Thursday, March 17, 2016

Making Limited Run Models with a 3d Printer: Lessons Learned

When I first thought about commercially making my Hart gondolas, my big question was “Can I profitably make models on a home 3d printer?”

It’s pretty amazing that I can even ask that question. If I was living in the 1950’s, I’d have needed to learn immense amounts about injection molding to make commercial models; to break even, I’d have to select models that could sell in the tens of thousands. If I was living in the 1990’s, I’d need skills at resin casting and making beautiful masters so I could sell hundreds of models. With 3d printing, I’ve got the chance to mass-produce the models I want in smaller runs.

Form One printer, with the kits it produces.

So can I really mass-produce freight cars with a 3d printer? Maybe. I’d like to share what I’ve learned over the last few months making these cars on my Formlabs Form One printer. If you ever wanted to share a particular prototype with the world, either commercially or for friends, my lessons learned might serve as inspiration and guidance for producing your own models.

First, let’s start with what went well for me.

  • I actually sold 3d printed freight cars, and got mostly positive feedback on the cars. (See, for example, Joe D’Amato’s experience building one of the cars.) I'll give numbers below.
  • I’ve been able to share freight cars that haven’t been available commercially for years, and where the previous models had been coarse and unprototypical.
  • I chose a freight car design that is well-suited to 3d printing. (Again, I'll say more about this later.)

However, getting these cars out the door was much more work than I expected. The difficulties meant that, for me, the project was always partially a labor of love. Some of the particular challenges were:

  • Cost of starting up dominated because there's so many one-off, time-consuming tasks to do.
  • Getting decent yield - enough salable cars - was always a challenge.
  • I'd assumed I could print on demand, but found I needed to print in batches for efficiency.
  • Making these models profitably was a struggle. The possible scale just is too small.

Costs of Starting Up

Startup costs. One my big surprises was how the costs of doing the setup for each model - the startup costs - dominate overall costs. It was easy to account for the cost of each model - cost of the box, cost to print each model, labor to finish and box each car. However, because of my small runs, I completely underestimated the time cost of doing a bunch of essential, mundane, one-off tasks, and how those costs contributed to the cost of making each model. I’d expected that research and design of the models would be time sinks. I was surprised at effort needed for simpler tasks: tracking down boxes, making labels, discovering the right way to print and finish the models, etc. The cost of doing things the first time is inherent with any new venture, but usually gets amortized over a huge number of items. When I’m selling the first model, or selling a small run of less than a hundred kits, those costs are much more obvious.

The solution, for me, was to be pretty careful about what prep work I did for these first models. I held off on finding boxes until I had strong reasons for selling kits in boxes. I avoided setting up an automated way to buy kits on the internet, both so I could control how many orders I took and so I could avoid setting up (and paying for) an e-commerce site. I chose the simplest approaches I could for taking payment (just PayPal), and setting up the web page. I tried hard not to obsess over making everything perfect, but worked to get models ready to sell.

Not all my choices on issues to defer were correct. I'd initially planned to sell the models in a ziplock bag so I didn't have to worry about boxes and labels. I assumed many of the purchasers were going to build the kits immediately... even though I certainly had lots of kits I'd "bought for later" and put on a shelf.

The infamous boxes

Instead, I realized many folks buying my kits for their stash. There’s lots of reasons we modelers buy kits and put them on a shelf. We’re not sure how long kit might be available (stockpile), we think we might need the model for a future project (planning), or we just think it’s a cool model we’d like to own (aspirational). When I realized many buyers weren’t building immediately, I knew I needed boxes to protect the kits. The boxes also needed nice labels to remind them of their purchase, and give them a reason to get excited each time they saw the model on their shelf.

I spent three or four hours searching the web for a reasonable box for my models, and spent similar time figuring out how to use labels for box art. At even minimal values for my time, that probably ate much of my first profits. (FYI: I used for my boxes. Recommended.)

Pilot models were another surprisingly large cost. Pilot models were essential for showing the car was real and attractive, and that folks needed to see these to buy. (I've often had a hard time buying models on Shapeways if the seller has no photos of actual models.) Making each pilot model took a lot of my time (10-15 hours) to build, paint, and photograph. That’s a lot of hours of labor to pay for. I had to redo my first pilot models when I saw that sloppiness I’d tolerate for models on my layout was way too visible in product photos. Those models also couldn't be used on my layout, because I'd need them undamaged and ready for display if I ever sold the kits at a convention.

