Searching \ for '[EE] Fast temp sensor?' in subject line. ()
Make payments with PayPal - it's fast, free and secure! Help us get a faster server
FAQ page: piclist.com/techref/io/sensors.htm?key=sensor
Search entire site for: 'Fast temp sensor?'.

Exact match. Not showing close matches.
PICList Thread
'[EE] Fast temp sensor?'
2008\06\25@001348 by PicDude

flavicon
face

I'm searching for a fast temp sensor for measuring ambient air temperature
(with some overhead) -- say -10deg-C to 120deg-C.  1% accuracy or better.

There are a lot of sensors that meet this numerical criteria, but what types
of temps sensors are *fast*?  I usually don't see response time listed in
the datasheets (and not even sure what the parameter is for response time),
and I don't have a specific value for how fast it should be, but I want to
compare various sensor types and start investigating/experimenting with the
fastest of these.

I did find "fast response RTD's" and wondering how they compare (speed-wise)
to other alternatives such as semiconductor sensors.  I expect the plastic
housings for TO-92 types and similar would be enough of an insulator to
significantly increase the response time.  Will the metal can versions
(TO-39 IIRC) be considerably faster?  Open element sensors and low-mass
sensors should be fast.  But I'm sure I am missing a lot of other
possibilities here.  

Any links/clues on this?

Thanks,
-Neil.

--
View this message in context: www.nabble.com/Fast-temp-sensor--tp18104658p18104658.html
Sent from the PIC - [EE] mailing list archive at Nabble.com.

2008\06\25@002527 by Sean Breheny

face picon face
Hi Neil,

I've been impressed by the response speed of tiny thermocouples. Their
response is on the order of a second. I'd say that it is on the order
of 20 seconds for typical semiconductor temp sensors.

RTDs are probably in the same range as thermocouples but tend to be
more expensive I think.

I have seen to-92 packaged temp sensors from National which list
thermal response time in their graphs section. The response is roughly
exponential, like an RC circuit with a step input, so the graph should
give you the time constant (at which time the temp difference is
reduced to 37% of the original difference).

You could also possibly greatly increase the response speed of any
temp sensor by mounting it to a small heatsink and blowing air over it
with a small fan.

Sean

On Wed, Jun 25, 2008 at 12:13 AM, PicDude <spam_OUTpicdude2TakeThisOuTspamavn-tech.com> wrote:
{Quote hidden}

> -

2008\06\25@003422 by Jinx

face picon face
> housings for TO-92 types and similar would be enough of an
> insulator to significantly increase the response time.  Will the
> metal can versions (TO-39 IIRC) be considerably faster?

'dude, the LM34/LM35 datasheets show that the TO-46 metal
can is significantly faster than the TO-92 for stirred oil. Handy if
you're measuring stirred oil

TO-46  = 3 seconds to 100%
TO-92  > 8 seconds to 100%

AD590 comes in a can too (pricey, last time I looked)

Response in still air though shows the same curve for both packages,
200s to 100%. There's a curve for Thermal Time Constant vs air
velocity as well

2008\06\25@003735 by John Coppens

flavicon
face
On Tue, 24 Jun 2008 21:13:44 -0700 (PDT)
PicDude <.....picdude2KILLspamspam@spam@avn-tech.com> wrote:

> But I'm sure I am missing a lot of other
> possibilities here.  

One of the most important factors you didn't mention is thermal
resistance. If the sensor is in free air, it'll take much more time than
when it is in contact with the thing to measured. It's very much like a
when you need to cool with heat sink, just the other way around.

If you can make the sensor have contact with the object you'll have much
better speed with metal housings. Else (air contact) it'll be pretty much
the same, and the time constant will mainly depend on the sensor's mass
and air circulation.

