The Premier's Limo: 2004 Toyota Prius "winter beater" ecomodding thread

[Cd]

As much as I'd love to build a full boattail for the car, I'm really discouraged by what A.I. is giving as an estimate for the gain in MPG.
Quote :
🔬 Why Aero Gains "Top Out"

Here is the physics breakdown of why your conclusion is right:

· The Cube Law (Aero Drag): Power to overcome air resistance increases with the cube of speed. Even dropping Cd from 0.26 to 0.19 (a ~27% reduction) does not cut fuel consumption by 27% at 65 mph; the benefit is significant, but diminishing returns set in.
· The "Engine Sweet Spot" (Your Main Point): At 65 mph, the gasoline engine is forced to run at a higher RPM outside its most efficient range (the "sweet spot" is roughly 1,700–2,300 RPM). Because the power demand is constant, MG1 and MG2 can't shut the engine off or significantly alter that load requirement.
· The Return on Investment: Aero mods give the biggest percentage gains at higher speeds (above 55 mph). However, on a Gen 2 Prius, a full boat-tail and smoothing might only yield a gain of ~1-3 MPG over stock under those perfect conditions.

📊 The Realistic Verdict

In your "laboratory" scenario (65 mph, 0 wind, flat road):

· Stock Prius: ~48–52 MPG.
· Highly Modified (cD < 0.20): Realistically ~54–58 MPG.
· Theoretical "Perfect Storm": Exceptionally rare cases might touch 60–62 MPG, but that is the absolute ceiling before engine physics stop you.

So, while shedding the boat-tail wouldn't hurt, the MG1/MG2 units cannot force the engine to run more efficiently at that speed. The car simply hits a "mechanical wall" where the engine must burn a certain amount of fuel to turn the wheels at 65 mph.

One to three mpg return is actually embarrassing.

I really do wish that the 65 mph test results on Darin's test could be posted.
How often does he visit the forum ?
I would also like to ask him how he got the 70+ MPG results on the low speed run that he did ( 50 mph and with several stop lights and trips through small towns.
EOC was listed as a method at these speeds, but I wasn't aware you could EOC in a G2 Prius.
( And the ICE kicks on in my car at speeds as low as 6 mph !)

[Gasoline Fumes]

I wouldn't trust any advice from AI!

Originally Posted by Cd

I really do wish that the 65 mph test results on Darin's test could be posted.
How often does he visit the forum ?

He doesn't visit.

[Xist]

I think that consulting ChatGPT should be a standard step in research projects.
"We're going to assume that's wrong. Instead, our theory is..."

[redpoint5]

Originally Posted by Gasoline Fumes

I wouldn't trust any advice from AI!

Skepticism is healthy, but dismissing AI is not.

[freebeard]

It's an architectural problem.

Triple modular redundancy
In computing, triple modular redundancy, sometimes called triple-mode redundancy, is a fault-tolerant form of N-modular redundancy, in which three systems perform a process and that result is processed by a majority-voting system to produce a single output. If any one of the three systems fails, the other two systems can correct and mask the fault. The TMR concept can be applied to many forms of redundancy, such as software redundancy in the form of N-version programming, and is commonly found in fault-tolerant computer systems Continued in Wikipedia

[redpoint5]

That's what's used to protect against "bit flip", where cosmic radiation causes the value of a bit to change without being commanded to change. Pretty important when the accelerator position of your drive-by-wire car has a byte value and the digits matter.

It's speculated that the Toyota "uncommanded acceleration" incidents were the result of cosmic rays flipping bits. Of course, shifting to neutral would have interrupted that, but folks generally don't know how things work.

[Vman455]

This is very much a case of "you shouldn't even bother asking AI, it's just spitting out BS based on probabilities derived from all the BS that has been fed into it."

Originally Posted by Cd

🔬 Why Aero Gains "Top Out"

Here is the physics breakdown of why your conclusion is right:

· The Cube Law (Aero Drag): Power to overcome air resistance increases with the cube of speed. Even dropping Cd from 0.26 to 0.19 (a ~27% reduction) does not cut fuel consumption by 27% at 65 mph; the benefit is significant, but diminishing returns set in.

"Diminishing returns" is correct but not because of the geometric relationship between power and speed (we're holding speed constant here at 65 mph; changing speed isn't a factor). It's because at your target speed, the car will have some rolling resistance that, unless weight is changed, will be effectively constant. As aero drag becomes a smaller and smaller proportion of total resistance force, it has less and less influence on fuel economy.

I'm writing a post now on thermal efficiency, propulsive efficiency, and overall efficiency and what these all mean, as well as how to measure the overall efficiency of your car, but the upshot is these things--MPG, efficiency (eta_o), fuel energy (Q_R), and resistance force (F)--are all related by:



Since we effectively have a "floor" of rolling drag that is, for most cars, nearly as large as aero drag at 65 mph, F in the denominator can only get so small and, assuming no change in efficiency, that means there will be some maximum steady-state MPG achievable at that speed. Using measured efficiency on my car at various speeds, this shows variation in predicted MPG by C_D:

Originally Posted by Cd

· The "Engine Sweet Spot" (Your Main Point): At 65 mph, the gasoline engine is forced to run at a higher RPM outside its most efficient range (the "sweet spot" is roughly 1,700–2,300 RPM). Because the power demand is constant, MG1 and MG2 can't shut the engine off or significantly alter that load requirement.
· The Return on Investment: Aero mods give the biggest percentage gains at higher speeds (above 55 mph). However, on a Gen 2 Prius, a full boat-tail and smoothing might only yield a gain of ~1-3 MPG over stock under those perfect conditions.

