2D-Linear Fit needed

[Mario Hirth]

Hello,

I have to find the best straight line in a x-y measurement array.

The "Linear Fit.vi" uses a linear regression method normal to the x-Axes which leads to bad results if the sollution would be near a vertical straight line.

e.g.: with a Pair of measurement points with:

1.: x=0 Y=+1

2.: x=0 Y=-1

I would like to get a straigth vertical line (x=0, k=inf or something which represents this), the Linear Fit.vi's result is NaN

Does someone know where to find a vi for this?

Thank you for any comments.

 Greetings

Mario Hirth

[dan_u]

Hi Mario,

one possible workaround is to fit in both directions, i.e. y vs. x and x vs. y.

If you know your data will usually be close to vertical just do the x vs. y fit.

Hope this helps,

Daniel

[Mario Hirth]

Hello Daniel,

thank you for your quick answer.

I allready had a workaround where I swap x<>y if the y elongation is bigger than the x elongation.

Leading to better results for straight lines in x OR y direction, but not in between (45deg?).

Since I use the residuum of the linear fitting to evaluate the measurement (which should represent a straight line, but I dont know in which direction), I have to be mathematically correct.

I also had a solution by iteration with: "find slope" => "turn measurement data into x-axis according to slope" => start again with finding slope => do this about 5 times etc...

in principle this works, but since I have to make a iteration with the result of this output (which is a different story), I have a iterationtask  inside an iteration task.

It works, but its neither a beatiful sollution nor a quick (means CPU-intensive).

But thank you anyhow

Mario

[altenbach]

---

Mario Hirth wrote:

Since I use the residuum of the linear fitting to evaluate the measurement (which should represent a straight line, but I dont know in which direction), I have to be mathematically correct.

I also had a solution by iteration with: "find slope" => "turn measurement data into x-axis according to slope" => start again with finding slope => do this about 5 times etc...

in principle this works, but since I have to make a iteration with the result of this output (which is a different story), I have a iterationtask  inside an iteration task.

It works, but its neither a beatiful sollution nor a quick (means CPU-intensive).

---

Mathematically, this makes no sense to me. Can you explain what it is supposed to achieve?

What is the error in x and y for each data point?

[Mario Hirth]

Hi altenbach,

Why should this make no sense?

The Linear Fit.vi trys to find a straight line which minimizes the Summ((Delta y)^2) , I want a line which minimizes Summ(Sqrt((Delta x)^2+(Delta y)^2))

In other words, not the deviation in y-direction but in a direction normal to the fitting line.

By turning the measurementpoints into the x-axis according to the slope of my fitting line, "y-direction" and "normal to the fitting line" is the same.

Shure I have to remember how much I have turned my measurement points, but this is not a real problem.

greeting Mario