John

2008\06\25@030201 by Vasile Surducan

face picon face
The reason why you'll don't find fast and accurate temperature sensors
is because the temperature variation is slowly by definition (faster
in air speed and slower in liquid). The speed of the analogic
temperature sensors is gaven by the size of the sensor. A thermocouple
without case or a thin film RTD is faster than a thermocouple with
case or a huge 5mm diameter RTD.

You can find at Omega some small and cheap RTD which could met your 1% request.
But from the sensor accuracy to the final measuring value accuraccy
it's a long way.

Vasile

On 6/25/08, PicDude <picdude2spamKILLspamavn-tech.com> wrote:
{Quote hidden}

> -

2008\06\25@082333 by Marc Nicholas

picon face
I'm doing a lot of work with thermistors these days....my current prototype
is <0.1c accuracy and 0.1c resolution on the digital side. This is from a $4
thermistor encased in copper.

You'll actually find "standing air" response times published by most
thermistor manufacturers -- as long as it's a thermistor truly meant for
instrumentation. In my experience, they're in the order of 25s in standing
air. Moving air, stirred oil, and water figures are sometimes supplied and
will all be faster.

Not sure if that really helps, my my 2 cents Canadian ;)

-marc

On Wed, Jun 25, 2008 at 12:13 AM, PicDude <.....picdude2KILLspamspam.....avn-tech.com> wrote:

{Quote hidden}

> -

2008\06\25@084721 by Alan B. Pearce

face picon face
>Open element sensors and low-mass sensors should be fast.

Well, you do not say just how fast you need them to be, but I would have
thought the fastest would be the tiny bead thermistor types would be ideal,
and well suited to the temp range you mention. One like the YSI 4400x series
is what I am considering here, although they may need some small
linearization for best results.

>But I'm sure I am missing a lot of other possibilities here.

Personally if you can stand the cost I like the Analog Devices AD590 device.
One of the package options is a lowish mass ceramic case that seems to have
reasonable thermal response times. Sensitivity is 1uA/K, and they are
linear, operating over -55 to 150C. Use a 10k resistor to ground, and a
supply high enough to have at least 5V across the sensor, and you have a DC
voltage just right for putting into the ADC pin on a micro, although for
best accuracy you need to offset the low side of the reference voltage.

2008\06\25@084819 by Spehro Pefhany

picon face
Quoting Marc Nicholas <EraseMEgeekythingspam_OUTspamTakeThisOuTgmail.com>:


Seems to me that you really should start with a specification (or some idea
of what you require) and go from there.

I have used thermocouples (open wire wire and open ribbon) for really fast
response with air (for example, to control a proportional valve for a  
combustion
heated air flow). But to get 0.1°C *accuracy* is probably optimistic unless
it's laboratory conditions. 0.1°C short term stability shouldn't be a problem
with good instrumentation and controls.


{Quote hidden}

>> -

2008\06\25@091810 by Ariel Rocholl

picon face
2008/6/25 Sean Breheny <@spam@shb7KILLspamspamcornell.edu>:
>
> You could also possibly greatly increase the response speed of any
> temp sensor by mounting it to a small heatsink and blowing air over it
> with a small fan.

Intuition tells me there is probably a difficult tradeoff to balance
between adding lot of "temperature inertia" to the device with a
heatsink and how to compensate it by additional airflow. I wouldn't
add any heatsink to get faster response time.

--
Ariel Rocholl
Madrid, Spain

2008\06\25@105340 by Sean Breheny

face picon face
Hi Ariel,

Yes, there would be a tradeoff. I used the term "heatsink" very
loosely. I had in mind a small piece of copper tape or foil, attached
to one of the leads of the part. This way, you enhance the thermal
conductivity from the air directly in to the die of the sensor. I have
not done tests here - it could well be that a bare sensor would be
faster.

