AC drive programming and pedal mapping

[weber]

When you put the regen slider down near zero it becomes essentially torque control (but with a square law on the torque and requested speed going to zero at zero pedal). Same formulas.

[weber]

Oh, but I forgot to mention that because torque(really current) requests are unsigned, what you actually have to send to the VF drive is the following.

[Edit: Tritium_James explains later why this is not necessary and the previously posted curves are just fine.]

Note that the speed curve must be very specific to ensure that the drive gets the correct sign for the torque. So you can see why I find unsigned torque strange.



[Edit: Changed chart to show more accurate curves]

[7circle]

Will this aproach be for the wavesculpter or an industrial drive.

It is useful for any EV in getting a good mapping of throttle positioning to motor load.

The Propulse (EV focused) controller talks in terms of Power not torque and has a dial to adjust max power output. But could be a translation issue and be torque in the way it controls the motor.

It makes more sense to have throttle controlling torque, with some compensation for speed so adjust regen levels.

[woody]

One crazy idea I had for ice-like pedal response was to do some rough sums on the ICE - e.g. throttle position X = Y sq cm of throttle opening.
at X rpm = Y litres of air per second
calculate suction on the cylinders => engine braking
air volume ~ power

I didn't get very far with this, obviously

Are you going to try and sync the motor with the gearbox when you push the clutch in? I.E. work out gear ratio by comparing motor and tailshaft speed, then on clutch pedal change motor speed to match next gear (with low torque demand)

[coulomb]

Oh, but I forgot to mention that because torque(really current) requests are unsigned, what you actually have to send to the VF drive is the following...

I wonder if you can add a term in there linear with pedal position such that the inflections occur along a line starting at 0,0 and sloping slightly down, so that when you do the calculation and reject negative values (oops, but having reflected the left part of the line already), you end up with short "dead spots" on the pedal graph, which allow for coasting if desired.

Or really, I guess what is wanted is to pull the curve apart at the reflection point, and add a short dead zone there. So you would shorten the graph along the x-axis by say 5%, and insert 5% of zero torque/current where the inflection hits the y-axis (where it previously just touched zero torque at a single point).

I don't know if this dead zone would ruin the cool feeling that TJ describes. If so, maybe the dead spot also needs to be adjustable or at least switchable. It starts to get difficult to present these options to the user in a reasonable way. What English-free "international" symbols would you put on the maximum regen slider and the dead zone switch?

[weber]

Will this aproach be for the wavesculpter or an industrial drive.

We are currently using the WaveSculptor, but it is equally applicable to industrial drives (like the Control Techniques drive we still have, which would be great for a van or ute, and can overvoltage a motor twice a much as a WaveSculptor can. Hint. Hint. ).

I suppose that means we've departed a little from the title of this thread. But I started it, and I'm quite happy that it has become the "Pedal Mapping" thread.

It is useful for any EV in getting a good mapping of throttle positioning to motor load.

I'm glad you think so. The more I look at it, the more obviously right it seems, but the real test will only come with the driving. I'm afraid that may yet be some months away. Tritium James could try it much sooner in the Civic.

The Propulse (EV focused) controller talks in terms of Power not torque and has a dial to adjust max power output. But could be a translation issue and be torque in the way it controls the motor.

Could be a translation problem, but if it's really power and not torque, then just take my requested torque function and multiply it by actual speed to get requested power.

T = p^2 + (p^2-1)*k*s
P = T * s
= p^2*s + (p^2-1)*k*s^2
where T is requested torque, P is requested power, p is pedal position, k is max regen and s is actual speed.

It makes more sense to have throttle controlling torque, with some compensation for speed so adjust regen levels.

Yes. That is a minimum requirement for anything resembling ICE response, and I believe the above curves do it as smoothly and simply as possible. Although I note that there is also an equivalent formulation where the accelerator pedal (there is no throttle any more) controls speed, with some compensation for actual torque to adjust regen levels. That's how Ross Pink did it. I may have a go at plotting a smoothed version of that scenario later.

And of course, the pedal controlling requested power with compensation for actual speed, can be made equivalent as above.

[weber]

One crazy idea I had for ice-like pedal response was to do some rough sums on the ICE - e.g. throttle position X = Y sq cm of throttle opening.
at X rpm = Y litres of air per second
calculate suction on the cylinders => engine braking
air volume ~ power
I didn't get very far with this, obviously

I can help you with the first part. A sketch and a bit of trig shows that (normalised) throttle opening cross-section is just 1-cos(pi/2*p). It's possible that the other factors are approximately linear (until they hit their limits).

There is no significant difference between a 1-cos(pi/2*p) pedal response and an x^2 pedal response, in the range 0 to 1. But x^2 is a lot easier to program on a microcontroller. Not so easy for an analog circuit. Easiest nonlinear functions for analog are log and exponential, using diodes. The following scaled and offset exponential function is very close: (8^p-1)/7. Or else it can be done piecewise-linear.

