Gill fuel flow sensor failure modes

[zcar]

Yes. Anyone who says they "know" the true value of a mass flow to a high precision by measuring it indirectly has never tried to repeat the measurements themselves. RB upper management may think they know, and they don't. I suspect RB engineers with more sensor experience are a bit red faced at their team's press releases.

I took the RB releases to mean that they have much confidence in the repeatability of their mass flow model and observed the shift against the mass flow sensor, i.e. “We had a fuel-flow sensor that was fitted to the car that we believed to be in error, and therefore based our calculation on the fuel that the injectors were providing to the engine, which is a calibrated piece of equipment that is consistent and standard across the weekend that we've seen zero variance in." Given a stable timebase (in the ECU) and repeatable injector motions (according to his claim) one would hope for consistent fuel flow against the model, otherwise you would see variance in other engine parameters. Granted, this estimation of mass flow depends on the some more measurements just as the velocity measurement of the ultrasonic. I understand there are many things at work at the fuel injector itself...shift time, fuel inertia, fuel viscosity, pressure differential across the nozzle, maybe a dozen more?

To be sure, this probably puts the instrumentation guys on the spot...

[Dragonfly]

Even every day desktop ink jet printers determine the amount of ink spent by counting the drops they deliver. I don't see why at the high end of engine management using injected fuel data should be less accurate than data from an external to the system (and volatile) flow meter.

[erikejw]

What striked me as humiliating to FIA and the GILL company is that they have no performance measurements of a sufficient standard deviation when it comes to bursts of 0.2 seconds of hard acceleration and fuel from going from none to full flow. That is what RB broke against according to the faulty sensor which is god in this case.

They have performance measurements under a steady state flow, guess what, steady state have nothing to do with F1.
A steady state flow, like a calm river flowing slowly compared to the extreme bursts in F1 during heavy acceleration.

I am sure GILL can deliver a sensor that can measure fuel flow for a full race but not reliably when it comes to fractions of a second.

I guess the GILL lawyers prepares for lawsuits in this very moment.
It is sad for everyone, GILL, F1, FIA and especially the fans that F1 is dominated by that crazy, nonreliable sensor.

The preparation have been years and they cannot handle it.
My passion for F1 and longing for a new fantastic season has been crushed.
Shame on everyone who have humiliated F1 this weekend.

[autogyro]

Looks to me that RB has used the unreliability of the fuel flow sensor to gain a power increase in an attempt to address their problems of having less power and less control over that power than other teams and power unit suppliers.
I suppose it was worth a try Adrian.
Looks like all those little tricks using aero are not the same when applied to the better defined area of fuel use.

[hardingfv32]

Looks to me that RB has used the unreliability of the fuel flow sensor to gain a power increase in an attempt to address their problems of having less power and less control over that power than other teams and power unit suppliers.

This is false. FIA mandated the use of the original 'unreliability of the fuel flow sensor' before the race.

Brian

[Sombrero]

New formula : first race, first controversy with the spiLL sensors...

[Dragonfly]

About sensor calibration. I suppose the unit has its own MCU or FPGA device which processes the raw data and the factory calibration values are written in the firmware.

[WhiteBlue]

About sensor calibration. I suppose the unit has its own MCU or FPGA device which processes the raw data and the factory calibration values are written in the firmware.

You can safely make such an assumption. The sensor is analogue and the output digital. So a digital processor of some type is required without doubt for many tasks. FPGA or ASIC would also be my guess.

[bill shoe]

Yes. Anyone who says they "know" the true value of a mass flow to a high precision by measuring it indirectly has never tried to repeat the measurements themselves. RB upper management may think they know, and they don't. I suspect RB engineers with more sensor experience are a bit red faced at their team's press releases.

As I've mentioned before I'm quite sure it can be done. I must admit I cannot give the technical details of how it works, but I can assure you that I've worked with the results. On off road vehicles fule consumption is measured in liters/hour. I used to be one of the developpers of controllers on (among others) this machine. Part of the datamining to the driver consisted of fuel consumption. I know for sure that from the measurements I've done the integration error was less than 0.5liters per tank (over a 1000 liters). If I remember well that was done without a fuel flow meter. Keep in mind this is not an F1 car and technology is a lot cheaper. I'm convinced that with a flow meter that kind of accuracy would not have been reached.

