Drivetrain measurements

[bwilson4web]

Having owned a Prius for six years, I'm still learning and a recent question had me go back and take more measurements:

• What is the Prius transaxle efficiency?

Fortunately, it is possible to measure it and this is a look at some of the preliminary data.

A modified Prius was driven from 'Big Ed's Pizzeria' to 'North Alabama Computers' in Huntsville AL while recording these data elements:

• MG1 torque (Nm)
• MG2 torque (Nm)
• MG1 rpm
• MG2 rpm
• Traction battery volts
• Traction battery amps

Then using a few, key relationships, intrinsic to the transaxle axle, we can calculate:

• ICE rpm
• ICE torque

Those relationships are:

• ICE rpm = (MG1_rpm + (MG2_rpm * 2.6)) / 3.6
• ICE torque = MG1_Nm / 0.28
• mph (Indicated) = MG2_rpm / 57
• power (W) = Nm * 2 * pi() * rpm / 60

But sampling is slow, ~0.50 seconds per sample, in this file. I will try a faster sampling rate which will help reduce the 'outliers.'

So this is what the 5 minute, ~300 second, trip looked like:


Initially, I had to wait in the parking lot for traffic to clear:

The initial engine start shows a peak of ~8kW to spin it up. Reverse is only via the traction battery and MG2. Then the Prius tries to use EV mode as much as possible until the catalytic converters reach operating temperature. Once stopped, the car recharges the traction battery and then shuts-down the ICE. Vehicle power is provided by the traction battery.

Then I accelerated to the 50 mph posted, speed limit:

The most efficient acceleration is one that does not draw on traction battery power. This is easily monitored in the cabin energy flow screen. Between seconds 80-85, the data points suggest MG1 ~= 9kW and MG2 ~=8kW giving roughly 88% efficiency. However, faster sampling and the actual numbers will give a more accurate reading and interpolating the traction battery load will also add to the energy picture.

Once at speed, cruise control handled the various overpasses:

The Parkway has a number of overpasses and this resulted in the traction battery doing a lot of swings from charge to discharge. This is not a flat route suitable for steady-state metrics plus I had to deal with lunch-time traffic.

Finally, I decellerated and came to a stop at North Alabama Computers:

With a lot of regenerative energy charging the traction battery, there was no need to run the engine. It was functionally an electric car as I exited, changed lanes, turned into the parking lot and found a parking place.

Sad to say, the Prius transaxle goes through many state changes that occur faster than the sampling interval. This results in some of the calculated values having 'impossible' or 'outliers' simply because we don't multichannel data recording. But these can be minimized by long runs in steady state conditions along with 'outlier' detection. Regardless, even this preliminary data provides insights to Prius transaxle efficiency.

Sad to say, my GPS mouse is missing the USB-to-PS2 adapter and the one I got at North Alabama Computers did not work out. I've ordered another GPS mouse and when it arrives, I'll be able to add true speed and altitude for kinetic and potential energy. The ultimate goal:

efficiency = output / input

Bob Wilson

[dcb]

I don't think the question was about the transaxle as a whole, as there are numerous gear losses not being represented here, but it was in relation to the relative electrical power going through mg1 and mg2 and the battery. It was not clear if each motor and the battery had their own current sensors even, because you cannot infer one and then use that inference to measure efficiency (otherwise they will all be 100% efficient).

[bwilson4web]

Originally Posted by dcb

I don't think the question was about the transaxle as a whole, as there are numerous gear losses not being represented here, but it was in relation to the relative electrical power going through mg1 and mg2 and the battery. It was not clear if each motor and the battery had their own current sensors even, because you cannot infer one and then use that inference to measure efficiency (otherwise they will all be 100% efficient).

Patience, there are more than one way to measure losses. Certainly current is interesting but it also needs voltage and these are pulsed and pulsed-width-modulated power flows. This poses an interesting technical challenge of trying to measure RMS values from a complex waveform.

For now, I'm after the 'low hanging fruit,' power generated and consumed ... the energy flow. The data is reasonably smoothed and will meet my interest. After all, it is measured in the field versus Appendix B of the Oak Ridge report, a lab bench test.

I'm also thinking about how to measure wheel torque and rpm via another channel. There are new strain gauges that make this entirely practical and affordable. I could either mount on a drive shaft, the wheel hub, possibly a lug-nut, or even tire strain sensor. Then there are the off-the-shelf solutions. Good, fast, cheap, pick two.

Bob Wilson

[dcb]

Originally Posted by bwilson4web

Patience, there are more than one way to measure losses.

