Best fit on profile of a line (technical question)

  • Filter
  • Time
  • Show
Clear All
new posts

  • Best fit on profile of a line (technical question)


    Does anyone have a useful link to learn more about the least squares best fit. I'm looking forward to do manual calculations for a line profile data set.

    Last edited by mespejel; 04-08-2020, 01:36 PM.

  • #2
    Could you describe a little more.
    The line is on the flat section, or is it an edge ?
    Do you "cut" the COP construct a line ?
    I would say profile in front of standards should use minmax vector.


    • #3
      The deviations can be calculated by different ways :
      - least square (alone) just uses distances between measured values and theo values :
      and then minimizes the sum of the square of deviations.
      - least square vector uses the definition of the deviation, which is the projection of the deviation along the theo vector :
      and then minimizes the sum of the square of deviations.
      - minmax vector uses also the definition of deviations, but the algorithm search to minimize the difference between the max deviation and the min deviation.

      ​​​​​​​You can try using the excel sheet in the #6 here, I'm not sure it works fine in english...


      • #4
        That's different values that I find (different than yours !):
        Minimax 0.4933
        max 0.5157
        min 0.0224
        LS vector 0.6377
        max 0.2516
        min -0.3270
        LS only 1.7577278
        max 0.4933
        min -0.3738
        Last edited by JEFMAN; 06-14-2021, 08:25 AM.


        • #5
          Yep, values are approximately the same (can change after the 4 decimal place)
          Init is used to start the calculation with 0 on each displacement.
          LS only is the minimization of distances with least squares, LS vector is distances along the vector and minmax also.


          • #6
            I think it's because of the deviations along z.
            On theo values, there's a variation between 5 and 33 mm, so the vectors shouldn't be <0,0,1>.
            Projecting along the vector or calculating the distance "point to point" (meas to theo) give very different results because the distance are not measured along the same direction.


            • #7
              I would use minmax vector...
              LS vector is intersting if there are some outliers hits...
              I can't answer to your question without seeing the part, the method...
              I just would say that you can give the result that you want if you add to the result the algo used...


              • #8
                I think it should be the same,exept if my prog is wrong !


                • JEFMAN
                  JEFMAN commented
                  Editing a comment
                  I made it on my own, just applying what I know...
                  I know that a french teacher wrote something about it, I don't know if he did it in english (Bernard Anselmetti).
                  You can try a google search with "using excel solver in gd&t" ?

              • #9


                Related Topics