Thanks for interesting response. I'm initially skeptical of the accuracy you mention, but I have to respect your hands-on experience.

Let me take a half-step back. If someone is certain about the accuracy of some absolute measurement, then they must have an even-more-accurate reference to compare their measurement to. A scale can be called accurate to within x% if you calibrate it with various reference weights that have weight/mass known to within something less than x%.

You mention 0.5L relative to 1000L, or 1 part in 2000. What was the nature of the even-more-accurate reference for this test? I've done simple fuel economy testing (perhaps nowhere near as involved as you) and I think it would be difficult in the real world to somehow measure the volume of a liquid reference to less than 1 part in 2000 (0.05%). Especially when the liquid is some kind of diesel fuel where tiny bits of evaporation or temperature-change will cause it to change volume more than 1 part in 2000. And when the volume is something big like 1000 L! Am I assuming you meant accuracy when you meant precision? Correct any of my assumptions if I'm getting them wrong.

I think these issues relate to the current controversy in F1. If RB is certain their software fuel-model is more accurate in an absolute sense than the FIA FFM, then RB must have an even-more-accurate reference of some kind that allows them to judge their fuel-model. And this even-more-accurate reference somehow works onboard their F1 car in its typical operating conditions. Given the limited/difficult running of RB's car and Renault's engine, it seems amazing they have somehow implemented this remarkable reference tool.

This is not verbal slight of hand. These are the "how do we ever know the true value" kind of questions that come up in real world testing where you seek an absolute value to within 1% or less. Most test people who go down this rabbit hole come back with much stronger skepticism for absolute certainty than the people who want to apply the results.

[DaveW]

If someone is certain about the accuracy of some absolute measurement, then they must have an even-more-accurate reference to compare their measurement to. A scale can be called accurate to within x% if you calibrate it with various reference weights that have weight/mass known to within something less than x%.

The quote above was taken from a memorable post that contains my all time favorite metaphor.

Sadly, I can't quite agree with the statement above. Man has long applied insight and ingenuity to measurement (starting, according to some, with Archimedes) to improve measurement accuracy. For example, it is now possible with very simple commercial equipment to place an oil rig in the middle an ocean with millimetric accuracy.

I imagine that controlling a current F1 power train would require significant modelling expertise to make the thing work efficiently. I suggested earlier that some form of Kalman filtering might be involved. if so, I have no doubt that energy transfer would be important, and fuel input would be one of those that would receive much attention. I would guess the engine people could predict fuel energy input very accurately without measuring fuel flow at all.

I agree with the metaphor, though.

[Jef Patat]

Sorry guys, maybe I was a bit unclear in my posting and caused more confusion than I intended. I was only talking about the integration error, because that was the part that concerned my part of the job. I forgot to mention the details of the ECU. It has been too long to remember the details that clear but it was something along these lines. For legal reasons the company could not display fuel consumption if it didn't meet certain criteria, and that only in certain countries. I believe it was similar but not the same as on fuel pumps. Now I'm talking about accuracy. (If I remember correctly it needed to be smaller than 0.5% in most countries on the pumps). For all clarity that kind of testing has not been done by me. That was between our engineering lab and the engine manufacturer. I don't remember anything about temperature compensation, probably there was, I know for sure there is a fuel temperature sensor. I only remember it had to be a similar measurement to fuel pumps. So the accuracy of the engine was along the lines of 0.5%. If I remember well there were two values from the ECU: total fuel consumption (in liters) with the guaranteed accuracy and another 'instant' consumption. To be able to detect peak usage I could not use the total consumption and had to use the instant value. The ECU was broadcasting its instant consumption several times per second on CAN. The measurements I did was checking for the integration error of this 'instant' value to the total (with guaranteed accuracy) to be sure things were still legal after this operation. Homologation probably did those measurements again, I've never heard anything about it so I guess it was ok.

Again, I do not know what model was use to do the measurement. But I hope it is clear that the model used was accurate to something like 0.5%. It has been 8 or 9 years by now, so I might have forgotten about some details.

