Computers back in the day, what value? 4

[ParabolicTet]

Just 30 years ago , workstations would cost millions of dollars , have less than 100 kb of RAM, and take several engineers working full time to operate and program.

My question is how on earth did large companies justify the cost of buying such expensive machines for such little return on investment. You would be limited to the most basic of analysis which you could do faster and cheaper by hand. And what basic part would be worth the multi-million dollar price tag to design it?

It seems many companys underuse FEA despite computers being so relatively cheap.

 

[GregLocock]

"Just 30 years ago , workstations would cost millions of dollars , have less than 100 kb of RAM, and take several engineers working full time to operate and program."

I think you are a bit out with your dates/costs/capabilities. I ran my first FEA in 1999.

FWIW I have built many useful FEA models with fewer than 1000 nodes. They also correlated well with the real world.

The reason car companies got involved at that time was that they could see the potential.



Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[GregLocock]

sorry 1979

Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[KENAT]

It seems many companies underuse FEA despite computers being so relatively cheap.

Funny, I was thinking just the opposite today. We have a relatively simple stress analysis to be performed on a seismic bracket (simplified to static loads), I previously helped an intern do similar with classical/hand calcs.

This time my manager and the senior guy whose design it is immediately jumped to FEA. One of our new interns has an interest in FEA so she's been given it to do. Trouble is I'm not sure there's anyone here to really supervise her properly, in my immediate group I'm probably the most qualified which is just plain scary.


KENAT, probably the least qualified checker you'll ever meet...

 

[corus]

I first did FEA in 1980 designing Nuclear thingies. The workstation didn't cost millions, and there is no way the same analysis could have been done by hand. The basic part being assessed was worth millions and worth the time and effort.

Even a small component may not cost much to manufacture but if in mass production or as part of a larger component, then the costs can be high if failure occurs. Generally I've found that hand calculations make such broad assumptions, take ages to do, and are full of mistakes, that they're very rarely worth doing. I'd prefer to use FEA and take the human factor out of it as much as possible.

corus

 

[40818]

FE was used way back in the day in aerospace, for determining loading in statically indeterminate structures. Dont think it cost quite that much but it allowed better aircraft which then sold better.

 

[cbrn]

Hi,
ParabolicTet, the company I work in began using such machines you mention in the late '70s. I've been told it was effectively painful (people travelling from Italy to Switzerland with bags of perforated cards...), but the fact is that your assertion of these machines being used for tasks which could be performed by hand is simply out of reality. There are LOTS of things that a hand-calc can only roughly approximate or even not approximate at all. In these cases, the decision was: either use a "supercomputer", or exit your business area. BTW, just an example from "my world": today, a single hydraulic turbine can cost milions of Euros. In the past, total cost was even higher, so the ROI of even a "super-machine" could be a few years.

Regards

 

[fecad]

it is good to know what happend in end of (19)70s & early (19)80s, i just born at that time and using FEA now at my work with couple of yrs. of experience behind me ..luckily we have fast computers available but less experience in hands..and when we will reach at your experienced stage..youngers at that generations may use palm computers for FEA which may cost some peanuts

cheers,
fecad.

 

[mrgoldthorpe]

I feel an anecdote coming on.....

I started FEA proper in 1977. The company I worked for used a program called PAFEC, sold by a spin-off company from the University of Nottingham, UK. The program cost a few hundred or so pounds sterling (GBP). That was a one-off cost, and you had the source code. We reported bugs back to them, and even sorted them out! We added our own code.

We ran FEA on a Honeywell mainframe that cost several hundred thousand pounds (about 400,000 GBP, I think) and cost a fortune to run and maintain, needing overnight operators for example. We struggled to do single 2D, non-linear analyses overnight. The only output was a line printer isting of results.

I recall we tried to get a VAX 11/780 costing about 80,000 (1979!) pounds (160,000 USD). No way!

Today the commercial program costs about 13,000 GBP per seat and runs on a 400 GBP computer, perhaps 10000 times faster with 10000 times the storage capacity of that 1970s machine.

Software and hardware costs have changed in a very big way!

Funny though, we had more thinking time, and I think in some ways the overall productivity was better then because you had to understand what the 'black box' did in those days.

 

[Denial]

MrGoldthorpe,

Your anecdote stirred a few old memories.

I too cut my FE teeth on PAFEC, but running it on a computer bureau in London in 1975. That way a failed run cost you real money, as well as a heap of time, so you learned the discipline of painstaking checking before launching the analysis. Perhaps this partly explains your (totally correct) feeling that "we had more thinking time" back then.

Such discipline is lost to me now, and I almost always adopt a "suck it and see" approach rather than a fully considered one. A pity really. But I would rather not return to the days of carrying a large box of punched cards along a rainy street, using the umbrella to keep the cards dry while I got soaked.

