LDU parameter tuning

[jon volk]

You should be able to ride some of the derating parameters, however, I have not played with them enough to understand how well they work.

[dougyip]

Good results at the dyno. Bottom Line: Max of 317 ft-lbs/353 HP - with low end torque limited by the dyno and conservative tune due to overcurrent errors.

Drive Unit: S85
Battery: 2/3 Chevrolet Bolt Pack 2P96S. Nominal 120Ah, 40kWh
Dyno: Dynapack Hub Dyno.

Things we learned:

- Max torque/power directly related to Fslipmax
- Fweak greatly affects peak power, not just point of max power. Looks like you would always want to be less than 300.
- Torque drops sharply after Fweak point. Fconst is very useful in boosting torque and power after fweak (thanks Jon Volk)
- With sufficient power at higher RPM stock LDU max RPM can be raised over stock Tesla. We managed to get to 17,250 RPM (fmax=575) but not 18,000 RPM

Effects of Varying FSlipMax (no use of Fconst):
Fslipmin=0, Fweak=275, Boost=1200, Fslipconstmax=Fslipmax, Fmax=550, Fconst=550, udcnom=0

Other pulls using Fconst
30- Jon Volk: Fweak=258, Fconst=450,fslipmin=0.87,fslipmax=3.06,fslipconst=450, fslipconstmax=9. Boost reduced from 1850 to 1500 to keep within dyno current limits.
31- Fweak=275, Fconst=450,fslipmin=1,fslipmax=3.25,fslipconst=450, fslipconstmax=6, Boost=1500
32 - Same as 32 but Boost = 1200
33 - Same as 32 but Fweak = 258
34 - Same as 31 but fslipconstmax=9

[dougyip]

Even with ocurlim, iacmax, and idcmax set to be out of play we were still getting overcurrent errors.

With fweak at 350, boost = 1200, couldn't raise fslipmax above 4.5
with fweak at 300, boost = 1200, couldn't raise fslipmax above 4.0
with fweak at 275, boost=1200, couldn't raise fslipmax above 3.75

Any ideas why? I see that Damien is using fslipmax =5 and the EVShop Porsche = 6

[johu]

I'm a bit surprised that torque drops off with speed below your fweak. Are you running version 4.55.R or higher?
Lowering fweak OR raising slip means increasing motor current. Thats why you have to compensate on with the other.
I'm not 100% if desat errors are reported reliably on Tesla hardware. So if you say comparator is taken out of the game you might be seing desat errors declared as over current

[dougyip]

Firmware version 4.57.R.

I am attaching a datalog file for run 23 (Fslipmax=4.0,Fweak=275). Pack volt/amps are measured at the battery by our Orion2 BMS, UDC/PotNom/Freq/IDC are logged via CAN from the open source controller. Pack Volts/UDC agree fairly well. There is a huge discrepancy between PackAmps and IDC. Either there is a reading error or huge inefficiencies.