One surprising cost and risk was keeping parts in stock. I'd initially planned to make true kits, and include everything needed for the kits - wire, styrene strip, chain, etc. I jettisoned this idea before producing the kits. Buying commercial parts opened up the risk of not being able to produce kits if a part fell out of stock; if I'm out of decals or wire, I'm dead in the water. The parts all increased my cost, and knocked my proposed price out of the price range of current resin kits. I instead omitted parts I thought that my customers might already have. I also suspected many modelers wouldn’t add all possible details (such as the chain for attaching the hand brake shaft to the brake gear), so including a few dollars in parts might not actually help most buyers. There were other parts I could 3d print myself so I didn’t need to buy commercial parts, such as the brake cylinder. One neat surprise was that some builders used better grab irons and stirrup steps than I would have supplied.


When I started the kits, I understood that I needed to keep the costs per kit low, but I didn't quite understand how many 3d prints I'd have to make in order to get a salable model. I'd made an initial guess that I'd have successful prints three out of four times (75% yield); my actual yield was around 50% (one out of two models bad). When I had problems printing models, I’d lose significant time trying to track down causes. Right after I declared the kits ready for sale, new prints started getting very rough. I spent a multiple days tracking down the cause to poor orientation (and oddities with the laser spot). Each problem episode meant a couple days of cleaning the printer, e-mailing support, and trying to diagnose causes myself. There were other times when I couldn't get a decent print for love or money.

Author poses with some of the rejected prints. I'm ready to design a realistic wreck scene at the bottom of a canyon.

Failed prints also had a bigger cost than I'd imagined for both inspecting the models and repairing (if possible) any flaws. I'd expected gross failures - the model emerged from the printer with pieces missing. About 20% failed like that; the rest had other problems which required closer inspection: hard-to-see flaws in detail, hollowed-out posts, insufficiently cured cross-braces, sticky finish from insufficient rinsing, warping after printing, or rough surfaces because of printer problems. These sorts of flaws were painful because they required a lot of effort to look over the model, fix, and judge the model against previously-printed models; sometimes, they required changes in my process for making the cars. I made checklists to make sure I looked for all the common flaws. I made gauges to test if car ends bent out excessively.

Comparing models was a challenge. Most of my print runs were in small multiples (1-4 car bodies), so there were always differences between each run. Small runs meant it was hard for me to consider which models had significant flaws. I found it hard to remember what counted as a major flaw, and what was still worth shipping. I ended up running several print runs, then comparing all the models against each other to decide which were decent. Often, I pulled my pilot model or cars on my layout aside to remember where those models weren't perfect.

Photo documenting particular car, along with miniature form for tracking a car batch.

Problems with yield also made me much more systematic about how I was making the cars. I made lots of test prints to figure out orientation and what printed successfully or not successfully, testing whether I could actually mass-produce the kits. I also adjusted the design several times to add support structures to make sure models printed well, and tweaked the support structures so they were easy to cut off of the final model. Problems with warping models, parts breaking during cleanup, or other problems (sometimes caused by orientation, sometimes by printer malfunctions, sometimes from environment, sometimes from weaknesses in the design) always derailed things until I could track down a cause. I created jigs to keep certain areas from warping after printing. I kept paperwork - spreadsheets, diaries, photos of previous models, and paper forms tracking specific batches to remember what I’d done. I kept records of which run each car came from, and which run each buyer received.

Can I print models on demand efficiently?

I'd started with the idea that 3d printing would let me make things just as they were needed. I could have six or seven different designs ready, take an order for a mix of cars, then 3d print them all within a day or two. Rather than keeping stock and spending money on resin, I'd just wait for the orders to come in, print exactly what the buyer wanted, and mail them off.

The idea of printing on demand ended quickly. I found it was a lot more efficient to print a bunch of cars at a time, and then to do the assembly and boxing in an assembly line fashion. Yield problems were one of the reasons, but just assembling the bits for a kit - instructions, boxes, parts - took non-trivial time.

Similarly, I'd assumed that 3d printing would allow me to adjust designs quickly and iterate. This was somewhat true - I could update designs and documentation between runs, and often did. Some changes were transparent - changes in support structures or orientation didn’t affect the future model. Some involved operations of the final cars, such as adjusting coupler height. Other changes affected designs in small or large ways, such as correcting the model when I learned that the side door latches were only on every other door.