Are you going to try and sync the motor with the gearbox when you push the clutch in? I.E. work out gear ratio by comparing motor and tailshaft speed, then on clutch pedal change motor speed to match next gear (with low torque demand)

No. We just plan to have pedal response close enough to ICE response that you will change gears in exactly the same way as with an ICE. I believe that the requested-speed curve shows exactly what speed you will get for your pedal position when the clutch pedal is in (or you're in neutral), except that there will also be maximum ramp rates that simulate an ICE flywheel to the minimum degree necessary.

There will also need to be a modification to the mapping to include an idle speed at zero pedal in the case where the aircon is on and the gearbox is in neutral or the clutch pedal is in.

[Edit: Added chart]

[weber]

I edited a chart into my previous post in this thread, showing how three different mathematical formulae give very similar curves.

I've also changed the thread title from "Industrial AC drive programming" to "AC drive programming and pedal mapping" at coulomb's suggestion. Thanks coulomb, for telling me how (just edit the first message).

I wonder if you can add a term in there ... [so] you end up with short "dead spots" on the pedal graph, which allow for coasting if desired.
...
I don't know if this dead zone would ruin the cool feeling that TJ describes. If so, maybe the dead spot also needs to be adjustable or at least switchable.

I don't understand the desire for such dead zones at zero torque at all. I don't experience them in an ICE. If coasting is good, regen is even better. I think they would ruin the "cool feeling".

In our case you can always coast by taking your foot off the accelerator and holding the clutch pedal in, or putting it in neutral, as with any ICE car.

I can however see the need for a small amount of hysteresis in the pedal response, which is a very different thing. It would prevent small bumps and jiggles from changing the set points, at any pedal position, and break possible nasty destabilising feedback loops between vehicle accelleration, your foot and the pedal position parameter.

It starts to get difficult to present these options to the user in a reasonable way. What English-free "international" symbols would you put on the maximum regen slider and the dead zone switch?

No dead zone switch required, but max regen slider could be:



[Edit: Added part about coasting in neutral]

[Tritium_James]

Weber, I reckon that mapping would be quite nice. It gives a similar effect to what the setup with an ACP system is, where the 'zero' point of the pedal moves along the range of travel as speed increases. This means no dead pedal travel when moving off from a stop, but regen controllable by the same pedal once you're moving.

FYI, the ACP system had a linear movement of the 'zero' point between pedal fully up at 0km/h and about 1/3 of the way down at 10-15km/h. So your proposed movement of 0 to full regen over 40% travel is slightly more physical travel, and up to 100% speed (compared to 10-15km/h) is a much bigger band. I think the only way to know if this is better or worse is to try it and see!

[coulomb]

I don't understand the desire for such dead zones at zero torque at all. I don't experience them in an ICE.

Right. But an ICE doesn't have extreme regen either (well, I suppose that first gear is an exception, but that can get annoying too, with those waves of decompression that sometimes happen).

In our case you can always coast by taking your foot off the accelerator

Are you considering a clutch switch then? Surely by putting in the clutch, the motor will go to zero speed (yes, the vehicle will coast), but then changing down a gear would be more difficult. Even changing up a gear might be more difficult than of the motor and vehicle could both coast. Then again, the low inertia of the motor may negate that problem.

and holding the clutch pedal in, or putting it in neutral, as with any ICE car.

With an ICE, the engine has a fair bit more inertia, and won't go to zero speed unless something goes wrong (stalled engine). I suspect that having the motor go almost instantaneously to zero RPM when the pedal is released will feel unnatural.

I can however see the need for a small amount of hysteresis in the pedal response,

Simulating slack in the pedal cable and/or throttle linkages, perhaps. Or perhaps our pedal cable will provide natural hysteresis as it is   

which is a very different thing.

Granted, but I'm not sure that hysteresis would feel familiar or desirable either.

It would prevent small bumps and jiggles from changing the set points, at any pedal position,

Well, it will prevent bumps from changing the set points in one direction, but not the other. If I understand what you want to do correctly (I probably don't).

and break possible nasty destabilising feedback loops between vehicle acceleration, your foot and the pedal position parameter.

Yes, those would be bad.

max regen slider could be:

The letter B is English; not allowed. Though I notice that some symbols have "ABS" in them.

I suggest something like



The red colour isn't necessary or probably even desirable; it's just the first brake symbol I found, and I added the lightning bolt to suggest electricity.

[Tritium_James]

'B' is used in plenty of mainstream EVs (Leaf, iMiev, etc) to indicate the drive mode that has increased regen (compared to 'D' mode).

Weber, the Wavesculptor uses the absolute value of the incoming command torque, so no need to reflect your torque commands in the Y axis. You're still only going to get regen if command rpm is less than actual rpm, though.

[a4x4kiwi]

Hi guys,

Just a reminder that the Azure Dynamics Pedal Application Manual makes good reading.

Mal.

[weber]

I don't understand the desire for such dead zones at zero torque at all. I don't experience them in an ICE.

Right. But an ICE doesn't have extreme regen either.