Dyno Run 23
Fweak=275 Fslipmin=1.0 Fslipmax=4.0 Boost=1200

PackVolts PackAmps Power (kW) UDC PotNom Freq IDC
372 19 7 371 35 41 0
371 17 6 370 42 249 8
371 17 6 370 49 416 17
369 116 43 370 52 750 12
362 194 70 366 54 2040 70
359 200 72 360 60 2625 108
358 211 76 357 72 2083 112
358 240 86 357 79 1916 114
358 248 89 355 81 1916 132
356 272 97 354 97 1749 133
355 278 99 352 100 1749 153
355 270 96 351 100 1665 161
354 283 100 352 100 1623 160
353 288 102 352 100 1623 157
353 300 106 351 100 1708 164
352 321 113 350 100 1831 171
349 330 115 349 100 1957 180
349 366 128 348 100 2165 200
349 385 134 346 100 2291 206
347 401 139 345 100 2415 215
345 411 142 343 100 2581 227
343 440 151 342 100 2748 244
341 451 154 339 100 2874 248
340 484 165 339 100 3041 262
338 495 167 337 100 3208 273
339 520 176 336 100 3375 284
335 537 180 335 100 3583 304
336 564 190 334 100 3665 294
332 594 197 332 100 3873 319
331 601 199 331 100 3999 326
329 615 202 330 100 4166 339
328 642 211 328 100 4333 351
329 660 217 326 100 4500 360
329 661 217 324 100 4665 372
328 682 224 323 100 4831 375
326 711 232 322 100 4957 393
324 728 236 320 100 5124 397
323 742 240 319 100 5291 407
320 768 246 318 100 5499 423
318 781 248 317 100 5666 435
319 798 255 314 100 5790 438
316 823 260 312 100 5874 438
314 836 263 314 100 6123 458
315 841 265 312 100 6249 461
314 871 273 310 100 6416 470
311 872 271 309 100 6583 487
310 894 277 310 100 6750 481
309 904 279 307 100 6958 507
306 958 293 305 100 7081 501
311 913 284 303 100 7207 505
306 975 298 302 100 7374 499
307 970 298 302 100 7582 536
305 966 295 301 100 7665 538
301 998 300 300 100 7916 551
299 1020 305 297 100 8083 561
300 994 298 296 100 8250 551
297 990 294 296 100 8332 556
299 970 290 296 100 8499 541
301 972 293 298 100 8625 535
301 962 290 297 100 8790 526
300 949 285 298 100 8915 520
303 933 283 301 100 9123 510
303 907 275 299 100 9123 497
305 890 271 303 100 9375 510
303 913 277 303 100 9583 502
308 855 263 302 100 9750 477
304 846 257 302 100 9873 469
309 856 265 301 100 10081 464
308 840 259 304 100 10248 462
307 824 253 304 100 10415 448
306 804 246 307 100 10582 435
308 799 246 307 100 10708 425
310 785 243 306 100 10875 427
312 783 244 309 100 10916 424
311 768 239 310 100 11206 413
310 757 235 308 100 11415 409
311 736 229 309 100 11581 404
314 732 230 309 100 11707 391
315 730 230 313 100 11874 392
315 715 225 311 100 12082 387
315 723 228 312 100 12208 376
314 690 217 310 100 12375 376
316 687 217 312 100 12540 370
315 690 217 313 100 12750 362
315 670 211 313 100 12706 358
315 666 210 313 100 12999 342
314 667 209 311 100 13207 364
317 658 209 315 100 13333 338
318 642 204 314 100 13500 343
318 640 204 316 100 13749 331
322 654 211 312 100 13831 341
321 608 195 319 100 13998 341
317 614 195 316 100 14165 335
316 643 203 315 100 14332 305
317 595 189 321 100 14332 331
319 590 188 317 100 14666 307
321 610 196 319 100 14833 319
322 606 195 316 100 15000 302
323 596 193 318 100 15165 282
325 563 183 320 100 15331 306
320 563 180 321 100 15457 283
323 558 180 323 100 15665 281
324 556 180 321 100 15832 303
324 543 176 318 100 15958 274
323 493 159 321 89 16125 296
329 439 144 322 100 16081 246
330 463 153 327 71 16207 220
329 485 160 325 97 16125 242
329 466 153 323 68 16207 233
328 483 158 325 83 16166 248
324 493 160 327 82 16166 245
324 420 136 324 81 16166 247
329 252 83 329 0 16125 201
340 96 33 337 0 15873 98
351 27 9 349 0 15583 29
355 15 5 353 0 15208 9

[jon volk]

Really great data! Thanks for doing this and sharing.

Was udcnom in use when running my parameters?

[dougyip]

Udcnom was not in use for any of the runs. Your power and torque should be better as two Volt packs have less sag and more Power capacity.

[jon volk]

Alrighty folks.

If you happen to follow my nonsense on Instagram, you'll know Ive been spending a bunch of time tuning the parameters on my LDU for more performance of the past few months. I have been using a Dragy 10hz gps box to record times for comparing changes. Throughout the process, I tried to make all timed runs at similar SOC and battery temps. Tuning for power was done in the same process that I would use for an ICE combination. Increases in 1/8 and 1/4 mph equate to power gains. If a change did not net any improvement, it was reset to the previous value. Similarly, 60' time and 0-60 were used as measures for not just power, but how effectivly/aggresively it was delivered.