For starters, I'm just trying to understand the source of the measurements and how much processing/inferring has already occured by the time you log them.

Would you mind going through these questions in order? I will assume anything other than "yes" is a long winded "no".

1. Is there an actual torque sensor on mg1? Or is it "inferred" by the ECU and current sensors or???

2. Is there an actual torque sensor on mg2? Or is it "inferred" by the ECU and current sensors or???

3. Is there an actual rpm sensor on mg1? Or is it "inferred" by the ECU and voltage sensors or???

4. Is there an actual rpm sensor on mg2? Or is it "inferred" by the ECU and voltage sensors or???

5. And I assume traction battery voltage and amps comes from actual voltage and current sensors on the battery. But if you could confirm that to please.

Thanks, Dave.

[bwilson4web]

Originally Posted by dcb

For starters, I'm just trying to understand the source of the measurements and how much processing/inferring has already occured by the time you log them.

Would you mind going through these questions in order? . . .

1. Is there an actual torque sensor on mg1? Or is it "inferred" by the ECU and current sensors or???

The maintenance manual does not identify a torque sensor on the shaft nor a sensor in the stator mounts. We just know the hybrid vehicle (HV) ECU reports both torque and rpm in the same MG1 response packet.

2. Is there an actual torque sensor on mg2? Or is it "inferred" by the ECU and current sensors or???

The maintenance manual does not identify a torque sensor on the shaft nor a sensor in the stator mounts. We just know the hybrid vehicle (HV) ECU reports both torque and rpm in the same MG2 response packet.

3. Is there an actual rpm sensor on mg1? Or is it "inferred" by the ECU and voltage sensors or???

The maintenance manual and papers show a position encoder that uses an elliptical metal disk with two, 90 degree coils. In fact, the coils are in a ring. Here is an example, not from a Prius, but similar architecture.

4. Is there an actual rpm sensor on mg2? Or is it "inferred" by the ECU and voltage sensors or???

We know that the motor generator encoder signals go directly to the HV ECU which in turn drives the power electronics in the inverter. These are power, brushless motor generators and the HV ECU handles all of the position sensing and drives the inverter gates that let them work as either motors or generators.

We also know there are Hall Effect current sensors for two of the three legs to the larger, MG2. We suspect these are used to detect unbalanced stator loads that might occur when there is a stator failure.

5. And I assume traction battery voltage and amps comes from actual voltage and current sensors on the battery. But if you could confirm that to please.

Correct. There is a Hall Effect current sensor connected to the dedicated battery controller. Both the battery controller and HV ECU can report traction battery current. However, the battery controller only reports 19, module-pair, voltages, no summary. The HV ECU reports the total voltage. Since the HV ECU pretty well runs the show and needs to modulate the inverter power pulses, it may pull the traction battery voltage from the inverter.

The best way to view the two MGs are as massive, stepper motors under control of the HV ECU that has a separate, bi-directional, 'dumb' inverter. The HV ECU is what makes the Prius drivetrain work with specialized functions distributed to other ECUs such as brakes, air-bag, electric power steering, battery, body, and the immobilizer.

Bob Wilson

[bwilson4web]

I've take the earlier data and used an averaging technique to go from point-to-point per sample to samples with points and the averaged values from earlier and later samples. Not quite a straight-line approximation, it is close enough to make the data more easily understood:


Starting:

The interesting aspects are:

• A - Starting the engine uses MG1 and a slug of power from the traction battery. However, the ~3 second period to make all six samples means in this case, we didn't sample the traction battery power until after the peak MG1 load. This coarse sampling interval means per Nyquist that we need at least seven seconds of steady-state data to even attempt to draw any conclusions.
• B - Shows movement in the parking lot coming to the street. Probably the car was in "N" and the slight spike in battery load comes from the electronic-assisted brake load.
• C - While waiting for traffic to clear, the engine started and put a charge back on the traction battery.
• D - The engine shutoff and the only load is the normal vehicle electrical demand.
• E - During car warm-up, until the catalytic converters get hot enough to work, the car uses EV mode as much as possible. The engine runs but just to heat up the converters.

Acceleration:

The main observations:

• A - To compare MG1 - MG2 efficiency, we need to factor out the traction battery energy and the vehicle overhead load. Then the ratios of MG1 and MG2 can be used for efficiency over the electrical, power path.
• B - Notice that the ratio of mechanical to electrical path load varies significantly. I had once thought it was fixed proportional to the torque ratio but clearly it is not.