However, making changes every few models opened up new problems. If I updated a design, should I keep selling the old? Could I make sure a purchaser got multiple kits with identical details? Did the documentation match the specific car? For some changes (such as correcting the latches), I stopped selling the incorrect cars. To make sure purchasers got identical cars, I wrote batch numbers on the boxes, then checked records to make sure different batches had the same features. Changing the instructions frequently also increased my workload. Rather than print a couple hundred sets of instructions at Kinko’s, I was printing the instructions on my slow inkjet printer, and collating the pages myself.


So I can sell cars... but is it a potential side business?

The answer is "probably not", I think. Doing small runs is great for a hobby business and to satisfy a niche market, but the effort needed to do runs of less than a hundred kits always means it's going to be a wash, financially. I can recount a bit of my costs to help explain the problem.

First, price. I chose the ~$35 price point for my kits to be close to the typical price of a resin kit these days. I assumed I was going after the same sort of buyers as resin kits - modelers interested in precise prototypes, and with the willingness to build kits. I also wanted to make a prototype that could be built as a long string, just like the real cars, and so wanted to make sure the price would allow someone to build five or ten cars. I found a lot of my buyers were also die-hard Southern Pacific fans, often interested in otherwise-unavailable prototypes; I suspect those buyers were less price-sensitive. However, I suspect higher prices would make it harder for these folks to buy from an unknown manufacturer. (For contrast, go to Shapeways, and note the $170 Hart gondola built from a Canadian narrow gauge prototype. The modeler was certainly hoping to get his development costs back, but I suspect most modelers would flinch at the cost.)

For the first model I put on the market (the Hart gondola as originally built), I sold around 40 kits in the first couple months. I suspect I will sell a similar number of the "modern" Hart gondolas. I got these sales through only three ways: advertising on the Espee mailing list on Yahoo Groups, a very nice article in a well-known SP modeler's blog, and word of mouth. The blog post generated the vast majority of the sales, both because the readers included most serious SP fans, and because folks trusted the well-known modeler's opinion. There was a blurb in one of the online model railroad magazines, but it generated a small number of visits to the web pages.

If I was pushing these more, I might consider advertising in some of the historical association newsletters. Because the SP's W-50-3 Hart gondola was also used by the Union Pacific and Pacific Electric, another angle is to advertise explicitly to these communities. I'd also guess I could sell a fair number of kits at the Southern Pacific Historical and Technical Society meeting.

In terms of costs, resin costs, machine costs, and labor dominate. Each car took around $3.75 in resin and $3.00 to cover depreciation of the machine per print; increase those costs to account for prints that failed. Packaging cost around 65 cents for box and labels (not including labor to print and apply) Detail parts (stirrup steps, wire, brake wheel, chain) probably would have been around $4.50 each, which explains why I gave up on including these quickly. 0.010 wire was one of the larger costs; if I had an easy way to straighten and harden brass wire, it might have been reasonable to include these in the kits. I initially estimated I'd need about 15 minutes of labor for kits for all the work: making the prints, cleaning, finishing, and inspecting, packaging. In reality, it took a lot more effort, especially including the costs of handling prints that failed. At $20-30/hour for my time, that's way too pricey per car; to really turn this into a business, I'd need to be able to go from printed model to box quicker. I'm not including additional time - bookkeeping, keeping the 3d printer running, etc.

Keeping printer working was particularly challenging; when it was printing well, I could produce cars efficiently; when it was having problems, I was dead in the water for significant time. My Form One Plus printer is pretty amazing; although it's pricey ($3500), it's much more affordable than the commercial machines, fits in a home setting, and the resin is relatively inexpensive at 15 cents per cubic centimeter. However, the commercial printers have service folks who can be out in a day; the "prosumer" printer won't be that easy to get repaired, leading to a ton of downtime.

I had the Form One fail three times: once, one of the galvanometers went bad and messed up dimensions in one axis; that was an obvious failure deserving a replacement. The second printer died when it slowly stopped printing anything useful. It took about a month of back-and-forth with support before they replaced the machine. The third machine printed 37 great cars (out of 55 attempts) before it started curing the wrong sections. I'm suspecting that's caused by dirt on one of the mirrors; after I finish this post, I'll probably be spending a day getting the printer running again.

