How much trial-and-error at your work location? 7

[DHambley]

I'm just curious as to the prevalence of trial and error in different industries. Below, I describe two situations with two different design methods. It may be sucessfully argued either way that one is better than the other.
Method 1:
Try a 1/4 inch shaft on the new conveyor belt design. Put the max load on it. It broke. Hmmmm, I wonder why. Let's try a 1/2 inch shaft. Install the 1/2 inch shaft. Put the max load on it. It worked for a few hours, then broke. Hmmmm, I wonder why. Let's try a 1 inch shaft. Install the 1 inch shaft on the new belt. Put the max load on it. It didn't break! Cool. We'll use a 1 inch shaft for production. Total design time: 8 hr.
Method 2:
An engineer models the shaft and coupling. It takes him about 12 hr to set up the CAD model. This model includes worse-case extremes of shaft misalignment, temperature, shock load, speed, etc. It takes him about 4 hours to go through the model with different shaft diameters. He concludes that a 7/8 inch shaft is the minimum acceptable to gaurantee the life of the product, so he adds a margin to increase the diameter to 1 inch and specifies a 1 inch shaft for production. Total design time: 16 hr.

How much of method-1 do you see at your work location?

 

[TXspace]

The allowable Trail-and-Error in a company is inversely proportional to the amount of "Error-then-Trial", if you get what I mean.

That is to say..., if failure is not an option and things are expected to work correctly right out of the gate then there may be a "trial" after any "error".

But if failures are of little consequence -- in terms of safety and cost -- then trial-and-error can be acceptable. Why design a $1000 solution for a 10-cent problem?

But most companies with engineering departments have them for a reason: they do want/need the design to work correctly right out of the gate. Overkill on the up-front design effort can avoid lots of costly errors/failures downstream.

16 hours of one designer's time is a lot cheaper than 4 hours of 5 techs and 2 managers staring at the failed design (i.e. 4 x 7 = 28 man-hours). Not to mention the additional cost due to "There is never time to do it right, but always times time to do it over" syndrome.

The good news here is that seasoned designers don't reinvent the wheel. Sizing a shaft, for example, will typically be calculated once and then future designs can be quickly WAG'ed based on the experience of the previous design calculations.

Sort like if a 300-lb man can walk across a 5-ft long 2x4 board calculated to withstand that load, then a year later a designer will confidently recommend using a 5-ft 2x4 to support a 200-lb man.

Unless there is an overpowering reason to optimize the design, the previous calculated design becomes the rule-of-thumb. And, sure, although testing will almost always be conducted to validate a design, the calculations minimize the number of test failures -- which would lead to another round of costly tests.

So pure trail-and-error should be considered a no-no except for rare situations of precious little significance and consequence.

 

[btrueblood]

"if failure is not an option "

"So pure trail-and-error should be considered a no-no except for rare situations of precious little significance and consequence. "

Then no new devices would ever get created. Real R+D is a lot of educated guessing, followed by testing, followed by analysis and then repeated as required until the error (what you've got vs. what you wanted) is reduced to an acceptable level. Sometimes, not very often I would agree, but often enough, you find yourself blazing truly new territory. It's pretty fun when it happens, and you have to be willing to stub your toes and break stuff in trying to get stuff working the way you think/hope it ought to.

I find it funny that a discrete control PID loop is essentially a trial and error process used to control all kinds of widgets in the real world.

 

[Compositepro]

Btrueblood makes a good point. Many people have a complete misconception as to what "trail and error" means. In this usage, error does not have the connotation of failure or catastrophe. It simply means a deviation from the desired result, as in error signal in a control loop. We drive to work using trial and error. When the car's path deviates from the desired path by more than a comfortable amount we take action to correct the error. Too many people take the term error personally and react by denying that there is an "error" because that would mean that they are "wrong". Trial and error is probably the most common and most useful method of problem solving, even for engineers. All that the analysis and calculations do, is allow for the size of the errors to be kept smaller. I guess the flip side of "trial and error" is "paralysis by analysis". There has to be balance and perspective.

 

[geesamand]

Where I work, we have no room for trying things. Except for the pure research in the R&D lab, it's expected that any/all mechanical designs should be done "correctly" on the first time through without failures. We are constantly doing variations of existing products ranging from minor alterations to completely custom, so especially with the custom cases a prototype with testing is required to ensure flawless operation but we're never given the opportunity. In the cases where a prototype is planned, sales sells the prototype before it's even been built.

David

 

[chicopee]

With method 1, it may take 8 hrs of design time but think of the cost in labor including supervision, injuries , parts, tool replacement,utility useage, unless your plant is in China, India etc...

 

[Cockroach]

As important as trial-and-error may have been in past centuries, personally I don't support the assertion that this method is of value in engineering. I do recognize that engineers are typically recipe types who reference closed form solution sets from technical references, this per-supposes there is an understanding in the model limits or the assumptions inherent to that paradigm anyways. So on that note, I guess we are all guilty of some measure in trial-and-error, philosophically speaking.

