Link Copied
@mecej4 with the latest iteration of ELSO provides a master class in how to reconfigure old Fortran. It is quiet a treat to read. The comments are sparse but clear.
!---------------------------------------------------------------
! The joint coordinates are adjusted
!---------------------------------------------------------------
lct = 0
do i = 1, jct
do j = 1, 3
if (jtype(i,j)<=0) cycle
lct = lct + 1
if (j>=3) cycle
cord(i, j) = scor(i, j) + defx(lct)
809 end do
end do
For example he provides a neat rearrangement of the arithmetic if's with the use of cycle.
Some one else looking at it might have offered:
!----------------------------------------------------------------------------------------------
!
! Add the deflections to the original positions
!
!----------------------------------------------------------------------------------------------
LCT = 0
DO I = 1,JCT
DO J = 1,3
IF(JTYPE(I,J) .gt. 0) then
LCT = LCT + 1
IF((J-3) .lt. 0) then
CORD(I,J) = SCOR(I,J) + DEFX(LCT)
end if
end if
end do
END DOThe difference often comes down to individual choice, I do not use cycle, never have. The 809 in the first code is the remnant of the use of a single end do, with the line numbers removed it has to be adjusted.
Harrison's flaw, for want of a better word is adding the extra row to the NxN stiffness matrix, but at that stage there are only academic papers on inversion, principally in structures by Wang. And this was the early days. But now if we want to use MKL we need to reduce the stiffness matrix. Interestingly the run time difference from Harrison's inverter to MKL is almost nothing.
The other issue is the need to check the answers on big problems, so as @mecej4 taught me many years ago, use allocate and a base module, so the problem is easy to scale up. And add a drawing routine so you have a visual output which makes it much easier to check on problems like this bridge over the Mississippi River, sadly it was taken on the 19th Ultimo to be replaced by an interesting structure. The visual code produced in AutoCAD allows for a check of large problems, this one below is about 300 members, and you can do all the checks, member type, stress resultant and lambda factors for a plastic analysis and plot a result for the client.
One can put a Monte Carlo system around the program and test across the full statistical spectrum, this one in Ularc will do 50,000 runs overnight. The standard output file for this number of runs is 50 GB. One of course has VEDIT to look inside the file. Thank God for VEDIT, it is impossible to make DXF file checks without it. The better way is to just collect the 50,000 lambda minimums and do a statistical analysis.
The actual plot is the following which you can give the client. It is much better than a long EXCEL file or 50 GB.
Harrison uses a single brace in one problem. If you need to put a door in you have to use a single brace, as one of the chaps I knew found out the hard way. But a single brace is not a good solution for cost and I would prefer not to have a brace in compression in an earthquake a recipe for dead people.
The secondary problem with AutoCAD is you need a large array, each element covering about 1 mm in size, when you occupy a space on the drawing you switch on the element, you then know not to put two items in the same place, above 5 and 6 were moved apart, but clearly incorrectly.
So we are exceedingly lucky to have such people.
I remember reading a story in the 80's that was the time of management consultants looking for efficiency, apparently the story goes the MC went into the IBM Watson Building. At the end of the work, they said we know what everyone does except the guy in the corner office on the ground floor. He sits there all day and occasionally people pop in and out.
The IBM manager said, he is the guy who solves the unsolvable problems, he is not to be touched, he saves us millions every year..
There are a small number of people like this, my favourite was my fourth year structures lecturer, Peter Kleeman, he was a font of wisdom and humility.
John