For context, the attached JSON file are the parameters I run in my 3,150lb with driver E30 with a standard base Tesla large drive unit and two Chevy Volt battery packs in parrallel. The car so far has been 3.0 seconds 0-60mph (2.81 if comparing with 1' rollout the magazines use) and 11.02@120mph 1/4 mile. This is on a 26x8.5" bias ply drag slick. There is a bit more power to be had, but I'm out of traction at this point.



Some basic notes/observations on the parameters and tuning them for your own combo.

Boost: This is one I struggled with a bit and now ultimately map over CAN based on pot value. More on that later. The LDU does NOT like higher boost values under any sort of high load, low to mid throttle input. Running a boost value at say 1700 or higher would result in overcurrent shut downs when attempting to start up a steep hill or a quick throttle transition shortly after inverter startup. I had also chased a low-ish parking lot speed oscillating surge that was downright annoying at times. Both of these issues have since gone away for good when setting boost to a lowly 1400.

Fweak:The LDU makes a bunch of torque. More than most reasonable cars can use. Fweak is really the key to regulating where in the rpm range the power is delivered. For me 238hz will stick my slicks and net the above posted acceleration times almost every hit. Bringing this to 233hz will just burn the slicks. 258hz worked well for me on street radials.

Fcost/Fslipconstmax:I've found these values net the highest 1/4 mile mph. There is not much point in changing these as you will not likely find any gains and reduction will only hurt higher speed freeway/highway/motorway performance.

Fslipmax: I run a parameter value of 3.18 to 3.3. Dont be fooled if you compare this to the parameter files on Damiens Github. With UDCnom factored in, this value is really 4.0-4.5 under voltage sag. Higher value=more power. Going too high will slam an overcurrent trip. Value needs to be a bit lower if battery temps are over 95-100F on a full SOC for me. I'm presently working on mapping this.

IDCmax: This is the DC current limit and a good parameter to use as a power ceiling and preventing overcurrent trips. I can get away with a slightly higher fslip value if this is used effectively. I do not have a standalone current sensor, but based on the hp calculated by mph and weight, I believe the DC current values shown in the software are roughly 50-60% of actual. Particularly what Tom has said they've seen their skyline pull for amps. All I know is the trap speed and weight of the car calculate to more power than the software says is being used. As such, an IDCmax of 760 seems to work well for me.

Throtramp:This is really a subjective setting but can be used to fine tune launching the car at WOT. If you go WOT while holding the brake and then release the brake pedal. Throtramp will control how fast throttle is applied. Think of it as another tool to control the onset of low end torque. I run anywhere from 65-75 and adjust based on tire temp. For regular driving, Id start off around 5 and move up from there if you want.

Idle:My parameters have a brake-off idle that will roughly simulate an ICE motor and cruise around a parking lot around 10-15mph without throttle input. Ive found this to be the most natural and comfortable setup. Be warned, boost, ampmin, slipstart, and idle throttle limit all play into how the car behaves in idle. If you start chaning these, it may take some playing with to get a satisfactory performance.

Regen: I use a 0-5v pressure transducer to control regen. The harder I press the brake, the more the motor regens. I do not have it set up to do any "one foot" driving off throttle regen. When driving the car hard, this feels most natural to me. It also feels more efficent being able to just coast throttle off. I have no emperical data on that. I just "feels" better. I believe Porsche didnt do any, or very little at most, throttle off regen on the Taycan for whatever thats worth. If you want throttle off regen, you are on your own with those settings.


Overcurrent faults: I've beat the shit out of my inverter and hit overcurrent shut down more times that I care to count in exploring how far I can take this thing. So much so that I had written a function meant to auto retart the car if it hit a fault. Doing so brought to light that these overcurrent events are more than likely desat faults from a gate driver on one or more inverter phases. I discovered this when the openinverter board would transition back into a run mode, but no PWM generation in the inverter woudl take place without a power cycle. This pointed to the issue being a Tesla hardware sourced fault rather than openinverter software triggered. From my original LDU with a bad IGBT that never turned an rpm in anger, I know that a desat fault will show as "overcurrent" in the software.


UDCnom: Ive posted this spreadsheet before, but here it is again so if you decide you want to deviate from my 390 value, you can see how it will affect other parameters. You can also input throtramp value and itll tell you how long in seconds from 0 to 100% throttle to compare scenarios.