📊 The Realistic Verdict

In your "laboratory" scenario (65 mph, 0 wind, flat road):

· Stock Prius: ~48–52 MPG.
· Highly Modified (cD < 0.20): Realistically ~54–58 MPG.
· Theoretical "Perfect Storm": Exceptionally rare cases might touch 60–62 MPG, but that is the absolute ceiling before engine physics stop you.

So, while shedding the boat-tail wouldn't hurt, the MG1/MG2 units cannot force the engine to run more efficiently at that speed. The car simply hits a "mechanical wall" where the engine must burn a certain amount of fuel to turn the wheels at 65 mph.

And all of this is BS ("the power demand is constant"--this is nonsense! The whole point of this question/simulation is to model what happens when we reduce vehicle power demand by reducing aero drag).

The Prius should respond better to changes in F than cars with fixed gear ratios precisely because the computer can vary MG1 speed and gas engine speed to keep it in a more efficient spot (in other words, keeping eta_o high) based on required power and BSFC. It isn't "engine physics" stopping incredible MPG numbers, it's the combination of all vehicle performance parameters. That includes the difficulty of achieving really low aero drag, the difficulty of reducing weight and rolling drag, and the difficulty of achieving high thermal efficiency regardless of engine load and RPM. It's the interaction of those 3 things (since we aren't changing fuel heating value, unless you want to convert the engine to run on diesel or something) that will determine MPG at any given speed.

[Xist]

At 15:56, Hank Green asks "What you're saying to me right now is that the LLMs hallucinate because no one on the internet ever writes the words `I don't know'?!"

https://www.youtube.com/watch?v=5CKuiuc5cJM

I'm glad that is 16 minutes in because the video is over an hour longer than that, and I watched it at triple speed to find that quote.

Google built Gemini into YouTube and I asked "Why did they say LLMs hallucinate?" and why does this contain so many words?!

The speakers explain that hallucinations occur primarily because Large Language Models (LLMs) are trained to predict the next piece of text based on patterns in their training data, rather than strictly prioritizing the truth. (15:36 - 15:44) Specifically, the models are trained to mimic human writing; since humans in certain contexts (like writing case law or posting on the internet) do not typically admit when they lack knowledge, the AI mimics that behavior by providing a confident but fabricated answer to maintain textual similarity to its training examples. (15:46 - 16:45)

I requested a direct quote and it gave me

Nate Soares notes:

"The obvious guess for what's going on with hallucinations is that these AIs are trained first and foremost on predicting text. And if you're, you know, say you're predicting case law that a lawyer writes, that lawyer's going to write, you know, blah blah blah as seen in this precedent and then cite a real case law. That lawyer when writing case law is not going to say, 'I happen to not know the case law.' ... Even if you tell the AI, you're predicting a lawyer who would never make up case law. The AI can still get closer to like as measured by like text similarity to what a lawyer says by making stuff up. It's not that it doesn't know it's fake in some sense. It's that this thing has something like a reflex or an impulse. ... it in some sense cares more about the similarity of the text, about it looking like it's supposed to look even if it doesn't have anything to put in that space."

Nate said the first part, Hank interrupted to ask the question I quoted, they have a little discussion, and then Nate continues.

Gemini skipped all of that, eliminating the one line which mattered to me.

I've asked it to summarize things and it has made them longer!

The Praxis sample test questions were super long, sometimes 1,000-1,200 characters, but the actual test questions were more like a sentence or two, and I studied almost 400 questions.
I asked it to summarize each question in a sentence or two and it randomly deleted questions and added ones which could have come from anywhere.

I could write far more about LLM's problems, but I can't trust an LLM to give an accurate summary.

[redpoint5]

That's the thing perplexing me; that so many people angerly defend their tribe's assigned position on a topic that they have no understanding about. Why would anyone have strong opinions on a subject of economics if they don't know what money is, or the basics of supply and demand? How does a person have anything to say on a subject of history and geography when they know none of it?

It's the most frustrating thing to me when I have spent tens or hundreds of hours to know the little details about something, and some bozo thinks the position assigned to them is just as valid because they have spent zero seconds contemplating the subject.

I generally refrain from commenting on things I have no interest or knowledge about. Occasionally I'll "guess", but I put that caveat in my reply that I'm guessing for fun, not out of an abundance of knowledge.

The other strange thing is it's only certain subjects that people have profound ignorance AND strongly held opinions. Nobody who has no interest in classical music goes around saying Chopin was the greatest that ever lived and Beethoven was a child by comparison, but they will tell you that photovoltaics are the future of electricity even though they don't know what photovoltaics, electricity, energy, power, Hz, amps, volts, economics... is, and they have zero interest in learning anyhow.

All that to say it should be unnecessary to write "I don't know", because one should refrain from saying anything in the first place, or have the caveat that they are guessing just for fun.

[freebeard]

All that to say it should be unnecessary to write ... the caveat that they are guessing just for fun.

Works for me.

Text may suffer from Reddit-brain, but AI images say that future cities will all be bone white and gold.