Cruse:

Dealing with traffic and overpasses, the data looks pretty noisy but:

• A - This is energy recirculate mode when MG2 acts like a generator and MG1 is a motor adding its power to the ICE and keeping it from spinning faster. The combined power travels over the mechanical path.
• B - Again, the ratio of mechanical and electrical path power is not fixed. This means any fixed assumptions about the electrical path power efficiency and total transaxle efficiency needs to be rethought.

Stopping:

Here we see classical, regenerative braking and sampling and possibly a new, previously undocumented MG1 mode:

• A - In classical regenerative braking, MG2 works as a generator and stuffs energy into the traction battery. However, the traction battery is limited to 20kW so any extra energy has to be handled by the mechanical brakes and/or engine braking.
• B - Here we see a mysterious 4kW with no apparent source. However, close examination of the source data suggests the current sample may have have been missed and/or a brief, no current condition. With a 3 second cycle time for all 6 samples, it is entirely possible a legitimate, no current condition was sampled.
• C - Again, no traction battery power but it is entirely possible that MG1 was just 'all windings tied to ground.' This would turn the motor into one heck of an electronic brake to stop the ICE. But given the heating of MG1, not the best approach. Again, this is probably a sampling issue, not a real operational mode.

Initial testing with a faster polling rate looks to be OK but I haven't recorded any data, yet. Also, converting point-samples into straight-line approximations is the way to go but it is not trivial. I've ordered a GPS receiver that should be here next week. Then I can start a series of benchmarks to accurately measure Prius transaxle energy flows.

Bob Wilson

[dcb]

I ask the questions just because it will be terribly hard to draw any conclusions without knowing precisely what you are measuring, especially regarding efficiency. Are there any sanity checks you can do on the readings? i.e. measure the cars weight/aero/rolling resistance and see if actual acceleration matches predicted based on all the torque readings and number and configuration of loaded gears?

[bwilson4web]

Originally Posted by dcb

I ask the questions just because it will be terribly hard to draw any conclusions without knowing precisely what you are measuring, especially regarding efficiency. Are there any sanity checks you can do on the readings? i.e. measure the cars weight/aero/rolling resistance and see if actual acceleration matches predicted based on all the torque readings and number and configuration of loaded gears?

We know MG2 torque in Newton meters and the gear ratio to the drive wheels. We can measure the tire/wheel radius and calculate the Newtons measured with the car in "D" against a scale or lifting a weight from the trailer hitch. Piece of cake, hold that thought.

In fact, why don't you make a list of things you want calibrated. Will that include the GPS data too? I look forward to a post experiment, calibration exercise . . . just hold on to that 'rope of hope' and I'll get right on it.

In the meanwhile, I've captured a fresh set of data using the higher sampling rate. Thanks to a happy accident, I retraced the same path as the earlier data and remembered I have a straight-line, approximation program to normalize the data.

Bob Wilson

[dcb]

I assume you are doing this in public to build some peer consensus. I am not sure you know precisely what you are measuring if looking at the output from the ECU, and it is critical to understand that if you are trying to measure transaxle efficiency. I have seen a lot of ECUs report bogus or computed data, so there isn't a lot of point in the data analysis without first figuring out what you are looking at, hence why I am asking for particulars about the system in question. I appreciate it is a bother to double check what the ECU says but there is good reason for it.

For example you did not mention current sensors on mg1, so is the ECU inferring the current there? and if so, it is not possible to use that reading to determine the efficiency of anything relating to mg1.

Also since it appears to use current to derive torque, we don't know if it is "electrical" torque, or shaft torque after converting for losses, or how accurate the current to torque conversion reading is. Could be a 30% discrepancy there depending how the ECU programmer interpreted the "requirements" if any.

Also, did you mention publishing data? Assuming the data is correct (or even if it isnt), it would be simpler to process things like "positive area under the blue line vs negative" with the actual values.

But I do need to understand how things like what the ECU reports as torque relate to the real world before I will accept ECU funged data. Just like I wouldn't accept airflow as an approximation for fuel flow where precision is concerned. I have no other purpose here accept to ensure the claims are valid, and taking measuring shortcuts is rightfully going to draw some criticisms. I'm glad you are willing to do it, but I will pressure you to do it "right" before "publishing" any conclusions.

[bwilson4web]

Originally Posted by dcb

I assume you are doing this in public . . .

I'm doing it because I want to know how my car works. Sharing doesn't change the data but if it bothers you, don't read the thread.

Bob Wilson