Sean


On Wed, Jun 25, 2008 at 9:17 AM, Ariel Rocholl <KILLspamarochollKILLspamspamgmail.com> wrote:
{Quote hidden}

> -

2008\06\25@110050 by Vasile Surducan

face picon face
On 6/25/08, Marc Nicholas <spamBeGonegeekythingspamBeGonespamgmail.com> wrote:
> I'm doing a lot of work with thermistors these days....my current prototype
> is <0.1c accuracy and 0.1c resolution on the digital side.

Could you elaborate how do you really know that your device
performance  is better than 0.1C accuracy as long the display
resolution is equal with the accuracy ?

Vasile

2008\06\25@114528 by Spehro Pefhany

picon face
Quoting Ariel Rocholl <TakeThisOuTarochollEraseMEspamspam_OUTgmail.com>:

> 2008/6/25 Sean Breheny <RemoveMEshb7spamTakeThisOuTcornell.edu>:
>>
>> You could also possibly greatly increase the response speed of any
>> temp sensor by mounting it to a small heatsink and blowing air over it
>> with a small fan.
>
> Intuition tells me there is probably a difficult tradeoff to balance
> between adding lot of "temperature inertia" to the device with a
> heatsink and how to compensate it by additional airflow. I wouldn't
> add any heatsink to get faster response time.

Just because a tradeoff is difficult to evaluate doesn't mean you  
shouldn't try to do it. ;-)   If the heatsink did not impede flow to  
the sensor, then anything you attached thermally to it that responds  
faster than the sensor itself should speed things up because heat will  
flow from the faster bits to/from the slower bits. Assuming a large  
amount of air flow etc.


Best regards,
Spehro Pefhany
--
"it's the network..."                          "The Journey is the reward"
s...EraseMEspam.....interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com


2008\06\25@142746 by Marc Nicholas

picon face
Hi Vasile,

That's a good question.

I actually have greater resolution available than 0.1c on the device via a
serial debug session and also have an industrial thermometer to check
accuracy of the target sample.

-marc

On Wed, Jun 25, 2008 at 11:00 AM, Vasile Surducan <EraseMEpiclist9spamgmail.com>
wrote:

> On 6/25/08, Marc Nicholas <RemoveMEgeekythingEraseMEspamEraseMEgmail.com> wrote:
> > I'm doing a lot of work with thermistors these days....my current
> prototype
> > is <0.1c accuracy and 0.1c resolution on the digital side.
>
> Could you elaborate how do you really know that your device
> performance  is better than 0.1C accuracy as long the display
> resolution is equal with the accuracy ?
>
> Vasile
> -

2008\06\26@153834 by Martin

face
flavicon
face
PicDude wrote:
{Quote hidden}

Find something in a very small package and then use more than one to
achieve high statistical accuracy.
i.e. n samples with 1 degree accuracy would give you 1/sqrt(n) accuracy
IIRC.

You could possibly do a derivative calculation to achieve quicker
estimates of the ending reading
i.e: You read 105 degrees. The dT/dt is -4 degrees. You might predict
that the future sample is likely to be closer to 101 degrees or that the
ACTUAL temperature is lower than 105 because you can account for your
sensor's thermal mass.
"YMMV"

-
Martin

2008\06\26@162029 by Gerhard Fiedler

picon face
Martin wrote:

> Find something in a very small package and then use more than one to
> achieve high statistical accuracy.

I second this.

> You could possibly do a derivative calculation to achieve quicker
> estimates of the ending reading

Unless you know very well what kind of input signal form you are getting or
actually model your sensor well, I think any calculations of this sort make
the readings slower (worst case) rather than faster. Then there's the
differential problem: errors get amplified.

Gerhard

2008\06\27@095733 by Martin

face
flavicon
face
Gerhard Fiedler wrote:

>> You could possibly do a derivative calculation to achieve quicker
>> estimates of the ending reading
>
> Unless you know very well what kind of input signal form you are getting or
> actually model your sensor well, I think any calculations of this sort make
> the readings slower (worst case) rather than faster. Then there's the
> differential problem: errors get amplified.
>
> Gerhard
>


Right it would take some testing. It would be interesting to see if it
worked. I'm sure it's not a new idea but I don't know the formal theory.

