back to the circuit.
Can R26 handle 72V to make 750V to 822V?
I guess thats silly if it is 100ohm.

Or R1 and R2 rated for more than 750V?

On silly style issue if you happen to make a new schemaric version, would it be clearer if OP3 was moved left to line up with OP2 and the relay dashed line extended up to the contacts. Can't help myself.. Image .. join up TouchV nets.

On the firmware side look forward to read how you implement all the features.

Have you got hour hands on the goods.
Great price from OSHpark.

(edit: don't show gif)
Last edited by 7circle on Mon, 13 Jan 2014, 10:30, edited 1 time in total.
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

(edit: The old double post)

so replace it with note that the png image of the CAD rendered PCB looks fine as a png.

I didn't look close at the layout because the track work on top and bottom is lost. (edit: "lost" as in can't see it clearly under the solder mask. PCB routers show layers as combination colours)
Image
Some times I wonder how far new technology gets us.

Any way I can clearly see R26 in bottom right.
Looks like 50V Surface mount.

These CAD renders do help pick through hole issues.

And your replies are far from snooty.
wish I could screen capture the pixeled view of the schematic for ya's

Cheers 7C
(edit: I'll try and let the dust settle now after the edits)
Last edited by 7circle on Mon, 13 Jan 2014, 10:37, edited 1 time in total.
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

7circle wrote:Can R26 handle 72V to make 750V to 822V?
I guess thats silly if it is 100ohm.

Or R1 and R2 rated for more than 750V?
Good point. I assumed that with 750 V working, they would have a higher withstand voltage (1 minute). But in fact they do not.
http://www.farnell.com/datasheets/1661907.pdf
So I have changed that max fault voltage on the schematic to 750 V. 832 V is the summed maximum output voltages of our two chargers. But the two half-packs are never actually in series when being charged. 750 V is 3.44 V per cell.
On silly style issue if you happen to make a new schemaric version, would it be clearer if OP3 was moved left to line up with OP2 and the relay dashed line extended up to the contacts. Can't help myself.. Image .. join up TouchV nets.
Good ideas. Done. Called rev 4.1. Won't bother posting until there are more significant changes, e.g. after testing.
On the firmware side look forward to read how you implement all the features.
So do I. Image
Have you got hour hands on the goods.
Great price from OSHpark.
They are done and on their way across the Pacific, but not quickly, at that price.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

7circle wrote:I didn't look close at the layout because the track work on top and bottom is lost. (edit: "lost" as in can't see it clearly under the solder mask. PCB routers show layers as combination colours)
Image
Some times I wonder how far new technology gets us.
Is this what you want.

Image
Any way I can clearly see R26 in bottom right.
Looks like 50V Surface mount.

Size is 0805 (imperial) so 150 V working, 300 V withstand. But it should never see more than a volt, since it is part of a 1000:1 voltage divider.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

Hi Weber

Ta for router view of PCB.
Looks nice and tidy.
Few things I think worth considering.
1: R17 does the I-to-V conversion, but it is a long way from the ADC input. And track work with gnd is a big loop. And the pcb is mounted on the Bat Shunt so Magnetic Fields will be big.
You mention TOUCH tests will be with no battery current, but the EMI could just be a problem on the uCro pins.
This may also be a similar issue for the FETs driving the relay coils.

2: If there is bad solder join or part failure on opto's or R26 there would be no load to hold relay ELV side low. Track work is very close to 12V signals. Single point failure may present too high probabilty of HV effecting 12V supply lines.
The loading of the DC-DC input is a clamp of types, but the fault source could be negative or positive HV. Is this a problem?
Sometimes just having zener diodes to the relay NO pins and chassis will reduce probabilty to double-point failure and perhaps be acceptable.

3: Looking into reed relays coil needs, I noticed the spec derates the 1000V open contacts to 1mA. So can it handle 8mA openned to 350V?
They also do the life operations test at a very soft 1V To 10mA contact load.
Further the insulation test is done at 100V.

I hope that feed back helps.
And thanks for using gif pictures.

