Ferrari Power Unit Hardware & Software

[seventhsin]

Could it be extra acceleration be as simple as deploying in the middle of the long straights, hence "reaching top speed" earlier, but then harvesting from the H at the end of the same long straight, even when at full throttle?
Say you tell the system that the H is to add 50 hp to the ES every time you are above 300km/h (and maybe say unless the third paddle is pressed?). You are effectively robbing the engine of power near top speed, but a 5% reduction in power near top speed is only a change from, say, 320km/h terminal speed to 314.4km/h.
Surely the time gained reaching 300 outweights the losses for the while over 300, while the extra harvest can be put to good use in deploying for longer in future straights?

Hmmm... not 100% sure this is making sense, I might have confused myself there.

That would certainly seem to be a useful approach.

The most likely extra power would come from extending the period that electric-supercharger mode is deployed. If,say, they could deploy from 250kph to 300 and others couldn’t it might require about 250kJ extra from the ES. From 300 on the 120kW minus whatever the MGU-H is making, then when you come to harvest it also relies on what the H can make.

So if you can make more from the H you can deploy the electric supercharger mode longer, get to speed quicker and sacrifice a little top speed to recharge the ES.

As an example if you can make 80kW from the H instead of 60kW. After 300 kph you could run 4 seconds at self sustain plus then 2 at ICE only, harvesting from the H. You would get an extra 120kJ plus you’d have higher part throttle charging and, perhaps, higher blowdown and so lower discharge rate during electric-supercharge. So not far from being able to deploy from 250 to 300.

Would it not be the most benefit to have the highest speed for longest, as it covers the most ground is shortest time?
Losing a second last thing before brake would loose more than a second while exiting the previous corner (or not?)

Drag squares with speed, so traditionally the most efficient way to decrease lap times is use your available energy in the slow parts of the track to get you up to speed as quickly as possible.
The higher your speed, the energy requirements to go faster increase exponentially.

However, is this still the best way to decrease lap time in F1's current technology and regulations? I don't know, that's why I listen and read twice as much as much as I speak. Advice I'd encourage a certain member here to take up.


Back on topic, did anyone find the specific regulation regarding each car having a 'single ES'?

Sent from my SM-G950F using Tapatalk

[henry]

...

Would it not be the most benefit to have the highest speed for longest, as it covers the most ground is shortest time?
Losing a second last thing before brake would loose more than a second while exiting the previous corner (or not?)

At 320kph you travel 89 m in 1 sec.

At 315kph it takes 1.016 sec to cover that same distance.

So a bit slower at the end of the straight doesn’t have a huge effect.

[Dr. Acula]

...
Would it not be the most benefit to have the highest speed for longest, as it covers the most ground is shortest time?
Losing a second last thing before brake would loose more than a second while exiting the previous corner (or not?)

Not necessarily. What you want to reach is the highest average speed over the whole straight. That doesn't mean you have to have the highest top speed at the end of it.
Also, with increasing speed you need more an more energy to increase the speed even more because the formulas for aerodynamic drag and kinetic energy are square functions which means to go twice as fast you need 4 times the energy. In a energy limited application, what F1 is since the introduction of the fuel limit, it can be hugely benefical to just accelerate as fast as possible to a relativ high speed and then just cruise along with that speed. Toyota demonstrated this wonderfully in LeMans this year. The top speeds of the private LMP1 cars were slightly higher, but they didn't had a chance on the Hunaudieres straight regardless.

[subcritical71]

Drag squares with speed, so traditionally the most efficient way to decrease lap times is use your available energy in the slow parts of the track to get you up to speed as quickly as possible.
The higher your speed, the energy requirements to go faster increase exponentially.

However, is this still the best way to decrease lap time in F1's current technology and regulations? I don't know, that's why I listen and read twice as much as much as I speak. Advice I'd encourage a certain member here to take up.


Back on topic, did anyone find the specific regulation regarding each car having a 'single ES'?

Sent from my SM-G950F using Tapatalk

I do not think the rules prohibit multiple ES, but I would think that the weight limit imposed would be for all ES and not just one.

5.4.3 The total weight of the part of the ES that stores energy, i.e. the cells (including any clamping plates) and electrical connections between cells, must be no less than 20kg and must not exceed 25kg.

[henry]

The ES and all the control electronics have to fit in a defined volume and be contained in the safety cell. So there’s no real advantage to making multiple components.

[Muniix]

The FIA has said that the Ferrari system is very complicated and considering that the power boost is only coming to some teams after a software upgrade, the mind immediately considers AI.

A theoretical software map might include machine learning software that can turn hardware (or other software) switches on and off depending on circumstance - that way, alternate functions of components can be concealed. Such an approach would confound the devil out anyone outside the implementers of the software and hardware. Machine learning software on its own is very complicated, talk less a system that might be part software, part electronic hardware, part mechanical/hydraulic/pneumatic.

Such a system might be able to learn and react to sensor readings as well as other parameters to know when to switch modes, so that the FIA (and you or I) would never be the wiser. For instance, if all the FIA has to monitor compliance are two DC sensors, then they might be beaten if circuits are made to behave in non-obvious ways, or their functionality changes completely.

