NonLinearFitMSE = NOVALUE

[smoothdurban]

Hello

I have been using the non linear curve fitting function for some time now and I have recently encountered a situation where it doesn't seem to be generating the correct result.

The ResultNonLinearFitModelFunction I am using is (1-e^(-a*x)) * (b*x+c)

Up until recently my ResultNonLinearFitCoef1-3 seemed to be correct however I have started to see negatives in these coefficients which triggered me to investigate further.

In addition the NonLinearFitMSE property returns a NOVALUE.

I have used another tool to generate my nonlinear fit coefficients  and I do not get results similar to what DIAdem is generating.

I have attached a set of data. 

The data I am having problem with is group 3 "Post Fade Stop 2"

The resulting non linear fit makes no sense to me.

[smoothdurban]

Another data set same issue.

[JDragon]

I can't say for sure what the issue is. I looked at the TDMS file in DIAdem and the calculations you used. I noticed that in the first 2 data sets, the coefficients were all fairly similar but the coefficients for the 3rd data set was very different.

What is the output you are expecting? 

[smoothdurban]

I am expecting an output to be similar to the other groups.  The point of of the non linear curve fit is to do just that, fit a non linear curve and supply the constants to that curves general equation.

If you plot the data in group 1 as follows, that is pressure on the X axis and APT and NonLinearFitY on the Y axis,  you can see how the function does a pretty good job of fitting that curves.

However if you perform the same overlay with data from group #3 the result is not fitted at all.  I don't understand why the function has no problem fitting the curve with the data in group 1 but fails miserably at fitting the data in group 3. 

Try it for yourself using this as you general equation (1-e^(-a*x)) * (b*x+c) in the non linear curve fitting function using three coefficient (a, b & c) independant variable of x and 10 maximum number of iterations

I am using 2015 DIAdem 32 bit.

[JDragon]

Hi Tim,

I looked through the data again and ran the calculations. I narrowed the issue down to the last value in the data set of Post Fade 2. It seems that the curve fit does not like the data point Pressure=(118.6,118.8) and APT=1.48

If either of those values go outside that range the functions works as expected. It may simply be a symptom of the equation as other equations also result in a valid fitting curve.

[smoothdurban]

I removed the values as you have suggested and the function does execute but the resulting line of best fit is very poor.

When I use Minitab to generate the non linear curve fit uisng the same equation I recieve the desired result and usable coefficents

a = 0.125677

b = 0.009092

c = 0.410816

[Brad_Turpin]

Hi Tim,

R&D has reproduced this problem and created a CAR for it (#615173) which is currently slated to be fixed in DIAdem 2017, due to release in May 2017.  I do not know which Beta version, if any, will contain this fix.  It's definitely under the microscope and on the docket for resolution, though.

Brad Turpin

DIAdem Product Support Engineer

National Instruments

[smoothdurban]

Thanks Brad

I will be looking forward to the 2017 release as we use this function quite extensively and it now has raised questions about all of the test data we have analyzed since February using this function.

[Brad_Turpin]

Hi Tim,

I just verified that DIAdem 2017 Beta3 has already corrected this bug.  Unfortunately, when I try to reproduce this bug in DIAdem 2015, I don't see it.  So we may be talking past each other.  The CAR in question was closed by R&D prior to DIAdem 2017 Beta3, though, so if you don't see it working the way you want in Beta3, we need to take another stab at it.  There will be a Beta4 pretty soon.

Brad Turpin

DIAdem Product Support Engineer

National Instruments

[smoothdurban]

Great!

How do I get copy of 2017 beta3?