7C

User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

Thanks 7Circle. Excellent points as usual. I have changed the track layout to address your first two points, as shown below with some 6 mm creepage circles. Higher resolution than last time (Use Ctrl-plus to zoom in).

Image

Re your 3rd point:

For others following this discussion, here's the datasheet for our reed relays:
http://www.farnell.com/datasheets/164037.pdf

I believe they can handle 7.2 mA opened to 720 V (100 kR test resistor). That's only 5.2 W and the contacts are rated for 10 W. But as you point out, at 1000 V you can only switch 1 mA which is only 1 W. So how do we know how to derate it from 10 W down to 1 W as we approach 1000 V?

This document shows the typical shape of the safe operating area (although from a different manufacturer).
http://www.comus-intl.com/PS%20339x%20HVR.pdf

It implies that our relay's 1 mA at 1000 V is really a bonus. Usually the current has to be zero at the maximum switching voltage, just as the voltage has to be zero at the maximum switching current (500 mA). I interpret that safe operating curve for our relay as

V^2/Rv + V*I + I^2*Ri <= 10 W

where Rv = 111k and Ri = 40R.

This smooth curve agrees with 1000 V, 1 mA at one end, 0 V, 500 mA at the other end, and 10 W in the middle (e.g. 140 V, 70 mA) so I'm reasonably confident of its prediction that 720 V, 7.2 mA will be OK.

That document also suggests we can expect 100,000 operations at rated voltages and currents. I'd be happy with 30,000 operations which would be 4 times a day for 20 years. I think the 300 million operations at 1V and 10mA is essentially the mechanical limit.

And yes, the insulation resistance test is done at only 100V, but we can live with a lot less than 10^11 ohms and it's clearly rated for 1000 V continuous operation, 2000 V dielectric strength.

Thanks to one of your earlier suggestions, we can detect contacts that are failed open or failed closed, and neither situation is dangerous in itself.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

Here's a log-log plot of the safe operating area I'm assuming for the Coto Technology 9104-05-11 Reed Relay. This was plotted by first solving the quadratic equation given in the previous message.

Image
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

We're a bit overdue for a progress report. The printed circuit boards for our 750 volt Insulation/Current/Voltage Monitoring Units (IMUs) finally arrived from OSH Park (replacement for BatchPCB). I don't think I'll be using the free shipping option again. It took 30 days, even though the envelope says "USPS International Priority Airmail"! [Edit: It only took 9 days from order to shipping.]

You may remember that we had a problem where the motor controller would intermittently refuse to transition the induction motor to overspeed (field weakening) and the MX-5 would bounce off a rev limit between 2000 and 3000 rpm. See the fourth and fifth paragraphs of this post back in early October. viewtopic.php?title=weber-and-coulombs- ... 953#p45953

Tritium kindly offered to look into the problem, and we left the MX-5 with James and Alex on the 17th of December. We all thought it would only take a few days to track it down. But it was a lot tougher than they thought. They eventually called in another engineer, Monte as well.

7 weeks later, we finally got the MX-5 back from Tritium, on Friday a week ago. Hoorah!

Coulomb and I drove to Tritium's workshop in the Nissan Leaf and James and Alex explained what they had discovered and what they had done to fix it.

They had found that our motor parameters, as set in the WaveSculptor, were out by a factor of about 5! This was partly my fault because the parameter extraction instructions in the manual specify using a DC voltage that is just sufficient to push 20 A thru the stator (typically less than 60 volts), whereas I found this too difficult to arrange and used the full 360 volts, so the PWM duty cycle was so low there was no resolution to speak of. The reason it took them so long to figure this out was that although they immediately suspected our parameters and restored the ones they had extracted when they had our motor on their dyno 4 years ago, it turned out that these parameters were bad too, apparently because their parameter extraction method back then wasn't very good either.

When they used the updated method with a low voltage power supply to extract the parameters, the problem seemed to have been fixed. James said his test drive had been quite enjoyable.