That brought me to the idea i posted in the second post of this topic. I'm quite sure the teams have have a better and faster estimation of the fuel consumption than the Gill sensor. I wouldn't be surprised if they try to trick it with aliasing, but that is pure speculation.

hope this clarified a bit

[alexx_88]

Great insight Jef!

If I understand correctly, you were gathering the data that was broadcast over CAN. That tells me that the data was calculated by the ECU of the engine. As previous posters mentioned, that value is calculated using the size of the injectors, the pressure and temperature of the fuel and the width of the pulse that controls each injector. Basically, you know the volume of fuel that the injector can put in (let's say 400cc/min), you know for how long you are keeping it open so it's simple maths to get a ballpark figure.

Because the error of this is both positive and negative, I'd expect that, over a long enough period of time, integrating these values would yield a pretty good accuracy, as you've said. However, when trying to accurately measure flow over a much shorter period of time, there is a much greater chance of a bad reading to influence the overall result.

[ppj13]

Sure you can estimate your semi-instantaneous fuel use just with the injection time plus fuel pressure plus fuel temp. Sure you don't make a big mistake. 0.5% is a bit optimistic to me. 1% probably. 2% for sure. I told you, every rally team does that. I think MOTOGP also and everywhere in the motosport. You need to set precisely how much fuel you are using, because there are many hp in that fine tunning. At every stroke.

The gill sensor has another big drawback, and this is lag. If the measure has whatever bigger than 0.2s lag, or variable lag, (both things being highly possible) then it's useless, specially if the team is "short-shifting" below 10500 (to seek for better efficiency or just by mistake). How do you correlate the actual fuel flow with the limit, which is rpm-dependent, at any given hundreth of second if the GILL thing is not reacting quick enough?

I remind you that when you upshift the whole drivetrain twists back and forth 3 or four times and the instantaneous rpm measurement is absolutely all over the place (plus and minus 1000 rpm is not strange) for the next 0.2 or 0.3 seconds.

I say again, only way out is to ignore the exact words of the rules. The allowed fuel flow is the one the GILL sensor say it's legal at every track, at every car, at every moment; regardless of how little that has to do with real fuel flow. Teams, adapt yourselves to that, at the end of the day it's the same for everyone.

[bill shoe]

Again, I do not know what model was use to do the measurement. But I hope it is clear that the model used was accurate to something like 0.5%. It has been 8 or 9 years by now, so I might have forgotten about some details.

That brought me to the idea i posted in the second post of this topic. I'm quite sure the teams have have a better and faster estimation of the fuel consumption than the Gill sensor. I wouldn't be surprised if they try to trick it with aliasing, but that is pure speculation.

hope this clarified a bit

Thanks for insight. And to Dave W for giving me the analogy award!

Your comment about tricking the sensor by intentionally aliasing the data- yea it occurred to me too, but I think the FIA takes in relatively raw data at a higher rate and then down-filters by averaging 0.2 second snippets. So they can check the higher freq data for tricks. I think this means people had that idea before me.

In lower and mid series, it's not uncommon to cheat rpm-limiters and spec-ECU's, so I would think the FFM is a new opportunity for illegal workarounds. The most basic way to cheat a fuel limit is to not run all your fuel through the sensor...

[chip engineer]

...
The gill sensor has another big drawback, and this is lag. If the measure has whatever bigger than 0.2s lag, or variable lag, (both things being highly possible) then it's useless, specially if the team is "short-shifting" below 10500 (to seek for better efficiency or just by mistake). How do you correlate the actual fuel flow with the limit, which is rpm-dependent, at any given hundreth of second if the GILL thing is not reacting quick enough?

I remind you that when you upshift the whole drivetrain twists back and forth 3 or four times and the instantaneous rpm measurement is absolutely all over the place (plus and minus 1000 rpm is not strange) for the next 0.2 or 0.3 seconds.
...

If they are really getting measurements at a high rate (say 100Hz) and averaging to get 5Hz data, lag should be small and not variable. If rpm is averaged the same way fuel flow is, I do not see why there should be a problem. Even if shifts caused some error below 10500 rpm, I doubt anyone would do extra shifts for a few % more fuel for 0.2 seconds.