RomRaider

Thanks for the post!

no problems, I will update when I log with the voltage the ecu sees. I'm not really sure how much the volt-meter would skew the results and am curious to see now...

I am using the AEM IAT sensor, it should be the same sensor as the GM Open element sensor. I am tuning the car right now with Cobb's speed density and I thought that their recommended calibration for the sensor gave me values that were too high. generally 12-20 degrees above ambient during cruise and idle. Also before I would fire up the car at all, temps were already 10 degrees above ambient. This spreadsheet seemed to give me more realistic values.
I started doing things from AEM's Resistance's value here. [url]http://www.aemelectronics.com/files/info/30-2010_2.pdf
[/url]
I made a quick and dirty spreadsheet based on AEM's data with adjustable Internal resistance.
I basically did this by setting everything up in Celcius according to these temps and calculated the Voltage. using the above equation. and 1229 for the resistance.

This is the resulting values in C.


I attached the quick and dirty spreadsheet below.

Toyota/Denso stuff from the early 90s, at least, uses 2.7k (measured 2.685k) 1% MF bias resistors, and the curve looks like this, except the twist at the end, which was error for sure:



That's my data, I measured it myself years and years ago. The sensors themselves look like this (the green one):



You can use any value bias vaguely near the centre of the curve, it just pushes the accurate area up and down, that's all. I provide preset calibrations of 2.0, 2.2, 2.4, 2.49(gm), 2.6 and 2.7k for the above, the gm and the standardish bosch sensor curves. If you want to be really accurate with your temperatures, then you need to measure your actual sensor and actual bias resistor and use a curve for the pair.

The math to generate a temperature curve based on the voltage lookup for a given sensor curve is fairly straight forward, but not suitable for doing in real time. It uses three points of resistance and temperature, with the bias value, to give you a temperature for any voltage. It's googleable/duckduckgoable, but I can post it if you want.

Fred.

Do you mean the 'linear' portion of the curve when you say 'accurate area'?

[off-topic] I'm using a GM Temp sensor in this thing and I just used a 5th order polynomial eqn to model the curve instead of a lookup table.

What linear point? The most "accurate" bit is where a large voltage change represents a small temperature change. And vice versa.

The 'linear portion' of the curve (temp vs voltage) has the highest resolution ('accurate area' as you call it).

But but but, there is no linear region! It's a curve, by definition! :-)

https://en.wikipedia.org/wiki/Thermisto ... t_equation

Sure there is. Have you plotted a temp vs voltage curve? There is a significant section in the middle where a linear model can be substituted in with over 99% correlation.

Unless your comment was tongue-in-cheek. Then I lose at teh internetz.

No, it's just that 99% isn't good enough for me :-p

99% seems like a dubious claim in combination with "significant section" too. Clearly if you zoom in enough any part of any curve starts to look straight. Just as when you drive fast enough, previously "straight" roads start to have a hell of a bend in them :-p

I think mrf582 was just referring to a part of the curve as the "linear part" to describe a certain area of the curve. Not stating that it was perfectly linear, but that it has "linear likeness" and that large changes in resistance are smaller changes in temperature, therefore it is more accurate in that portion of the curve. There is a definite linear "like" section from ~3500 to ~8000 ohms

He threw me off with "the middle" which to me is between 3k and 50C and the curviest bit.

There are two "most linear" bits, the really hot bit, and the really cold bit. In the middle it's very curvy.

The really cold bit has the accuracy/granularity, and the really hot bit is worst in that respect.

To model any of it with a straight line or low point count curve is rough/ghetto IMO. I was poking through one of the subby curves a week or two back, and it looked as though they use a 16 point table or similar. I have to say, I was quite shocked to see that. :-/

Fred.

You seem to be describing the Temperature versus Resistance curve. The ECU doesn't see that. It only sees a Temperature versus Voltage curve. And when you take a Negative Temperature Coefficient thermistor like the stock IAT sensor and run it through a pullup resistor, you get the following.

See that part in the middle from about 1.5V to 3.0V? That is the most linear part of that curve. It has the most resolution because a larger change in Voltage correlates to a smaller change in Temperature.

I think what you are really getting confused over is the following. The ECU or any electronic device really, can't measure resistance directly. They can only read the voltage using a circuit with a known resistor then back-calculate the actual resistance. Look at the following Temp vs Resistance curve and how it looks as a Temp vs Voltage curve after using a pull-up resistor of 1000 ohms.


The bottom curve is how the ECU sees it. Again, the part in the middle has the most resolution. You could model the curve from the 1V to 4V region using a linear fit and it would be over 99% accurate.

Well now, don't I both look and feel like an idiot! :-)

This is my data, measured by me, generated with my code, and never, in 5 years of using it, visualised like this before! Degrees C vs. 10 bit ADC count.

Yep, I apologise! Thank you for setting me straight, I appreciate it.

You'd still have crappy accuracy at the end without lots of points, though. And it's still faster and easier to just grab a number and use it. But the point that I was arguing about was correct.

Woops.

Fred.

Glad we're on the same page now.

Yeah, the end points still need more data points. But since we know the data points in the middle are redundant, we can rescale the temp scaling map accordingly. Remove some from the center and add some to the ends.

Anyway, looks like we can conclude this bit of off-topic tangent.