CAN mapped boost: Running a measly 1400 for boost means there is some degree of decreased perfromance up higher. Its still more than enough power to kill yourself if you're not used to decently quick cars, but not gonna cut it for me. To deal with this, I map boost to pot value over CAN. The following code is on a Teensy3.6 with flexcan and works very well. My pot min and max are 830 and 4095, respectively. I map boost from 1400 to 1840 between pot values 3700 and 4095. This allows me to get full axle abusing startup torque when holidng WOT and brake on, yet smooth reliable take off on inclines or otherwise with lower throttle input. I had to mess with the range and scaling a bit to get the transition I wanted, but the below works well for me.

Im only including the mapping function as there is enough elsewhere on the site in regards to mapping CAN values.

Code: Select all

void boostMap()
{
    if (pot > 3700) {

        boost = map(pot, 2700, 4095, 100, 230);

        msg.id = 0x601; //set parameter ID
        msg.len = 8;
        msg.buf[0] = 0x40; //CMD
        msg.buf[1] = 0x00; //index:boost=0, count down for index number
        msg.buf[2] = 0x20; //
        msg.buf[3] = 0x00;
        msg.buf[4] = 0x00;
        msg.buf[5] = boost;//value x 32
        msg.buf[6] = 0x00;
        msg.buf[7] = 0x00;
        Can0.write(msg);
    }

    else {
        msg.id = 0x601; //set parameter ID
        msg.len = 8;
        msg.buf[0] = 0x40; //CMD
        msg.buf[1] = 0x00; //index:boost=0, count down for index number
        msg.buf[2] = 0x20; //
        msg.buf[3] = 0x00;
        msg.buf[4] = 0x00;
        msg.buf[5] = 175;//value x 32
        msg.buf[6] = 0x00;
        msg.buf[7] = 0x00;
        Can0.write(msg);
    }

}

Go forth and burn tire.

updated tesla LDU parameters.json

(1.06 KiB) Downloaded 164 times

[Isaac96]

Bookmarked. Thank you Jon!

[P.S.Mangelsdorf]

Jon, this is amazing information! This much context is truly appreciated.

This will easily save myself (and others) months of headaches. We all owe you a drink or 5.

[Boxster EV]

Very useful post and summary, Jon.

Further to our separate discussions, I have two pieces of feedback from my testing over the past few days:

My OC issues which have been a nuisance from the beginning, but since dropping the BOOST parameter to 1375, they seem to have a gone. I can pull away at any speed and on any steep gradient without tripping. There has been just one exception: During some testing yesterday with 50% charge, I changed direction quickly (R to F), went to pull away hit OC.

I still need to drill into IDLEMODE. As you’ve previously reported, having it set to ON, seems to help and makes the car feel like an OEM auto. However when I enable that parameter and set IDLESPEED into a positive number it works as expected. However when I power-cycle, the logic board won’t close the positive contactor.

[Jack Bauer]

Thanks for your hard work Jon. I never try to promote the Openinverter Tesla board as being better than other solutions despite them seemingly being quite happy to do the reverse. However, your efforts have proven beyond any doubt that there is no loss of any "Tesla Magic". It really is the EV version of a standalone ecu.

[johu]

There has been just one exception: During some testing yesterday with 50% charge, I changed direction quickly (R to F), went to pull away hit OC.

Yes, quick direction changes were never a development goal since mister lead developer uses the mechanical transmission to change direction

I still need to drill into IDLEMODE. As you’ve previously reported, having it set to ON, seems to help and makes the car feel like an OEM auto. However when I enable that parameter and set IDLESPEED into a positive number it works as expected. However when I power-cycle, the logic board won’t close the positive contactor.

That is intended and also happens in OEM vehicles. As soon as idle mode is enabled you need to press the brake pedal to start.

[Boxster EV]

There has been just one exception: During some testing yesterday with 50% charge, I changed direction quickly (R to F), went to pull away hit OC.

Yes, quick direction changes were never a development goal since mister lead developer uses the mechanical transmission to change direction

I still need to drill into IDLEMODE. As you’ve previously reported, having it set to ON, seems to help and makes the car feel like an OEM auto. However when I enable that parameter and set IDLESPEED into a positive number it works as expected. However when I power-cycle, the logic board won’t close the positive contactor.