You don't have that sort of compute power available on the car, not even close.

It doesn't take much computational power to run machine learning (little devices e.g. phones, camera's come loaded with it these days). The only aspect that requires computational power is the building the model, which wouldn't be done in the car, but by code writers in an office.

It DOES take huge amounts of compute to do ML/DL the phones you refer to have special TPU tensor processing units, the Huawei Mate 10 Pro, uses their own Kirin 970 that uses the GPU graphics processing units and the dedicated TPU cores for AI workloads. This Kirin 970 SoC is the same one I use for powertrain control atm, to do AI and physics compute loads.

The standard hardware the Formula One teams are forced to use have less than 1% of the compute of average mobile phones.

Ferrari may be putting in some intelligence into the TJI in engine, in that it's controlling it's own spark timing based on cylinder and pre-chamber MEMS pressure sensors, maybe ion sensing, and using some models with thermodynamics and detecting the reaction rate of radicals as they can be sensed this way with TJI and placing a 10mm spark plug just for ION sensing in main chamber, Ferrari could be a little bit more elegant in implementation.

[Muniix]

If MR Moderator wants to stop this discussion it is of course his prerogative, but if I am permitted, I was under the impression that the gentleman I answered was saying “that for example after a corner and on the straight they deploy their 4MJ, and before the next straight they would for example harvested back another 4MJ, and on the next straight they will again deploy those 4 MJ". Can anybody tell me where would those harvested back 4MJ would have been stored if not in the ES? And so, how many 4MJ per lap is one allowed to deploy from ES? We all know that deployment id only possible through K.

It's a 4MJ window the ES SOC has to be operated in "at any time on track" as long as you stay within this rule your ok as long as you don't violate any other rules, rules are inclusive, not exclusive, i.e. just because your following one doens't allow you to break another.

[subcritical71]

Along the same lines as AI/processing, I remember hearing that the MotoGP factory teams have figured out a way around traction control limits by conditioning the signal between the sensors and ECU. That way, the FIM mandated ECU gets a signal which has already taken into account everything necessary to limit wheel-spin. Its a nice workaround which goes against the spirit, but not letter of the rules. And sense these control units are manufacturer specific its hard for the FIM to police.
With the F1 manufacturers using their own CU for the H, K, ES, ICE(?). Do we think that there could be some meaningful workarounds in this regard. I can already think that the K is being used as a sort of traction control system as the motor could be controlled in a way to not exceed a set motor acceleration for a given speed.

[PhillipM]

If MR Moderator wants to stop this discussion it is of course his prerogative, but if I am permitted, I was under the impression that the gentleman I answered was saying “that for example after a corner and on the straight they deploy their 4MJ, and before the next straight they would for example harvested back another 4MJ, and on the next straight they will again deploy those 4 MJ". Can anybody tell me where would those harvested back 4MJ would have been stored if not in the ES? And so, how many 4MJ per lap is one allowed to deploy from ES? We all know that deployment id only possible through K.

So you just send it through the H to the K, doesn't matter, the way the rules are written you can deploy as much as you have at 120kw.

The trick is that Ferrari have suddenly gained either 30bhp in the ICE or 30bhp through the K/H somehow, which is why other teams are questioning it. It's not that they're deploying longer, it's that they're deploying more power for short periods as soon as they get out of the traction zones.

[saviour stivala]

Could you please point me in the right direction where the regulations state this limit, please? Note that we're not talking ERS->MGU-K deployment here, but total deployment of ERS.

“ lets assume a regular circuit with corners and straights. We exit a corner with fully charged ERS (it is the ES which can be fully charged or not) and deploy the full 4MJ while accelerating on the straight. While on the straight, we’re able to harvest 4MJ again and charge the ERS (ES), now if we exit the next corner, we’re again able to fully deploy 4MJ from the ERS without breaking the rules”. Deployment is only possible by ERS-K, It is the ES that is charged, so it means that your deployment to the ERS-K through which deployment is only permitted can only be 4MJ per lap.

Please refrain from posting incorrect or misleading information. This will only confuse others that come to this forum and don’t realize the source.
[/quote
That in my opinion is unfair/unjust and uncalled for accusation. But that’s just my opinion, anyhow, I believe that I have read that accusation somewhere else before, I believe it originated from the US OF A, am I correct SUB?.
Also from on your other post, you know pretty well that we were not talking about the free load mode subject this time, so why push at that?.

[saviour stivala]

If MR Moderator wants to stop this discussion it is of course his prerogative, but if I am permitted, I was under the impression that the gentleman I answered was saying “that for example after a corner and on the straight they deploy their 4MJ, and before the next straight they would for example harvested back another 4MJ, and on the next straight they will again deploy those 4 MJ". Can anybody tell me where would those harvested back 4MJ would have been stored if not in the ES? And so, how many 4MJ per lap is one allowed to deploy from ES? We all know that deployment id only possible through K.