-
Martin

2008\06\29@101435 by PicDude

flavicon
face

Lots learned a lot over the past few days, and it really does seem that
thermocouples (especially open-elements) are very very fast.  Type T
thermocouples are apparently the most accurate and available in two classes,
with 0.5 deg-C error versions commonly available.  But thermocouples need a
lot (relatively) of support circuitry, especially since this is in an
automotive application (measuring air temp at a couple points in the intake
tract).  1sec to 2sec works fine, since this is a monitoring application,
and 1% error would be nice, although a bit more would still be acceptable.

Mounting (for the end-user) will be a major factor as well, so any basic
element/component that needs to be PCB-mounted becomes less desirable, as
I'd have to factor in the labor cost to build those to a user-friendly
format ... and that's still be decided upon.

I had missed the graphs with response time in those datasheets (LM35 etc) --
I was looking for electrical parameters.  But checking back now, they're
relatively quite slow.  I found RTD's that are available in better than 1
sec response time using open elements, and they require minimal support
circuitry.  They also come mounted to a stem/flange/etc so end-user install
is ready-to go.  The problem is cost -- so far they're in the $35-$40 price
range each (in 50-100pc quantities).

I always thought that thermistors were not-so-accurate, but I've been
corrected again it seems.  I'm looking at various datasheets now in the 0.5%
to 1% range, and they seem promising.  Very low cost (<$1 ), and I'm working
out a mounting structure to see just how much it would cost if assembled
in-house.  I also found some suppliers of thermistors in flanged probes, so
I have some calls to make this week.

Thanks,
-Neil.



--
View this message in context: www.nabble.com/Fast-temp-sensor--tp18104658p18181248.html
Sent from the PIC - [EE] mailing list archive at Nabble.com.

2008\06\29@141925 by Michael Hagen

flavicon
face
Talk about FAST temperature sensors.  I worked on a medical instrument many
years ago.
I made a test machine based on a PC and software in Pascal that dipped the
thermistors in ice water.
The thermal time constant had to be 20 - 40 mS to work in the final
application.
Resistance went to 66% of final temperature in less than .04s!  Some were
thrown out because they were too fast.

This was a very special thermistor made for us by Thermometrics (NJ , I
think).  It also had to meet low leakage  and a high voltage spec.

They are still going and will custom make you a special themistor if needed.
They will help you specify your part so it always meets a certain accuracy
under specified conditions.

Mike

{Original Message removed}

2008\06\29@145807 by Spehro Pefhany

picon face
At 10:14 AM 6/29/2008, you wrote:

>Lots learned a lot over the past few days, and it really does seem that
>thermocouples (especially open-elements) are very very fast.  Type T
>thermocouples are apparently the most accurate and available in two classes,
>with 0.5 deg-C error versions commonly available.  But thermocouples need a
>lot (relatively) of support circuitry,

The cold-junction compensation also adds an error term that could be
a problem in applications where the ambient varies widely (such as
automotive). It adds an error that is approximately 1:1 (eg. a 1 degree
error in the secondary sensor causes approximately a 1 degree error in the
reading. For this reason, T/Cs are better suited for EGT than intake. ;-)

>  especially since this is in an
>automotive application (measuring air temp at a couple points in the intake
>tract).  1sec to 2sec works fine, since this is a monitoring application,

Hmm.. what would cause the air temperature to change so quickly? Could
you use a milled out PCB as a combination mounting method, electrical
connection and heatsink? Maybe they you could use a thermistor or SMT
semiconductor sensor.

Monitoring is a *lot* easier than control. In control of air temperature
the sensor adds a pole to the response that can cause overshoot/undershoot
of the actual (not the measured) air temperature unless you grossly detune
the controller to make the response sluggish. In cases where there
is an object being heated or cooled that responds faster than the sensor
you can get serious problems if the actual air temperature does that.