There was also a problem with oscillations at low speed, e.g. when parking, that seem to relate to springiness in the drive-train and our accelerator-pedal regen algorithm that uses feedback of the actual rpm to determine the zero-torque pedal position. They made a huge improvement in this by increasing the rate at which the WaveSculptor 200 sends its rpm telemetry, from 5 Hz to 25 Hz.

Coulomb drove the MX-5 home and I drove the Leaf. I got home first and waited to see Coulomb's face when he arrived.

I was expecting an "EV grin" but instead I got this:

Image

The problem was still there, albeit at a much lower level. You could get past the barrier now, but there was still some jerkiness as you pass through it. The problem was completely absent when James drove it that morning, but it had come back again.

Tritium had reduced the magnitude of our problem enormously but it was still there -- intermittently -- sometimes staying good for whole days, and sometimes bad for whole days.

Another company might well have said, by now: "Prove that the problem isn't due to your own software or hardware". But no, James stuck with it -- still trying to help us figure out what was going on.

Many thanks to James, Alex and Monte. They have upheld my high opinion of the WaveSculptor 200 and the Tritium company in general.

[Edit: Added photo. Improved grammar.]
Last edited by weber on Sun, 16 Feb 2014, 20:06, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
coulomb
Site Admin
Posts: 6357
Joined: Thu, 22 Jan 2009, 20:32
Real Name: Mike Van Emmerik
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by coulomb »

So it was finally my turn to drive the MX-5 for a week. It turned out to be five days, because my wife needed the larger car I left at Weber's place. Logistics.

I got to use "Mexy" as a real car - getting the groceries and hardware supplies, driving my daughter to school, and parking half way between where I live and where I work; there is no free parking handy to buses closer than that.

Early in my stewardship, the car was mostly good. I recall one drive from one shopping centre to another (hey - why not?) I had the top down. I was slipping past four wheel drives twice the height of the MX-5 with the wind in my hair - I felt like a rock star! But alas it didn't last, and "bad Mexy came out". Image As Weber mentioned above, she would get jerky at certain rev ranges. Fortunately there was a work-around - use a higher gear. I formulated the idea that it was motor temperature causing the problem - when the temperature reached a certain level, it seemed to play up. The controller does monitor motor temperature, and does cut back the power level when a certain temperature is reached. This seemed to explain the "memory effect" - why when it started going bad, it usually stayed that way, at least for some hours. This was the first of many theories!

On the first work day, she was good when I drove my daughter to school. This was her first drive in an electric car, so I was very pleased that it went well. She was impressed, and learned the term "EV grin" Image But soon after it went bad again. Bad enough, that I decided it was worth driving back via Weber's place, even though it's quite a detour. This was because we hadn't planned ahead, and had left the Olimex programmer with Weber (this is needed to update the Driver Controls Units (DCUs) that do miscellaneous jobs like sending pedal information to the motor controller, and driving the instruments. There were some instrument changes I particularly wanted. We also wanted to sort out some issues with the CAN-ethernet bridge that we use for data logging. To cut a long story short, someone at Tritium had made a change to suit their network, and we overlooked a checkbox that would have quickly restored it. Things like that can burn up time. On the way back from Weber's house, there is some 100 km/h freeway driving (in retrospect, I probably should have avoided this and taken another route). At above around 90 km/h, the jerkiness was so bad that I could occasionally hear the tailshaft hitting something metallic, probably a battery box. But by staying only a little above 90 km/h and staying in fifth gear, I could avoid the worst of it.

Fortunately for the rest of the time, I didn't need high speed. After the first minute or so after a charge, there would be a transition from "reasonable Mexy" to "bad Mexy". The motor temperature theory was discarded because the temperature limit before limiting power is 140C, and we never got near that figure. We know the temperature sensor isn't way off, because motor temperature is logged.

One of the instrumentation improvements was the addition of a voltmeter, displayed as one of 5 "virtual instruments" (well, one real and 4 virtual) on the tachometer. I noticed that at times the pack voltage would drop some 8 volts in a single update. Despite our battery management system, I thought somehow the pack must have a weak cell or two, and some weird communications problem must be preventing the BMS from telling us about it, and limiting motor power. Or maybe it was limiting power but in a jerky way, somehow without displaying abnormal stress on our "stress-o-meter" (we have the battery stress or discomfort displayed on what used to be the oil pressure gauge). But besides being unthinkable, it just didn't fit the facts that well.

