Irregular grids

[Ian1]

Not sure if this is the right forum - but

I have a set of data from thermocouple positions and I want to poduce a contour plot. The thermocouples were are not in a regular grid. In effect they are on a 20x30 grid, but only 20 nodes are used.

To produce a contour plot I need a regular grid that is filled - so my question is about 2d interpolation from a sparsely populated grid

any ideas ????

[ptc-1368288]

On 3/31/2010 9:25:01 AM, ianjohnston wrote:
>Not sure if this is the right
>forum - but
>
>I have a set of data from
>thermocouple positions and I
>want to produce a contour plot.
>The thermocouples were are not
>in a regular grid. In effect
>they are on a 20x30 grid, but
>only 20 nodes are used.
>
>To produce a contour plot I
>need a regular grid that is
>filled - so my question is
>about 2d interpolation from a
>sparsely populated grid
>
>any ideas ????
______________________________

Just attach the data grid from a data file Mathcad can read or make one in your version. If you have Mathcad 14, "Save as" 11 for + collab to read.

It sounds an RHSMD contour plot.

jmG

[ptc-1368288]

It sounds an RHSMD contour plot.
rather RHMSD

[RichardJ]

You only have data for 20 points on a 600 point grid? I don't think the right approach is likely to be interpolation. I don't think the resulting contour plot would mean much. Do you have any kind of theoretical equation that should represent the data? If you do, the best approach would be to least squares fit that to the data.

Richard

[ptc-1368288]

On 4/2/2010 10:43:14 AM, rijackson wrote:
>You only have data for 20
>points on a 600 point grid? I
>don't think the right approach
>is likely to be interpolation.
>I don't think the resulting
>contour plot would mean much.
>Do you have any kind of
>theoretical equation that
>should represent the data? If
>you do, the best approach
>would be to least squares fit
>that to the data.
>
>Richard
_______________________

>In effect they are on a 20x30 grid, but only 20 nodes are used.<<br> ___________________

I read this way:

a 20 x 30 rectangular matrix of which 20 columns out of 30 are validated , therefore a complete square 20 x 20 square matrix that can be surface plot as finely meshed as desired for a complete contour plot as well as the surface plot.
The data set is needed c/w the columns that are valued.

Few minute project = done

jmG

[ptc-1368288]

On 3/31/2010 9:25:01 AM, ianjohnston wrote:
>Not sure if this is the right
>forum - but
>
>I have a set of data from
>thermocouple positions and I
>want to poduce a contour plot.
>The thermocouples were are not
>in a regular grid. In effect
>they are on a 20x30 grid, but
>only 20 nodes are used.
>
>To produce a contour plot I
>need a regular grid that is
>filled - so my question is
>about 2d interpolation from a
>sparsely populated grid
>
>any ideas ????
______________________________

I will leave this thread as you didn't provide data. Your project is not abstracted, as such I have an immediate comment: I you use ordinary grade T/C's there is several �C uncertainty between them. If you use special grade, there is less discrepancy, but nothing is true in either case. As you are mentioning a matrix of T/C's, whether a micromatrix for biological work or a matrix for geological work where the T/C' might be km apart from each other it is logical to consider that all T/C' do respond equally. Rosemount used to charge huge $ for their module to read their T/C. At the time of the Mathcad 11 DAEP, the special T/C calibration was implemented for the Mathcad users and this special linearisation technique was passed to NI [National Instrument], they were "delighted" [as they reported]. NI has that special linearisation in one or several of their input card. If you don't revisit this collab with data, you might have advantage to contact NI and explain your project to them.

The method was applied to valid T/C, i.e: J = large range, T = cryogenic ... each type in their complete range.

jmG

[Ian1]

Just returned to this thread, and realise I probably didn't make the problem clear. I didn't really want to get into a discussion about calibrating thermocouples - I was just trying to set the scene. Anyway, my thermocouples were arranged as in attached file "thermocouples 2.jpg". I interrogated the thermocouples 4 times a second for 100 seconds and have generated an animation - see attached "EC71 thermal.wmv".

In the end I filled in the missing values by interpolation and then reset all values where the interpolation became extrapolation. Might not be mathematically as rigourous as I wanted, but the results looked reasonable and the customer was happy!

Thanks for advice and I must try and look at the RHMSD in more detail .....

[RichardJ]

The results do look reasonable. Very nice in fact.

How did you do the interpolation?

[Ian1]

Hi Richard

First I discarded the thermocouples on the left of the centre line as shown on the map I posted earlier - these were for redundancy.

Then I created a 15x16 matrix and mapped the values I knew onto the relevant positions.