{Quote hidden}

They can be okay (sometimes very good), but not so trustworthy at higher
temperatures. If your maximum is going to be 40 or 50°C, that's much less
of a worry.

>  I'm looking at various datasheets now in the 0.5%
>to 1% range, and they seem promising.  Very low cost (<$1 ),

1% of a thermistor is a lot fewer degrees than 1% of a substantially linear
sensor such as a thermocouple or RTD. Thermistors are extremely nonlinear,
so they're more appropriately used over a relatively narrow temperature range.
 Watch that there is also a tolerance on the beta. The tolerance is kind of a
conical tunnel that is narrowest at the reference temperature (eg. 25°C) and
quite a bit wider at both ends.


{Quote hidden}

2008\06\30@002804 by PicDude

flavicon
face

I actually had some discussions with those folks a year or more ago about
thermocouples (for a different app).  They're on my list of calls for this
week, although I'm also looking into custom-making something simple.

Cheers,
-Neil.



Mike Hagen wrote:
>
> ...
> This was a very special thermistor made for us by Thermometrics (NJ , I
> think).  It also had to meet low leakage  and a high voltage spec.
>
>

--
View this message in context: www.nabble.com/Fast-temp-sensor--tp18104658p18188507.html
Sent from the PIC - [EE] mailing list archive at Nabble.com.

2008\06\30@004916 by PicDude

flavicon
face


Are you suggesting that cold-junction compensation is a problem in the
automotive environment even with compensation?  For EGT probes, I have the
junction inside a sealed enclosure (so the probe cannot be removed), and the
temp compensation is done in there.  But because of that, plus amps, etc,
I'm leaning away from that currently.

I don't mind non-linear outputs, as long as they're repeatable --  I can
always use tables.

The temp after a turbo can change quite quickly, but I'm going on specs from
the customer.

> Monitoring is a *lot* easier than control.

And don't forget control has a lot more liability associated with it.

I used to use the DS18B20 (8-soic or 8-ttsop) and I had it on a small 1/32"
thick PCB with a hole right under the chip, but it was time-intensive to
solder the wires etc to the small PCB.  Currently I'm working on
custom-making a sensor probe using a small bead-type thermistor (see
615-1001-ND at digikey as an example), and it would be mounted in a tube
with the bead mostly protruding out the end.  The tube would be filled with
some type of epoxy and the other end would have a couple wires on it.  I'd
need to solder/weld a flange onto the tube, but it would be sooo much easier
if I could find a small-diameter tube (say 0.1" to 0.125" ID) that's
threaded on the outside.  That way, two nuts on the outside would be used to
secure it.  That's what I'm searching for currently.

Cheers,
-Neil.




Spehro Pefhany wrote:
{Quote hidden}

intake
{Quote hidden}

--
>>I was looking for electrical parameters.  But checking back now, they're
>>relatively quite slow.  I found RTD's that are available in better than 1
>>sec response time using open elements, and they require minimal support
>>circuitry.  They also come mounted to a stem/flange/etc so end-user
install
>>is ready-to go.  The problem is cost -- so far they're in the $35-$40
price
{Quote hidden}

so
{Quote hidden}

>>

2008\06\30@015144 by Spehro Pefhany

picon face
At 12:48 AM 6/30/2008, you wrote:

No CRLFs.

>Are you suggesting that cold-junction compensation is a problem in
>the automotive environment even with compensation?

Yes, the accuracy of the compensation. With a T/C system you really
have two temperature sensors, and a low level
amplifier, and in general the errors of all three can add. If you
don't need T/C characteristics (ruggedness, wide range,
maybe fast response), you're far better off to just use the CJC
sensor as your thermometer and forget the T/C entirely!