Wednesday, my last day with Mexy, was not a great one for me. I managed to lock my keys in the MX-5, and the only other key was with Weber. By a comical sequence of errors, I managed to get my key back, the rescue EV almost ran out of charge, but it all came good in the end. Weber ended up driving Mexy to his home, and it was basically bad all the way.

As luck would have it, Mrs Weber needed the other EV that night, but part of the comedy of errors found it at my place charging. So despite an alternative ICE being available, and knowing the problems, she chose to drive Mexy, and again Mexy was bad all the way.

We had to find a solution to this problem!

Stay tuned for the next instalment!
[ Edit: some minor edits for clarity. Shipping past -> slipping past. ]
Last edited by coulomb on Sat, 15 Feb 2014, 03:44, edited 1 time in total.
MG ZS EV 2021 April 2021. Nissan Leaf 2012 with new battery May 2019.
5650 W solar, 2xPIP-4048MS inverters, 16 kWh battery.
Patching PIP-4048/5048 inverter-chargers.
If you appreciate my work, you can buy me a coffee.
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

I must add what a nice touch it was that Tritium_James arranged for the MX-5 to be washed and vacuumed before we came to pick her up.

It is possible that the MX-5 was never actually "Bad Mexy" when Tritium had her. I was very frustrated that she did not play up at all on the day I delivered her, when I took James and Alex for drives while they did CAN-bus data logging.

However, even when she was "Good Mexy" it was evident to them (but not me) that there was a problem with overspeed. They corrected this when they got the motor parameters right. But apparently they had not touched on the long-period-intermittent problem that made the difference between Dr Jekyl and Ms Hyde, although they had certainly made Ms Hyde less violent.

Image

When she was in her "Bad Mexy" phase, the difficult transition to field-weakening (overspeed) was evident even when revving the engine in neutral, but it then happened at higher rpm (around 4000) because there's no significant torque required in neutral.

We tried everything we could over the past week to figure out what it was that caused her to go from good to bad. As Coulomb described, there were many theories that were eventually tested and discarded as superstitions.

At some stage, Jeff Owen phoned and brainstormed possible causes. Among them was his observation that, although he knew we had a hall-effect pedal transducer and an AC motor, it sounded very much like what happened with his DC conversion when he had a scratchy potentiometer in his pedal transducer (pot box).

The turning point came late Monday night when we restored the ability to use the CAN-bus logging software, and Coulomb set off home from my place with logging on. I got an email from him next morning with the log file attached saying that Mexy had started off Good but went Bad around 10 to 18 minutes into the drive, but when he checked the netbook that was supposed to be doing the logging, he found it had put itself to sleep! And he had no time to look at the log.

When I looked at the log, I found that the netbook had gone to sleep after 15 minutes. Fortunately the change occurred at around 12 minutes so there it was in the log! Massive noise suddenly appearing on both the torque and speed setpoints (SP) that were being sent to the WaveSculptor. The noise was mostly around 10%, but sometimes up to 30% of full scale. The poor WaveSculptor was just trying its damndest to do as it was told! Imagine how your car would respond if you jiggled the accelerator pedal back and forth about 5 times a second. That's effectively what was happening, but electronically. The following plot shows the transition. The horizontal axis gives the sample number. Samples are 200 ms apart

Image

So what the hell was causing this! We needed to keep following the chain of causality back until we found something we could actually do something about. Was it coming from the pedal transducer, or from a dodgy cable or connector, or was it being generated by our pedal mapping software, and what made it come and go at such long intervals?
Last edited by weber on Mon, 17 Feb 2014, 08:17, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

[Check out the new photo, and the caption above it, two messages back]

The next step in tracking down the torque noise problem was to log the accelerator pedal input. We discovered that we could get an additional floating point value to appear in the WSConfig log by putting it into the unused field of the CAN packet that sets the DC Current limit.

