How to not fuck up your manufacturing startup • Hacker News Board

Re: How to not fuck up your manufacturing startup

#41

Post by fencepost » Tue, Dec 06, 2016, 8:15 AM UTC

leodeid wrote: post #24

fencepost wrote: post #9

From the article: "A good, defensible manufacturing strategy is one where you’re applying and protecting (ideally via patent) a faster, cheaper, more reliable way of doing something in your industry, by borrowing a proven approach from a parallel industry." If you're looking for a formalized system designed to help with some of this, take a look at TRIZ[1][2]. I'll just steal one note from the "What Is TRIZ" article…

I read the wikipedia page and that "what is TRIZ" article, and I still don't understand what it is. At times, it sounds like an automated program (especially with statements like "More than three million patents have been analyzed to discover the patterns that predict breakthrough solutions to problems"). But at other points, it seems like a human-centric problem solving strategy, but without the strategy. It describ…

I had some exposure to it and a few classes more than a decade ago now, but it didn't really end up being anything that was a good fit for me and I never got into the group of folks trying to do TRIZ for software development.

Probably in part because of the background of the creator and the problem datasets used for the traditional contradiction matrix it always seemed to me to be a better fit for manufacturing and physical goods.

The "Interactive Contradiction Matrix Beta" linked at the top of the TRIZ Journal site may be worth looking at, but it's kind of cryptic. Basically you pick out a few areas of concern - as an example I picked (on both axes) 1: Weight of Moving Object, 9: Speed, 15: Duration of Action of Moving Object, and 27: Reliability. Based on that, the recommended areas that I should be looking at for possible improvement potential would be 35: Parameter Changes (turns up 6 times), 3: Local Quality (5 times), several others at 4 times, etc. Hitting the Analyze button on that tool will give expandable examples for the various areas - for example "Parameter changes" includes a lot of changes to temperature, state (solid/liquid/gas) and consistency. An example might be making liquid-filled chocolates - do you have to fill the chocolates? Can you have frozen chunks of filling that you coat with chocolate instead?

Re: How to not fuck up your manufacturing startup

#42

Post by beachy » Tue, Dec 06, 2016, 9:00 AM UTC

avn2109 wrote: post #28

Animats wrote: post #27

The startup mentioned is an urban indoor farm for lettuce. As a manufacturing process, it's a good case. One product. Few changes. No need to retool for the 2017 Lettuce. Few operations. (In a manufacturing plant, an "operation" is one step in the process.) This is the best case for classic mechanization. You just need to do the same thing over and over while holding the process parameters within tolerance, and do it…

>> "...No need to retool for the 2017 Lettuce." These indoor farm outfits have nothing to do with lettuce. There's a reason they're building them in/around NYC, aka in the densest concentration of pot smokers the world has ever seen, far from Humboldt county's fields and Colorado's manual hydroponics. The business model is just to get the automation figured out with some low-value crop, e.g. salad greens, while waiti…

We toured a hydroponic lettuce farm when we stayed in the South Island here in NZ recently. Pretty low tech, staffed by a small cadre of minimum wage people.

My first thoughts about this article were just as you say - how can that pay off?

Now I get it.

Re: How to not fuck up your manufacturing startup

#43

Post by ju-st » Tue, Dec 06, 2016, 9:25 AM UTC

unixhero wrote: post #34

He forgot to mention: Big clients with aggressive legal stange who promise millions then strangle and cancel the contract you had with them when you've sunk 1000s or hours into R&D.

A lesson from history and geopolitics: Never trust more powerful entities.

Re: How to not fuck up your manufacturing startup

#44

Post by pjc50 » Tue, Dec 06, 2016, 9:37 AM UTC

kriro wrote: post #36

banjomascot wrote: post #18

Earlier quoted context omitted.

I disagree with you - Figure out your process and automate with software. Your software will only be as good as your process. There are mid-market ERPs that are 30 years old or more (specifically, Dynamics NAV comes to mind). These software solutions implement the processes that have been honed by humans for more than a thousand years. It is much easier to innovate elsewhere. I am curious which specific ERP packages…

Figuring out the process is a complex task. A complex task that (imo) is very likely to be solved better by some sort of AI approach than human tinkering. That's where I see software innovation in production. Having an efficient lab/simulation toolkit where you can constantly improve your processes (probably a human/computer hybrid).

Automated Taylorism is going to be the bane of our lives. The non-AI version of "gamification" and endless metrics is bad enough.

Re: How to not fuck up your manufacturing startup

#45

Post by pjc50 » Tue, Dec 06, 2016, 9:38 AM UTC

amk_ wrote: post #30

leodeid wrote: post #24

Earlier quoted context omitted.

I read the wikipedia page and that "what is TRIZ" article, and I still don't understand what it is. At times, it sounds like an automated program (especially with statements like "More than three million patents have been analyzed to discover the patterns that predict breakthrough solutions to problems"). But at other points, it seems like a human-centric problem solving strategy, but without the strategy. It describ…

They teach you this stuff in product design classes. TRIZ is a way of breaking down an engineering design problem into the thing you want to change, and the thing you can't change (a "contradiction"), then resolving it. A "TRIZ Matrix" is a reference tool that suggests ways of resolving conflicts between common design parameters (strength, weight, durability, manufacturing tolerance, etc.) based on a number of princi…

This reminds me of Brian Eno's "oblique strategies".