Could I have mass-produced models with narrow interest without the Form One? Perhaps. Others certainly have cut molds and injection-molded kits. For the cars I've done, I'm not sure there's enough demand to make cutting molds worthwhile. I also suspect designing a plastic kit for assembly is an immense amount of work. Some cars (like flat cars) perhaps would be better done as resin casting, but doing so requires me to learn a new trade. Finally, I could fall back to Shapeways if necessary. Shapeways price for just the body (not including any profit) would be around $70, raising my retail price significantly.

Best Models for 3d Printing

I think I had one really big advantage with the Hart convertible gondolas: they simply couldn't be made easily with other manufacturing techniques. Anything I produced was likely to be special to folks buying the kits.

I’ve found that my 3d printed models easily had details approaching the quality of injection molded or resin models, and certainly meet my standards for cars on my layout. The models always felt a bit “softer” than injection molded kits, with features often a little rounded rather than being extremely sharp. This was also true with Shapeways models printed on top-of-the-line printers. Though the Shapeways prints were a couple notches better than my "prosumer" (halfway between consumer and professional) 3d printer, the Shapeways prints were cleaner and more consistent. However, I could imagine some modelers would have been disappointed that the models weren’t “as good as” other techniques. If I’d sold a 3d-printed car that was easily done in plastic or cast resin - a boxcar or flat car, for instance - I could imagine some modelers being disappointed that models weren't any better than existing kits.

Luckily, I chose models that can’t easily be done in resin or injection molding - the truss structures on the Hart gondola just couldn’t be done in one piece in any other way. I think this was an important choice for me; my models got compared against either 1960’s craftsman kits, or against the fact that no alternative existed. I got several comments from buyers about how the cars were unlike anything else they’d seen. Their delight meant that folks weren’t comparing the models straight against a kit using other manufacturing techniques. Making a single-piece kit also gives me an advantage; 3d printing kept folks from having to do the fiddly assembly work required by plastic or resin kits.

Future plans

So what did I learn through all of this? Well, my kits show it's possible to make commercial models using a 3d printer like the Form One resin printer. If you're doing this for income, I'd say be very careful - the you really do need to sell hundreds of cars with many different designs to make decent money. Hiccups with the printer can also drive your income to zero, and can take weeks to solve. I suspect model railroad pioneers like Irv Athearn or the brothers who started Kadee were successful because they had skill at keeping their home-brew machinery and injection molding machines running at all costs, not by waiting for the injection molding machine repairman to show up. For a real test, try printing forty or fifty of the model you're intending to sell, and see if your printer (and your processes) can handle it. If you're going to seriously start producing models with a 3d printer, consider buying two so that you have both a backup and a way to swap parts and narrow down the causes of problems.

If you're a fan of a particular railroad and aren't too concerned about making a profit, then 3d printing special cars like this is great fun. I'd still suggest going after cars or details that can't easily be done by resin kit makers or injection molding, because it'll be easier to delight your buyers with something they can't get elsewhere. Definitely make the kits in a big run; make a guess about how many you may sell, and do all the printing and kit assembly at one time. You'll work faster, and you'll have less stress if the 3d printer stops working well.

Even if these models weren't profitable for me, it’s been a fun vanity project. I’d started the project because the material costs looked like a win ($3-$4 in resin, and maybe $3 a model to amortize the cost of the 3d printer). However, producing the models, packing, and shipping the models was a lot more work than I expected. Some of the time required might go down as I get more practice at producing the kits. However, other costs - such as cutting off support structures from printing, handling the models for drilling and tapping holes while the material is less brittle, curing the final model, and inspecting models - were really required by 3d printing and by the materials I'm using.

My future for 3d-printed kits is a little hazy; I'm starting a new day job soon, and that'll be taking my energy for a while. I'll continue to sell the existing stock of cars; when the stock runs out, I'm still undecided about whether to keep producing cars, or instead start selling the Hart gondolas on Shapeways. Either way, I suspect I'll still be doing a ton of projects on my Form One.

I’m still glad I’ve done these cars. I’ve justified the work by thinking of it as a fun project and as a contribution to the hobby. I’ve often described the process as similar to writing the book. There's not much profit for the work required, but I'm able to share details about a very interesting and unusual car with the Southern Pacific modeler community.