This showed us that when it appeared at all, the noise did indeed appear on the pedal input to the EV Control Unit (EVCU - like an ECU but for an EV).

Last Friday we pulled out the Hall-effect pot-box (accelerator pedal transducer) and took it apart and inspected it under a mag lamp. We found no problem. Then we hooked it up to a power supply and 'scope, and wedged its lever at about 2/3 and pressed and twisted and banged it. We found no noise to speak of.

Then we looked at the wires from the pot-box to the EVCU. We had repurposed the wires that used to go from the throttle position sensor on the Infernal Combustion Engine, to the ECU. They were all thin wires. The ground wire was black with a light-green stripe (BLK-LGN). When we looked at the EVCU end, we found the ground wire was a thick wire, but still BLK-LGN. So what was happening in between? How did it change from thin to thick?

Eventually we figured out that the thick wire came from a connector nearby that had 6 BLK-LGN wires that all connected together there, and that the thin wire from our pot box went into this connector. So the connector was a star point for all the signal grounds from all the ICE sensors. None of these were in use and they were just chopped off at their sensors, and hopefully taped up. But what if one of them wasn't taped up, or not very well, and was intermittently making contact with the chassis? This would create a ground loop which would act like an antenna for magnetic noise from the motor or power cables. So we separated our ground wire from all the others by cutting out this Medusa and using a single crimp joiner instead.

Image

There was also a pot that we added just before we took MeXy to Canberra. It was a quick'n'dirty way of adjusting our charge current, to take the maximum allowed current from whatever power points we might find on our trip (1 x 10A, 1 x 15A or 2 x 10A). The pot was connected to spare analog inputs on both of our EVCU's and left dangling by its wires under the glovebox. It might conceivably have been a source of the intermittent noise although it's hard to see how, since its power and ground were only connected to the EVCU that the accelerator pedal was not connected to. But we removed it for good measure since we now change charge current by flipping the key to the start position while charging. The charge current is shown on the tacho.

Anyway, so far so good. 3 days with no problem. But since it was intermittent anyway, it's hard to know if we've really fixed it. We figure that due to the law of maximum inconvenience, if the problem is still there, the best way to reveal it will be to demonstrate the car to someone we really want to impress. Image

[Edit: Grammar and punctuation]
Last edited by weber on Sun, 16 Feb 2014, 20:15, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
coulomb
Site Admin
Posts: 6357
Joined: Thu, 22 Jan 2009, 20:32
Real Name: Mike Van Emmerik
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by coulomb »

weber wrote: We figure that due to the law of maximum inconvenience, if the problem is still there, the best way is to reveal it will be to demonstrate the car to someone we really want to impress. Image

Though arguably the second best way may well be posting that we've fixed it on the AEVA Image    Murphy, Lucifer, Gremlins.... are you listening?   Image
Last edited by coulomb on Sun, 16 Feb 2014, 17:26, edited 1 time in total.
MG ZS EV 2021 April 2021. Nissan Leaf 2012 with new battery May 2019.
5650 W solar, 2xPIP-4048MS inverters, 16 kWh battery.
Patching PIP-4048/5048 inverter-chargers.
If you appreciate my work, you can buy me a coffee.
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

Wow, Murphy's Law is ruthless!

If it is that BLK-LGN wiring, I'd lock that connector gremlin up in the Ghostbuster Vault.
Or nail it to a plaque.

I'm still a bit head scratchy about the "limiter" signal shown on the datalogger plot?
Would this signal be toggling more rapidly than shown on the plot?
Can you describe its formula?

And does the "Current SP" go negative for regen?
Or is this the throttle pedal position?
Wondering why the Velocity SP dips when the current sp goes below about 0.15
Is the Velocity SP linked to the Wheels or Motor RPM?
What scaling does it use?

Great to see how the Id, Iq and Idc respond to varying Current SP and RPM.

Again thanks for sharing and putting so much effort into making a great DIY EV thread.