 

[johnhors]

Some of the opening remarks in this article echo the thoughts of MrGoldthorpe and Denial

http://www.etm-project.com/Roshaz_Art_1a.htm

 

[rb1957]

ah, the 70s and the state of computers ... at the place I was at we had three teletype terminals (none of this new fangled screen dispalys, also saved on a printer !) for 100+ engineers; computer was called (of course) "George"; big runs were done in London (a daytrip) often with no usefull result.

yeah, we're all getting lazy these days !

waiting for the 1st post "luxury! ..."

 

[johnhors]

rb1957

Wasn't "George" the name of the operating system used by ICL computers (such as the ICL1900 series) ?

But at least the ICL1900 allowed you to use a card deck, similarly my first experience with FEA was also to type the whole thing in at a teletype, then some hours later it would print out the results, from which you would manually create plots and graphs !

 

[KENAT]

johnhors

Star for you, excellent link, thankyou. I plan on sending it to my boss, not sure how that will work out for me though ;-).

The first section very much reflects my understanding of the current state of play of FEA.

I'm not an FEA user as such, I've played around with the one embedded in our CAD software but given my complete lack of training or experience am always highly dubious of what I get out. I'm very aware of the garbage in garbage out idea.

As mentioned before I'm involved in supervising interns that get asked to do FEA, previously only with the inbuilt tool in our CAD (FEMAP Express) but now one is going to use ANSYS. Most of what they are analyzing is very simple and I try and get them to do classical calculations first, including the FBD.

Does that articly reflect what the majority of experienced posters here believe?

KENAT, probably the least qualified checker you'll ever meet...

 

[GregLocock]

"Does that articly reflect what the majority of experienced posters here believe? "

Sure. My prediction (which I made a couple of years back) is that these integrated FEA tools will be withdrawn immediately after the first lawsuit resulting from a fatal accident where pushbutton FEA was a cause. the sad thing is that the draughtie will get it in the neck, it really is not his fault.




Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[johnhors]

Thanks Kenat, in my opinion there is a fundamental lack of understanding among middle and senior managers for what FEA is about. Fifteen to twenty or more years ago FEA was not a main stream activity, but the reserve of geeky specialists, who just happened to know what they were doing and had to work on a first time, every time right basis or lose their employers lots of cash. Now with FEA embedded within CAD there is a perception that it is just another module for the CAD jockey to run with. The real problem is the wysiswyg (what you see is what you get) nature of CAD itself, after all the CAD system will not allow the user to perform impossible solid boolean operations, so why should the FEA module be viewed any differently? I know of at least two CEO's of leading software vendors who have publically boasted that their user interfaces for FEA are so easy to use that even non-engineers can run them! So how long before Greg's prediction come true ?

 

[KENAT]

Yeah, I seem to recal the chairman of then UGS saying something similar at a conference last year.

The fact is there are designers that turn out rubish with CAD, it's just often less critical and/or easier to spot, if nothing else it often gets spotted on assembly of first article and while it may cost money/schedule rarely if ever hurts anyone.

Doing the same on something safety critical like stress calcs, which may not be spotted before reaching the customer, is scary.

KENAT, probably the least qualified checker you'll ever meet...

 

[jte]

I agree with the sentiments above. I've dealt with (like, yesterday) experienced engineers who cannot figure out the deflection of a simply supported beam with a point load in the middle combined with *gasp* a uniformly distrubuted load. Give 'em a quick FEA button and all the worlds problems will be reduced to a rainbow of red thru blue. From about 12 years ago: "Red is bad, right?" Sheesh. "Yes, unless I'm showing you a deflection plot or compressive stress plot." From the days when a 386-25 was a hot machine and a color inkjet printer was a genuine novelty. Oh, well, as GregLocock pointed out, we need to keep our friends in the lawsuit business employed.

 

[johnhors]

Then that makes three CEO's !

 

[jmagri]

In many cases the companies in the 70's did not own their own hardware or license the software. In my instance we rented time from Control Data ( Old line supercomputer company). This meant a trip to their facility with a data deck of punch cards, submitting this pile of cards, and running a check run. If that ran then you submitted the data for a full run. The analysis would be run on a priority depending on how fast you wanted it and how much you were willing to pay. Then you would also be charged for the run time. This was an excellent way to burn through some very large amounts of cash. I don't know what those machines had for power but large runs could take hours.

Since many of the analysts had been doing work by hand for years with handbooks this was a giant step forward. For the most part until Algor came up with a PC based system in the early 80's only the biggest companies or the ones that had the most riding on a design were using FEM.

By the way in the States an auditorium roof collapse in the 70's was partly attributed to an insufficiently conservative design that was confirmed by computer FEM calculation. The design failed to consider buckling.