Then running along each of the columns with data in, I filled in the missing values with linear interpolation - I then ran along each row and linterp(ed) the missing values again to fill the empty columns. I had to tidy-up before the first and after the last point in each line as linterp starts extrapolating and was giving silly results

Finally I generated a mirror image and stuck the arrays together to create the symetrical pattern

Obviously I then just FRAMEd through the data file to get the time steps for the animation

If anyone is interested I will tidy up the sheet and post it

Ian

[RichardJ]

Ian Johnston wrote:

If anyone is interested I will tidy up the sheet and post it

I would be interested in seeing it.

[jeanGiraud]

Simple questions beside what you think you might have done:

1. where are the T/C's inside on a perpendicular plane,

on an inclined plane, which plane [X,Y ?], what's their spatial position ?

2. are they on a single external helix or two helix like DNA ?

You can really see how important it is on helix !!!!

That would involved the "Fourier DNA" maths. No problem, this

type of work exists in the tool box.

Click on the image to view the cylinder

jmG

[jeanGiraud]

>Then I created a 15x16 matrix and mapped

the values I knew onto the relevant positions.<

_____________________________________

You must provide the XY coordinates.

I can understand the principle of what you are trying.

Let say, insertion T/C's but why so weirdly distributed

in a tube, for what purpose ?

Please abstract the project .

If not possible to unzip your work sheet, then ?

If you are sure of your mapping, then attach or show

in one way or another that 15 x 16 matrix. It will take

less than one minute to get the surface plot and contour

plot from the tool box.

Click left on the image .

jmG

[jeanGiraud]

>Thanks for advice and I must try and look at the RHMSD in more detail .....<

Reading back, I made the suggestion RHMSD [jmG]

I can see your matrix and T/C location , though I don't understand what

it means in a tube 553 mm ? If you would want an "average °C"

of the fluid flow, the Instrumentation man wouldn't do this way.

We could dispose 4/5 T/C's in the cross section, at the best points

based on the Chebyshev location [like in the Annubar flow DP].

The other way is to locate a single [or two for redundancy] T/C's

at the "Moulinet distance" [the distance at which we install the

moulinet flow device]. By itself the moulinet isn't very accurate

5 ... 10% error of max flow. This magnitude of error is about

what you can expect from grade 2 T/C ... a bit better from

grade 1. It is simple to personalise type "J" and "T" based on the

Mathcad "Calibration of Themocouples" that has been implemented

in some of the NI input card.

You revisited the PTC community on June 25 2010,

Richard replied on the same day,

I downloaded your Mathcad work sheet but can't open.

"The compressed [zipped] Folder is invalid or corrupted"

My WinZip works fine as it did extract Dumitru latest work sheet.

So, we are in the mud. What to do ? And we can't "e-mail"

Try to post the work sheet in the Mathcad collab,

I will read for few more days. Plug it in one of the two left

working conferences. In your file name there is an extra dot

"EC71 thermal.WMV", removing the dot does help WinZip ?

Sorry, Ian ... that's the best for today.

jmG

My e-mail is not private, but we must try the natural ways first.

Expect to read you soon.

[Ian1]

Hi Jean

I wasn't really planning to go into the details of my particular problem,but it seems important for your understanding.

First, the file I attached "EC71 thermal.wmv" is not a worksheet - it is a media file, it is the output from the animation.

The thermocouple positions are on the inside surface of a thin-walled tube 553mm circumference and 312mm long and the illustration shows the inside of the tube flattened - imagine joining the left and right sides together - then you have a cylinder. A hot jet is in close proximity to the outside of the tube for about 100s and I wanted to know the temperature distribution on the inside surface. Nothing is flowing inside the tube. I have attached a couple of photos.

Img_2419 shows the thermocouples fixed to the inside of the tube

Img_2470 is after the test, and shows the tube with the hot gas nozzle

I hope this makes it clear.

I have sorted out my problem for now and got good looking results - that was the "wmv" file. I will tidy up my worrksheet, annotoate it and post it with the data file over the next few days.

Ian

[jeanGiraud]

>I have sorted out my problem for now and got good looking results <

Often, good looking results aren't true at all !

One may wonder why you didn't distribute the T/C's in a regular grid ?

Square or rectangular, no matter you get a traceable technique.

At that presumed high °C, the ± error between T/C's might not matter

much, still surprising like in the steel industry, in molten metals.

Thanks for the details, will look back for the work sheet.

You are just guessing °C where there are no T/C's.

jmG

Amazing: several 50k *.gif just go away on the click,

but *.jpg take nearly a minute. So, it unpacks *.jpg

just on the click, but to pack it back to PTC it then

waits they have finished breafast.. .