7C
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

Thanks for your kind words, 7circle. They make it all worthwhile. And with all the help you've given us with our electronic designs in the past, you're practically an honorary SI unit. I hereby dub you "14pi radians". Image

The "limiter" signal is an integer that tells you which control loop is limiting the motor output at any given time. You can read more about it, and all the other logged data, in the WaveSculptor200 manual on pages 33 thru 37 here:
http://tritium.com.au/wp-content/upload ... Manual.pdf

What you see when she misbehaves, is that the noise on the motor current setpoint (essentially the torque setpoint) causes the WaveSculptor to flip back and forth between being limited by the motor current setpoint (limiter = 1 = "current") and being limited by the difference between the back-EMF and the maximum PWM sinewave available given the battery voltage (limiter = 0 = "PWM"). When the limiter is "PWM" the motor is said to be in "overspeed" or "field weakening".

Yes, it could be toggling faster than the 200 ms sample time. Normally it just flips once as either actual rpm or requested motor current passes thru a threshold. The threshold for either one of them depends on the value of the other.

The "Velocity SP" is actually the motor rpm setpoint. It has the same scaling as motor rpm on the graph. i.e. it relates to the vertical axis on the right of the graph.

I too would expect the current setpoint to go negative for regen, but it is standard for variable speed drives to ignore the sign of the current SP that you send. Its sign is instead determined by the sign of the difference between the actual rpm and the rpm setpoint ("Velocity SP"). So when you see these two cross over you know you are going into or out of regen.

The pedal position is not shown on that graph as we were not logging it at that stage. You can read here:
viewtopic.php?title=ac-drive-programmin ... 859#p30613
about how we derive both the current and rpm setpoints from the pedal position and actual rpm, in a way that gives accelerator-pedal regen that behaves like the engine-braking of an internal combustion engine.

[Edit:] "Rdc" is not a logged value but a calculated one. It is the approximate internal resistance of the battery in ohms, read from the left-hand scale. This was to see if the "weak cells" theory that Coulomb mentions above, had any legs. It did not.
Last edited by weber on Sun, 16 Feb 2014, 20:23, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
7circle
Senior Member
Posts: 540
Joined: Thu, 04 Feb 2010, 02:18
Real Name: Ken
Location: Melbourne

Weber and Coulomb's MX-5

Post by 7circle »

After a fair bit Image of duck diving on foc'ing dq I'm confident to say that letting your Id change around will seriously effect your Iq's (...sure mine's (that's my mind is) decelerating..) ability to torque.

So hear is a link to an small interesting research paper Nonlinear Torque Control of the Induction Motor in Hybrid Electric ...

This article looks at effieciency improvements by field weakening during low rpm's to keep the motor in a better sweet spot of less losses.

The wavesculpter200 gives you three setpoints.(edit please ignore see below --) Id,Iq and Ib (for battery but lower priority it appears)

Re: sect 18.2.1 for Motor drive for first two, then 18.2.2 for power control using Ibat.

So assuming the Rotor flux is tracked accurately ( using correct motor parameters and temp comp )
A higher layer control could target torque more accurately and take away the jitters.

Just concerned so many web hits on dq say Iq sets current (edit oops meant torque), but it's only if Id is hold the flux vector constant


mmm 14pi its a squidge of 44rads... and I don't mind being a non-dimensional being. Or are circles hyper-dimensional.... Image

(edit: 18.2.1 is for current setpoint and RPM)
( and Ib is bus current in 18.2.2 )
(Haven't found info on how the wavesculpter algorithim works ... they may be controlling torque not just Iq)
(edit: again Image )
Last edited by 7circle on Mon, 17 Feb 2014, 15:05, edited 1 time in total.
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Weber and Coulomb's MX-5

Post by weber »

I just came across this video interview and article by "Crazy" Al Bunzel, with the MX-5 and yours truly, at the Canberra EV festival in October.

Under The Covers Of A Topless Electric Mazda MX5
Revealed At The 2013 Canberra EV Festival


Thanks Al.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
Post Reply