Thanks to the folks who have bought Dry Creek Models kits for their support and encouragement. Great thanks to Tom Dill for feedback on assembling an early model. Tom Dill and Joe D'Amato also photos of their completed models, which was a nice reminder of why I'd gone to all this effort. Tony Thompson introduced me to the term "aspirational" for model purchases done because the buyer got a thrill just thinking about building the kit. It's a great term explaining a pretty common way that all us hobbyists behave. Thanks to Jason Hill for insights into 3d printing and injection molding. Finally, thanks to the DCC Lunch crowd in the Bay Area who had to listen to me chattering about 3d printing and gondolas for too long.

Friday, March 11, 2016

Second Kit: "Modernized" Hart Convertible gondolas

If you liked the Hart convertible gondolas I've been making (and selling) recently, but found the cars too old-fashioned for your layout, then today might be your lucky day. I'm now making a "modernized" version of Southern Pacific's W-50-3 Hart Convertible gondola, suitable for model railroads from the 1920's to the 1950's.

As you've read in the past, Southern Pacific's W-50-3 ballast gondolas were neat cars. The cars, built in 1911, used a patented design from the Rodgers Ballast Car Co. to make them useful in many different kinds of maintenance-of-way service. This "Hart Convertible" design looked like a normal gondola, and could be used for hauling ties or other loose material. Open some doors in the gondola's bed, and you'd expose a hopper, ready to dump ballast between the tracks. Close the hopper doors and unlock side dump doors, and you could push dirt out of the gondolas onto the sides of the track. Close the side dump doors and remove the end bulkheads, and you could easily pull rail out of the car for track-laying.

By 1926, plans show that the SP started removing the side dump doors, replacing them with solid wood sheathing on each side of the car. This change also cut the capacity of the car, and also removed the locking mechanisms that held the side dump doors shut. The SP didn't remove the brackets on each post that held the side locking mechanisms - the brackets also supported steel rod holding the car posts to the frame that probably kept the sides square. As a result, the modified cars had odd, unused castings on each post of the car.

SP kept using the Hart gondolas into the 1950's. A hundred of the original five hundred cars were still on the roster by 1950. By 1955. all the W-50-3s were gone. I've heard stories of cars abandoned along the Cuesta Grade near San Luis Obispo, and trackside in the Oregon Cascades.

If you'd like one of these modern cars for your layout, check out the Dry Creek Models website for photos and ordering details.

Thursday, March 10, 2016

3D Printing Trackside Details

3d printing is really fun for filling out a scene; if I spot something interesting in a prototype photo, it's not too difficult to turn that detail into something physical and on the layout. Here's two tiny examples.

This first item is a Southern Pacific luggage cart. I'd certainly seen the baggage wagons in old pictures, but I'd never seen a model like this. I found this cart in an old Model Railroader construction article from the 1990's. That article suggested how to solder the cart together from wire, but it wasn't too bad to model in 3d. My biggest challenge was figuring out how to support the cart while printing, and then how to carefully cut that support structure off afterwards. It's not the cleanest model, but it looks great in the station clutter at my San Jose Market Street station scene. I also liked adding in luggage directly onto the model, giving me one less bit of assembly to do.

This second model is a dwarf semaphore. Again, it came from an unlikely source; a local railroad history magazine (the California Trolley and Rail Corporation's June 1998 Clearboard) had a picture of a 1932 exhibition trip for the London, Midland, and Scottish's "Royal Scot" locomotive as it passed through San Jose. That photo shows the Fourth Street tower, the tracks approaching the station, and a bunch of other detail along the right of way.

My eye immediately noticed a few small dwarf signals protecting the main line track down Fourth Street and towards San Jose. Those little signals were both cute and prototypical for my Market Street shelf layout. I did a few Google searches to try to identify a manufacturer, and I'm guessing the dwarf was a General Railway Signal product. A bit more work, and I'd sketched out a decent signal for my shelf layout.

(Ooops, just noticed that I got distracted during my research, and ended up making a model of one of the dwarf signals I'd seen on a semaphores site, rather than the neat upper-quadrant semaphore shown in the photo. That was silly. I suspect the inaccurate semaphore's going on the layout anyway.)

Now, making each of these models wasn't exactly trivial; the angles and curves on each meant the computer model alone took a couple hours. I also certainly could have done these models with traditional materials such as styrene and card stock. However, for an active layout, quickly drawing and printing details lets me quickly add all the clutter that is in realistic scenes. Better yet, it's easy to make new ones. If one of my operators accidentally bashes a model, all I need to do is print and paint a new copy.