multi channel nonlinear fitting

[punkmonkey]

Hi all.

I'm trying to do analysis of J2263 data. The final equation is in the form of

F=A+Bv+Cv^2+a0+GW^2(a0+a1Y+a2Y^2+a3Y^3+a4Y^4) where

v=Velocity channel

W=Wind Speed Channel

and Y=Anemometer Yaw channel.

G is a constant 

A,B,C,a0 through a4 are coefficients that I am solving for.

It'd be swell if we could write multi channel non linear fit equations.

I am using the nonlinear fit equations to find an initial seed of A,B, and C and then passing it to a grid search of least squres fit for the final equation. It works but it is slow.

[Brad_Turpin]

Hi punkmonkey,

What's "J2263" data?  It sounds like you have implemented your own multichannel non-linear fit with the existing single channel non-linear fit function in DIAdem and a VBScript to programmatically (brute force) sweep the dependent parameter space.  I will pass along to R&D the request to turn what you're currently doing with an involved VBScript into a shipping ANALYSIS function-- that makes sense.

In the meantime you might want to check out the new "Calculation Manager" in DIAdem 11.1.  My guess is that this will be a much nicer way to host your existing solution, particularly if you want to share it with others or generalize the channel input process and integrate your analysis into the DIAdem user interface.

Thanks for the feedback,

Brad Turpin

DIAdem Product Support Engineer

National Instruments

[punkmonkey]

SAE J2263 EPA vehicle coast down data.

I've got DIAdem 11.0 right now. I'll check it out when I upgrade.

[punkmonkey]

How would you go about using "Calculation Manager" to calculate an equation of the form

F(x,y,z)=A+Bx+C(x*x)+(Const)(y*y)(d+ez+f(z*z)+g(z*z*z)+h(z*z*z*z))

where x,y and z are data channels?

Thanx.

[Brad_Turpin]

Hi punkmonkey,

The key is to have your X, Y, and Z channels arranged in what DIAdem call "triplet" order.  Lets say the valid values of these variables are:

X = {0, 1, 2}

Y = {4, 5, 6}

Z = {7, 8, 9}

You need to have the following values in those channels:

X   Y   Z

--  --  --

0   4   7

0   4   8

0   4   9

0   5   7

0   5   8

0   5   9

0   6   7

0   6   8

0   6   9

1   4   7

1   4   8

1   4   9

:    :    :

2   6   7

2   6   8

2   6   9

The formula you listed can be used directly in the Calculation Manager with no edits, 

Brad Turpin

DIAdem Product Support Engineer

National Instruments

[punkmonkey]

Here's a different way of explaining what I want to do.

I need to solve for the b's.

My Y and X channels are tens of thousands of sample long. I have been doing a general grid search least squares fit but it takes along time to get convergence. I haven't found anything in the help for multiple regressions.

Thanx for the help.

[Christian_S]

Hi punkmonkey,

Have you already tried the ANALYSIS > Curve Fitting > General LS Linear Fitting (ChnGenLSFit function)?

It is worth pointing out that the term "linear" refers to the parameters of the fit, not the basis functions.

Unfortunately there is no detailed backgound information in the DIAdem online help but you may have a look at http://zone.ni.com/devzone/cda/tut/p/id/6954 (search for "General Linear Fit").

Christian