That is intended and also happens in OEM vehicles. As soon as idle mode is enabled you need to press the brake pedal to start.

Thanks for the steer. I’ll give it a whirl.

To continue enhancements in firmware development happy to support through Patreon.

[jon volk]

Thanks for your hard work Jon. I never try to promote the Openinverter Tesla board as being better than other solutions despite them seemingly being quite happy to do the reverse. However, your efforts have proven beyond any doubt that there is no loss of any "Tesla Magic". It really is the EV version of a standalone ecu.

Thanks Damien. Like a traditional standalone, its more work up front and theres plenty of opportunity to mess things up, but the control-ability over multiple platforms is way more appealing to me.

[Roadstercycle]

Hi, I am trying to change a couple parameters on my LDU but it seems they will not save. It has worked fine in the past and I'm getting a good wifi signal. I change one setting at a time then hit enter, refresh page and it is OK. Then cycle power and then it's back to the original setting. Any one else had this issue. I've tried 2 computers to eliminate that possible issue of the enter button not working. Any ideas?

[jon volk]

Did you click save to flash at the top of the page?

[Roadstercycle]

Jon, Thank you, I'm sure that's it. Just forgot, it's been awhile since I played with the parameters but your latest parameter post and explanation got me motivated to see if I can do a little better... Again thank you.

[P.S.Mangelsdorf]

I've begun to dig into optimizing the tune for my car. I started out with Jon's updated parameters and tuning guide, and the first major change was to adjust the udcgain because it was way off and causing issues with regen, phantom spinning etc.

Since this change, I now slam into an over current fault whenever I stand on the throttle. I think I need to adjust the current limit, but I first started examining how far off the current reading was from reality. Based on comparisons to my ISA shunt, output via the CHAdeMO web interface, the LDU is reporting a calculated idc of roughly 16% of what the ISA shunt is seeing. Being that it is this far off, I'd like to get the current reading correct before playing with settings.

I assume the adjustment needs to be in the two current sensor gains, but because those are on the AC side, I don't have a good way to figure out how far off they are.

Has anyone figured out what the current gains need to be for a base LDU?

[jon volk]

I should probably update the parameters now that there’s been some fundamental change to the newest firmware with fweak stuff.

[P.S.Mangelsdorf]

I should probably update the parameters now that there’s been some fundamental change to the newest firmware with fweak stuff.

I'm still running the 4.97 firmware at the moment, but I (and I'm sure others) would appreciate it. Your parameters were a great starting point since I'm running a similar setup.

[johu]

There will be an update to idc calculation in the next release, I think the current calculation is like 30% off. But yours seems 600% off? Default ilXgain is 1 for LDU boards, that would mean 2048A maximum swing. Sounds plausible actually, no?

[P.S.Mangelsdorf]

There will be an update to idc calculation in the next release, I think the current calculation is like 30% off. But yours seems 600% off? Default ilXgain is 1 for LDU boards, that would mean 2048A maximum swing. Sounds plausible actually, no?

Seeing as my ilgain setting is currently 4.68, that's way different than 1.

Edit to add: fixing this seems to have me at least close to the ballpark in terms of DC current reading.

[P.S.Mangelsdorf]

So I took the car out to make a few pulls, and tried several different settings. I continue to slam an overcurrent fault when launching hard. It feels like it is occurring right as the throttle ramp ends, i.e. when it should be delivering full power.

Oddly (to me at least), it also shows an encoder fault when it trips the overcurrent fault (see below screenshots below, yes this is from this event, I cleared the errors before the test).

The screen shots are from an attempt with idcmax of 740, but I tried as low as 700 with the same results.

Anyone see something glaringly wrong or do I just need to keep pulling back?

[Isaac96]

idcmax is a pretty slow limiter, probably won't prevent an OC event. It's more often used for reducing power long-term to protect batteries, and even then only works well with FOC (and thus PM motors).

Raising fweak might do the trick, as it will move power delivery further to the right on a torque curve.

Mr. Volk's modified software with the new parameters made significant driveability improvements to my SDU-powered vehicle (before the third inverter blew), it ought to improve your experience a lot.