But don't forget, that with my mapping (as with Ross Pink's and AC Propulsion's but not Azure Dynamics'), at zero "throttle" as the speed decreases so does the torque, so the speed will decrease at an ever-decreasing rate. If it wasn't for losses it might be like Zeno's paradox and never actually get to zero.

Are you considering a clutch switch then?

As mentioned in reply to Woody, a clutch switch would be good for aircon idle. We already have a neutral switch and a reverse switch on the standard MX-5 manual gearbox. But it might be a good idea to use the clutch switch also to reduce the max regen to something small when the clutch pedal is in.

Surely by putting in the clutch, the motor will go to zero speed (yes, the vehicle will coast), but then changing down a gear would be more difficult. Even changing up a gear might be more difficult than of the motor and vehicle could both coast. Then again, the low inertia of the motor may negate that problem.

Right. The low inertia may work to your advantage as well. With your foot off the accelerator and the motor stopped, the torque is zero, so the synchro-mesh may spin it up again more easily. I'll worry about it after we've tried it.

With an ICE, the engine has a fair bit more inertia, and won't go to zero speed unless something goes wrong (stalled engine). I suspect that having the motor go almost instantaneously to zero RPM when the pedal is released will feel unnatural.

You must have missed my post in reply to Woody above. I mentioned setting maximum ramp rates to simulate a light flywheel. Then again, we do actually have a light flywheel.

The letter B is English; not allowed. Though I notice that some symbols have "ABS" in them.

It's also German and a few other European languages and Japanese(phonetically). What's next to the "gear lever" position for increased regen braking in your Prius?

[weber]

Here's a chart showing the mapping when the max regen slider is in an intermediate position. This time 10% max regen. Possibly more like a real ICE.

[weber]

Here's another way of looking at the same data. Instead of putting pedal position on the horizontal axis and showing torque curves for various actual speeds, we put actual speed on the horizontal axis and show torque curves for various pedal positions.

[weber]

Here's the C source-code for the parts of "process_pedal()" we changed in Tritium's Driver Controls unit:

void process_pedal( unsigned int analog_a, unsigned int analog_b, unsigned int analog_c, float motor_rpm)
...
        // Scale pedal input
        pedal = pedal / PEDAL_TRAVEL;
        // Check pedal limit and clip upper travel region
        if(pedal > 1.0) pedal = 1.0;
...
                 // Dave Keenan's quadratic accelerator-pedal regen algorithm
                 // See viewtopic.php?p=30613&t=1859#p30613
                 float normalised_rpm = motor_rpm / RPM_FWD_MAX;
                 float p2 = pedal*pedal;                                  // p2 is pedal squared
                 command.current = (CURRENT_MAX * (p2 + (p2-1)*regen*normalised_rpm));
                 if (p2 >= ((1-p2)*regen))
                          command.rpm = RPM_FWD_MAX;
                 else
                          command.rpm = RPM_FWD_MAX * p2/((1-p2)*regen);

I note that the above avoids divide-by-zero and floating-point-overflow problems that a literal implementation of the previously given algebraic equations would have.

However there are some minor jerky artifacts at present when driving the MX-5 with this code. These are apparently due to the fact that drive commands are sent to the WaveSculptor every 100 ms but the WaveScupltor only sends motor_rpm every 200 ms. James is working on a solution.

[Edit: URL font change]
[Edit: Fixed broken link to Tritium. Corrected spelling.]

[Renard]

Weber has kindly made the re-writes of the code for me to load into my Tritium Driver Controls Unit.
After some hiccups, I have got the code compiled and into the DCU.
I can report that according to Tritium's DCtest program it seems to be working as it should. For example:



shows that the requested speed at about 10% pedal position is 1880rpm as predicted by the formula for a speed of zero. That is, 1880rpm is the speed at which the request for torque (current) is zero. At higher speeds the requested current is negative.
At zero speed, [Edit No, I mean zero pedal.] the program shows requested current is linearly negative up to a maximum of 25% of maximum current at 4000rpm. (Thus verifying the relation T = - k*s when p=0.)
Now 4000 rpm is about 150km/hr for my car, but if I need more regen at any given speed, I can always shift down a gear thereby increasing the speed as seen by the motor.

In relation to the comment of Weber's about 5Hz speed reports from the controller, I checked with Tritium whether anything had been done about the problem Weber mentions. Apparently not.

[weber]

In relation to the comment of Weber's about 5Hz speed reports from the controller, I checked with Tritium whether anything had been done about the problem Weber mentions. Apparently not.

Tritium have recently increased the telemetry rate for "velocity" (rpm) to 25 Hz in our WaveSculptor. This has given us a significant reduction in the magnitude of oscillations due to drive-train springiness during low-speed low-gear parking manoeuvres, when using our quadratic regen pedal mapping.

After further testing and documentation, this should eventually make its way into a WaveSculptor200 DSP firmware update, with a version number greater than 1.11, on the Tritium website.
http://tritium.com.au/products/wavescul ... -inverter/

[Edit: Removed whitespace]