Hacker News Reader

Comment by WalterBright | Hacker News Reader

> extend JWST's life span

I recall a thread here when it was launched where I suggested building a twin simultaneously, and that the increment in cost was likely to be 10% of the cost of one.

One of the critics of this idea said there was no need for another, as there was only so much the JWST could discover. But it's hardly been turned on before people are trying to figure out how to make it last longer. Sigh.

P.S. I found out later that in the past NASA would build probes in pairs, and made extra parts in case one was damaged or didn't work. So it really couldn't be that expensive to have built a twin.

View on HN

You have indeed raised this idea before. The replies in other threads, among people familiar with this design space, will be the same here, as they were before (https://news.ycombinator.com/item?id=29855830).

Namely:

These devices are one of a kind items, fabricated, integrated, and tested manually, not on some kind of assembly line. There isn’t an economy of scale.

Your conjecture is just not correct. It’s remarkably hubristic to think the people who design these space telescopes have continued to do it wrong because making duplicates has not occurred to them.

Even if building a 2nd one cost the same as building the first, how much of that $10 billion is design and development cost? $0 of that will factor in to the cost of the second.

I've fabricated many things with my hands and machine tools. The second one takes dramatically less time, in every case. Even the 2nd set of materials cost less. For example, I ordered a needle bearing the other day for $7, but with shipping the total came out to $20. If I ordered two bearings, the total cost would have been $27, not $40.

It took me 20 minutes or so to install it. If I installed it a second time, I could have done it in 5 minutes.

The reason is simple. I had the right tools laid out, and I knew exactly what to do the second time.

So, yeah, I was quite unconvinced in the last thread.

P.S. I did not say they were doing it wrong.

I agree entirely. It's not like JWST components were built from diamonds or something where the material costs dominated the budget.

It wouldn't surprise me if there were already more than one made for many of the bespoke components used in the telescope. Nobody makes a one-off component without some iteration and covering of their own ass in case something goes awry in shipping or assembly.

There is no reason a priori to believe that integration and test of one of a kind space cryocoolers scales the same way postage does, or batching out parts at a drill press.

I’d suggest that it’s on you to demonstrate why these analogies should hold.

> integration and test of one of a kind space cryocoolers

The point is they wouldn't be one of a kind if #2 was built. Planning, designing, iterative prototypes, designing tests, designing test equipment, building test equipment, devising test plans, writing the enormous amount of software require for all of that, for the ground stations, etc., all add nothing to the cost of building #2.

Normally, people take the cost of a program and divide it by the number of units built, and call that the per-unit cost. That's an accounting fiction. The first one costs the bulk, the rest cost far less per unit.

> no reason

The reason is I can't think of any endeavor where the incremental cost of #2 doesn't drop dramatically.

When is the second copy of anything ever the same price as the first? That thread talks about the second being maybe 30% of the cost of the first. Is that not “economies of scale”? And I think that seems like a great deal, given how much universe there is to look at, and the potential for the first to not survive until the end of its expected lifetime. Hell, why not put 10 of them up there, and iterate on the design?

Perseverance was originally a twin of Curiosity, but it cost more than Curiosity. There's no way that a twin of JWST would cost only 10%.

> Perseverance was originally a twin of Curiosity, but it cost more than Curiosity.

I can't imagine how that could come about.

> There's no way that a twin of JWST would cost only 10%.

It's kind of the way building things works. The cost of the prototype is enormous compared to the next one. For one thing, the additional R+D cost is $0. The additional cost of the software (and I bet the custom software is a big chunk) is $0. The additional cost of committee meetings to discuss competing alternatives is $0. And on and on.

When cutting parts, the cost isn't in cutting the parts. The cost is setting up the machine to cut the parts. The cutting cost is trivial.

I still can't figure out why this "economies of scale" misconception is so popular on HN with respect to JWST. The major costs of building JWST are in the testing + validation + refinement phases, which must be done meticulously for every unit that is built. A JWST "out of the box" is guaranteed to fail: literally, you could launch a million "unrefined" JWSTs and every single one of them will experience a critical mission failure.

Most components of the JWST are not within spec as they leave the factory floor; for many components, the precision required cannot be achieved with machining metrology alone. Remember that system error compounds with every new component that is incorporated. Components have to be constructed, integrated, and then measured/validated with sophisticated metrology equipment after full assembly. If you're lucky, you can modify the components you have to achieve the desired overall tolerances. But a lot of the time, you have to bin the same component a dozen times until you get a batch which happens to be correct (much like in microchip manufacturing).

And this is just for physical manufacturing -- there are multiple other dimensions which are impossible to get right the first time, requiring multiple iterations until your integration tests pass. Many of these test scenarios are extremely expensive to simulate (e.g. full-size vacuum chambers, launch and zero-g simulators), and must be done to validate every single phase of a 5-year mission to an extremely high chance of success (from transport to launch site -> launch -> full deployment -> science operations). Something as simple as a wrongly-tensioned cable is enough to scrap an entire mission -- the validation is absolutely essential to ensure that anything from a manufacturing defect to a simple human error doesn't make it through to launch.

Even in spite of all the lessons learned from JWST 1, I would be surprised if JWST 2's cost was less than 50% of JWST 1 (realistically, I'd peg it at ~80%). The testing costs are a very high fixed cost that must be paid for every unit you make. There's no other way around it.

I remember seeing a documentary on designing the parachute for one of the Mars landers. JPL build this huge building solely for the purpose of testing the parachute designs. Design after design after design failed, and the engineers were worried that they'd never figure out how to make a working Mars parachute.

But they did come up with a design that worked. Phew!

The cost of that special building, the building's design, the special machinery that filled it, and all those months of testing various parachute designs must have been enormous, and all count for the cost of parachute #1. The construction of parachute #2, after all that, was likely insignificant in comparison.

Again, you'd have $0 in research and development and software and test rig costs of #2.

> The testing costs are a very high fixed cost

I'm sure they are. But you won't have to design the tests and build the test rigs and validate the test procedures a second time. Secondly, you'll inevitably learn from the first test runs to need less iteration.

For example, the full-size vacuum chamber. You would already have it on hand, and not need to build another one. Having already just run #1 through it, you'd know just what to do to get #2 through.

For example, the first time I took the heads off my Mustang it took 4 hours. The second time 2 hours. The third time 20 minutes. The procedure was already all laid out for me in the shop manual. But knowing just what to do cut the time enormously.

> When cutting parts, the cost isn't in cutting the parts. The cost is setting up the machine to cut the parts. The cutting cost is trivial.

When you're building a one-off, you aren't setting up a machine to cut the parts. You're just cutting the parts more or less by hand. It'd be too expensive to set up the machine, and calibrate, and run all of the prototypes to make sure it works, if all you need is a couple of pieces out of it.

I've made parts on milling machines and lathes. The time spent is on the setup, not the cut.

Reply on news.ycombinator.com