Regards,
Cockroach

 

[GregLocock]

There's a lot of trial and error when you are fine tuning something. For example, crankshaft bending dampers are fitted to improve sound quality, and despite the attempts of many, there is no conclusive way to compare the sound of two engines graphically or by calculation . Similarly, tuning shock absorbers for ride and handling largely comes down to suck it and see. Indeed, tuning the boost curve for the power assisted steering also comes down to a mixture of measurement, calculation and observation.

To some extent all of these are open loop investigations, there is often no specific target and the result is the best you can get as a judgement, in the time available, with the hardware to hand.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[Walterke]

In the cases where a prototype is planned, sales sells the prototype before it's even been built.

Welcome to every company ever

NX 7.5.5.4 with Teamcenter 8 on win7 64
Intel Xeon @3.2GHz
8GB RAM
Nvidia Quadro 2000

 

[btrueblood]

FWIW, my definition of trial and error, and I think Greg's, does not align with the OP's example. Nobody I know uses trial and error as a first design method; instead it is a method to refine a model beyond the limits of analysis available. The following comments continue using my definition:

Every custom design outfit, even geesamand' shop, is using trial and error, but they refuse to admit it. Their trials, in the form of one-off designs, get tested in the field, by their customers.

Cockroach, would you say that testing is also of no value then? I.e. why bother to test a product, since you "know" that you have designed it correctly? We test products because we know there are deviations that occur between real world objects and our simplified analytical models, and that these are beyond our limited ability and/or finances to completely control, and we hope that those deviations don't reduce the performance of the product beyond acceptable limits. But if you don't test it, your customer will. Just hope that the feedback loop gets closed by the customer directly, and not via his lawyers or those of his survivors.

 

[Cockroach]

I'm mechanical design - oilfield equipment, BTrueBlood. On occasion we might test for shear screw strength only because of the uncertainty in the brass chemistry, but for trial-and-error in prototype development, never. So I agree with you on your first design methodology comment.

But don't get trial-and-error mixed up with "testing". We test all the time, usually destructively for on the first-off. This is consistent with ISO 9001 Design Control were the interest is in "verification", not "validation". By the time the prototype is build, we have a real good handle on validation. If something doesn't test within our strict set of limits, then the model is in error and we search for the understanding as a design error. Validation has failed. This essentially means we go back to the drawing board and after a full review, design again to get it right. Hopefully it is a mathematical error, maybe a poor model or something, but more times than not it is a fundamental oversight in the input design statement(s).

Perhaps it is my mathematical physics background, but I just don't buy into taking a good guess and homing in on reality following a few good tests. That's how I regard trial-and-error, aside from Greg's comment on "fine tuning".

Regards,
Cockroach

 

[flash3780]

I'd argue that trial and error is done all of the time in engineering (though not in the way that is described in the initial post). Generally there are tons of unknowns that go into a design. Often loading is not well-understood when designing a machine, or it is simplified to make the calculations possible. Heat transfer and temperature distributions are especially difficult to predict, since we often rely on empirical models and environments are never any more than an approximation of what the device will actually see. Flow network models assume flow coefficients (which have been part of engineering since the Romans) for idealized geometries. Tons of approximations.

So, we (engineers) make educated assumptions about such things and refine our assumptions as a design progresses. Critical areas of a machine which push the mechanical or thermal limits of the machine's components are often validated with component tests, which further help us to understand their behavior. Finally, once we have done the best we can at producing a working design, we build the machine and test it as well.

At the end of the day, designing is all about trial and error. We use the best tools we have to make the smartest decisions about our designs that we're able to... and as we learn more about the thing we're designing, we iterate. Finally, after a lot of iteration, crumpled up papers, and frustration, we arrive at a design that we think is adequate and meets its design requirements... and we build it. From there, we see how we did, and how the device performs. We monitor it in service and listen to what customers like or dislike about it. We monitor performance, follow maintenance issues, wear issues, and other serviceability issues. We learn what worked well and what could be improved, and we use that input when designing our next machine.

Engineering is all about trial and error, actually. However, it's also about using your knowledge of physical sciences to make smart design decisions and minimize design iterations.

Could you imagine building an automobile by making parts through trial and error as was described in the initial post? There are thousands of parts which need to be sized. How much would it cost to have so many failed tries? How would you ever arrive at a competitive design?

 

[KENAT]

flash3780, arriving at a competative design of an IC now primarily by trial and error most likely would be impractical. However, 100+ years ago one suspects quite a bit of what we might now consider trial and error went into such things.

Hence the various comments of there still being trial and error on some 'cutting edge' applications - as well as on those which might not be cutting edge as such but we still have trouble fully defining/analyzing.

Posting guidelines faq731-376

http://eng-tips.com/market.cfm?

(probably not aimed specifically at you)
What is Engineering anyway: faq1088-1484​

 

[flash3780]

KENAT, I agree... though 100 years ago machines were often simpler (though not all were; steam locomotives are quite complex). There were innovative methods that engineers used to determine design adequacy, though. Gustave Eiffel, for example, used Tresca's maximum shear failure criteria when designing the Eiffel tower. Clearly there was no room for trial and error in such a venture.

Test components were often constructed for machines, however. One method for evaluating them involved painting parts with shellac; areas where the shellac cracked were assumed to have excessive strain and the design was modified to correct it.

Design has always been a bit scientific... and always has involved some level of trial and error.