So you just send it through the H to the K, doesn't matter, the way the rules are written you can deploy as much as you have at 120kw.

The trick is that Ferrari have suddenly gained either 30bhp in the ICE or 30bhp through the K/H somehow, which is why other teams are questioning it. It's not that they're deploying longer, it's that they're deploying more power for short periods as soon as they get out of the traction zones.

the point was were do you store it when harvested back and till you send it out (deploy it out on the next straight).

that FERRARI gained some BHP, i do not know the numbers, i have no doubt.

[subcritical71]

“ lets assume a regular circuit with corners and straights. We exit a corner with fully charged ERS (it is the ES which can be fully charged or not) and deploy the full 4MJ while accelerating on the straight. While on the straight, we’re able to harvest 4MJ again and charge the ERS (ES), now if we exit the next corner, we’re again able to fully deploy 4MJ from the ERS without breaking the rules”. Deployment is only possible by ERS-K, It is the ES that is charged, so it means that your deployment to the ERS-K through which deployment is only permitted can only be 4MJ per lap.

Please refrain from posting incorrect or misleading information. This will only confuse others that come to this forum and don’t realize the source.

That in my opinion is unfair/unjust and uncalled for accusation. But that’s just my opinion, anyhow, I believe that I have read that accusation somewhere else before, I believe it originated from the US OF A, am I correct SUB?.
Also from on your other post, you know pretty well that we were not talking about the free load mode subject this time, so why push at that?.

USA... sub... hmm... are you going to guess my birth year next? (hint, I'm 47)

[GrandAxe]

You don't have that sort of compute power available on the car, not even close.

It doesn't take much computational power to run machine learning (little devices e.g. phones, camera's come loaded with it these days). The only aspect that requires computational power is the building the model, which wouldn't be done in the car, but by code writers in an office.

It DOES take huge amounts of compute to do ML/DL the phones you refer to have special TPU tensor processing units, the Huawei Mate 10 Pro, uses their own Kirin 970 that uses the GPU graphics processing units and the dedicated TPU cores for AI workloads. This Kirin 970 SoC is the same one I use for powertrain control atm, to do AI and physics compute loads.

The standard hardware the Formula One teams are forced to use have less than 1% of the compute of average mobile phones.

Ferrari may be putting in some intelligence into the TJI in engine, in that it's controlling it's own spark timing based on cylinder and pre-chamber MEMS pressure sensors, maybe ion sensing, and using some models with thermodynamics and detecting the reaction rate of radicals as they can be sensed this way with TJI and placing a 10mm spark plug just for ION sensing in main chamber, Ferrari could be a little bit more elegant in implementation.

1% the compute average of mobiles phones is pretty handy where the the dataset being processed is small and has few dimensions (unlike, say millions of dimensions in a jpeg or video) as will be the case with turning switches on or off based on a narrow set of parameters in an F1 car.

Machine learning algorithms are quite varied - from those that basically require more storage space and RAM than computational power and can run on bog standard devices (e.g. pos taggers); to systems that require super computers (e.g. weather forecasting and gene sequencing). Not everything requires GPU and other specialist hardware with frightening names.

[Muniix]

It doesn't take much computational power to run machine learning (little devices e.g. phones, camera's come loaded with it these days). The only aspect that requires computational power is the building the model, which wouldn't be done in the car, but by code writers in an office.

It DOES take huge amounts of compute to do ML/DL the phones you refer to have special TPU tensor processing units, the Huawei Mate 10 Pro, uses their own Kirin 970 that uses the GPU graphics processing units and the dedicated TPU cores for AI workloads. This Kirin 970 SoC is the same one I use for powertrain control atm, to do AI and physics compute loads.

The standard hardware the Formula One teams are forced to use have less than 1% of the compute of average mobile phones.

Ferrari may be putting in some intelligence into the TJI in engine, in that it's controlling it's own spark timing based on cylinder and pre-chamber MEMS pressure sensors, maybe ion sensing, and using some models with thermodynamics and detecting the reaction rate of radicals as they can be sensed this way with TJI and placing a 10mm spark plug just for ION sensing in main chamber, Ferrari could be a little bit more elegant in implementation.

1% the compute average of mobiles phones is pretty handy where the the dataset being processed is small and has few dimensions (unlike, say millions of dimensions in a jpeg or video) as will be the case with turning switches on or off based on a narrow set of parameters in an F1 car.

Machine learning algorithms are quite varied - from those that basically require more storage space and RAM than computational power and can run on bog standard devices (e.g. pos taggers); to systems that require super computers (e.g. weather forecasting and gene sequencing). Not everything requires GPU and other specialist hardware with frightening names.

Your reading sensor data, and processing it, it's not on/off single bit digital data, it's potentially more complex than recognising 'cats'.

[Tzk]

the point was were do you store it when harvested back and till you send it out (deploy it out on the next straight).

As pointed out several times, you just harvest and deploy from/to the ES but to circumvent the 4mj Limit, you do it via for example via the mgu-h.

This has nothing to do with your proposed "free load mode". Which is basically motoring the turbo via the mgu-h with open wastegate(s).