>For EGT probes, I have the junction inside a sealed enclosure (so
>the probe cannot be removed), and the temp compensation is done in
>there.  But because of that, plus amps, etc, I'm leaning away from
>that currently. I don't mind non-linear outputs, as long as they're
>repeatable --  I can always use tables.

The problem isn't the difficulty  of digital linearization--that's
generally a triviality-- it's loss of resolution. You might have a 100:1
range in resistance for -20C to 100 C range. Even if it's repeatable,
you will have problems trying to use even a 12-bit ADC
to get 0.1C resolution even though there are only 1200 0.1 C steps
between -20 and 100. With a straightforward measurement
of resistance you'd need more like 16 bits or some kind of range
switching or linearization circuitry *ahead* of the ADC.

{Quote hidden}

You have something against compression fittings?

Best regards,

Spehro Pefhany --"it's the network..."            "The Journey is the reward"
RemoveMEspeffspam_OUTspamKILLspaminterlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com



2008\06\30@064919 by Michael Rigby-Jones

picon face


>
> At 10:14 AM 6/29/2008, you wrote:
>
> >  especially since this is in an
> >automotive application (measuring air temp at a couple points in the
> intake
> >tract).  1sec to 2sec works fine, since this is a monitoring
application,

> {Original Message removed}

2008\06\30@141048 by Alex Harford

face picon face
On Sun, Jun 29, 2008 at 7:14 AM, PicDude <RemoveMEpicdude2TakeThisOuTspamspamavn-tech.com> wrote:
>
> Lots learned a lot over the past few days, and it really does seem that
> thermocouples (especially open-elements) are very very fast.  Type T
> thermocouples are apparently the most accurate and available in two classes,
> with 0.5 deg-C error versions commonly available.  But thermocouples need a
> lot (relatively) of support circuitry, especially since this is in an
> automotive application (measuring air temp at a couple points in the intake
> tract).  1sec to 2sec works fine, since this is a monitoring application,
> and 1% error would be nice, although a bit more would still be acceptable.

Hmm... GM has IAT (intake air temp) sensors.  I'm not sure if they are
fast enough for you.  But, they have the advantage of being built to
live in an automotive environment.

GM parts Direct: http://www.gmpartsdirect.com/
GM IAT Sensor : 25036751
Pig Tail : 12102620

It has been a while since I've been digging around in GM ECM stuff,
but I believe the temp monitoring was in the outer loops (every
second).

2008\06\30@145323 by Carl Denk

flavicon
face
Ford also.

I think typically nearly all vehicles stating 1996, which was the main
launch of OBDII have them. Possibly pick one that you can buy the
connector matching. The shop manuals (at least the Ford) have
troubleshooting that includes the pinouts and power and signal values
expected. There are all sorts of sensors available in the auto area
including absolute barometric pressure, various pressures and vacuums,
float switches, etc. Select one from a common model, will be easy to
duplicate if needed.

From the aviation area, high temperature thermocouples that respond
quickly are used for exhaust temperature on piston engines. It's common
to have one thermocouple in each exhaust stack, makes it easy to
diagnose a fouled sparkplug ,clogged fuel injector nozzle, or broken
exhaust valve!

Alex Harford wrote:
{Quote hidden}


'[EE] Fast temp sensor?'
2008\07\02@090050 by PicDude
flavicon
face


Yes, but there's no way this would meet the accuracy spec.  SMP's catalog
shows accuracies of some OEM sensor in the 30% range (yes, 70% error!), and
I can't remember anything over 80% accuracy.  These are the guys that supply
to the OEM's.  And ECU's don't really need to know the actual temp, they
just put in fuel & timing compensation factors in a table at *that* temp,
whatever *that* temp may be.  This is why OBD-type gauges for temps &
pressures are useless.  (But they're great for things like injector
duty-cycles, etc, and manifold/baro pressure is generally fairly accurate).




Alex Harford wrote:
{Quote hidden}

> --

More... (looser matching)
- Last day of these posts
- In 2008 , 2009 only
- Today
- New search...