Re: How to not fuck up your manufacturing startup

#46

Post by arethuza » Tue, Dec 06, 2016, 9:42 AM UTC

kriro wrote: post #36

banjomascot wrote: post #18

Earlier quoted context omitted.

I disagree with you - Figure out your process and automate with software. Your software will only be as good as your process. There are mid-market ERPs that are 30 years old or more (specifically, Dynamics NAV comes to mind). These software solutions implement the processes that have been honed by humans for more than a thousand years. It is much easier to innovate elsewhere. I am curious which specific ERP packages…

Figuring out the process is a complex task. A complex task that (imo) is very likely to be solved better by some sort of AI approach than human tinkering. That's where I see software innovation in production. Having an efficient lab/simulation toolkit where you can constantly improve your processes (probably a human/computer hybrid).

I'm not sure that the actual manufacturing processes are that amenable to the application of AI as they are typically dictated by the actual physical work being done. However, scheduling and, in some environments, pricing in ERP systems can benefit from more advanced techniques though not sure that the ones I have seen would qualify as "AI".

Re: How to not fuck up your manufacturing startup

#47

Post by arethuza » Tue, Dec 06, 2016, 9:44 AM UTC

banjomascot wrote: post #18

AbrahamParangi wrote: post #4

"Figure out software after everything else" I also work at a manufacturing startup ( https://markforged.com ) and I disagree. Manufacturing is ancient. Software as a field has been around for 45 years (give or take). If you're not working at a pure software company, the software for your particular field is probably in its infancy. If you're looking to innovate, it is orders of magnitude more fruitful to look to the…

I disagree with you - Figure out your process and automate with software. Your software will only be as good as your process. There are mid-market ERPs that are 30 years old or more (specifically, Dynamics NAV comes to mind). These software solutions implement the processes that have been honed by humans for more than a thousand years. It is much easier to innovate elsewhere. I am curious which specific ERP packages…

Saying that NAV is 30 years old is like saying Unix is almost 50 years old - while true in one sense it is missing the fact that these systems have been in constant development since then.

Re: How to not fuck up your manufacturing startup

#48

Post by m_eiman » Tue, Dec 06, 2016, 11:00 AM UTC

Strangework wrote: post #26

So wild! Such a surprise to find this on the front page of Hacker News. If it weren't for the author's username, I might have missed this gem. I had worked at Edenworks as a software engineer for a few years, alongside Johnny. I don't pretend to know enough to speak on manufacturing itself, but Johnny's remarks on software development are spot-on. When I had come onboard, I had naively suspected a typical software st…

When the main product being developed is a tangible system, redos are way more costly.

Even when primarily doing software, this can come into play.

I know of a customer who had about 5000 card readers in an access control system, where bugs in the card reader firmware required a firmware update. Doing this required walking to a reader, unmounting it from the wall, updating the firmware, re-mounting the reader. Rinse and repeat 5000 times.

A rough estimate is that upgrading all the readers on the site took about a man-year of technicians walking around updating things. Let's just say that when more bugs were discovered, the customer who paid for it all was less than pleased…

Lesson learned: always make secure remote firmware updates possible on your devices.

Re: How to not fuck up your manufacturing startup

#49

Post by michaelt » Tue, Dec 06, 2016, 11:05 AM UTC

avn2109 wrote: post #28

Animats wrote: post #27

The startup mentioned is an urban indoor farm for lettuce. As a manufacturing process, it's a good case. One product. Few changes. No need to retool for the 2017 Lettuce. Few operations. (In a manufacturing plant, an "operation" is one step in the process.) This is the best case for classic mechanization. You just need to do the same thing over and over while holding the process parameters within tolerance, and do it…

>> "...No need to retool for the 2017 Lettuce." These indoor farm outfits have nothing to do with lettuce. There's a reason they're building them in/around NYC, aka in the densest concentration of pot smokers the world has ever seen, far from Humboldt county's fields and Colorado's manual hydroponics. The business model is just to get the automation figured out with some low-value crop, e.g. salad greens, while waiti…

Do you think this explains the reports of Toshiba [1] Fujitsu and Olympus [2] growing lettuce in clean rooms? I was under the impression that weed legalisation was low on the political agenda in Japan.

It might be that these producers expect higher human-labour costs, making automation more profitable - such as due to rising nationalism reducing the supply of cheap migrant labour.

[1] http://qz.com/295936/toshibas-high-tech-grow-rooms-are-churn... [2] http://www.japantimes.co.jp/news/2014/05/13/national/science...

Re: How to not fuck up your manufacturing startup

#50

Post by pipio21 » Tue, Dec 06, 2016, 11:31 AM UTC

Yeah, the main difference is that in the Software world companies like google make products in beta form and use their customers as testers. They could improve the product "on the fly".

On the other hand in manufacturing having a big recall for a defective or incomplete product usually means "instant bankruptcy".

Even big companies like Samsung(Galaxy recall) or Volkswagen(Dieselgate) or BP(Deepwater oil spill) have to suffer immense loses from "moving fast and breaking things".