Every day, we come in contact with objects that are injection molded – one of the cornerstone technologies that define our modern world. Injection molding allows designers to create extremely complicated parts with relatively low part cost after a (usually expensive) initial investment in an injection mold.
Injection molds are often closely guarded trade secrets – Legos is said to bury worn-out molds in the concrete of its new buildings. However, since we are an open sourced project, working closely with our manufacturing parters, I am able to give you a behind-the-scenes look at how injection molds are built.
Here’s the facility at Western Tool and Mold - my manufacturing partner for all the plastic brackets and components. It is run by an American gentlemen by the name of Collin Wilkerson. Their company specializes in aerospace injection mold construction, as well as specialized engineering resins, such as the EMS Grivory material used in the OpenBeam brackets. I’ve worked with Collin and his team for over three years with two different companies and would trust them with my most demanding projects.
The injection molding process starts with CAD data, transmitted electronically via email. Often, a drawing package will accompany the 3D data. Younger engineers tend to neglect the importance of a good engineering print, but this is a formality that cannot be overlooked. For one, the drawing is part of a legal contract between the buyer and the supplier. It governs everything from material selection, color and texture information, to packaging material, inspection standards, and delivery terms. (Oh, you didn’t want all your parts to be stuffed into a box in newspaper? You should have put something in your drawing…) The drawing also functions as a tool to convey features not in the cad model – such as locations for the gate (where the plastic is injected into the mold), allowable positions for the ejector pins (to remove the part from the mold), and cosmetic requirements on the part (the bottom of a toaster has less of a cosmetic requirement than the front display housing on an iMac).
An injection mold can essentially be thought of as two big blocks of steel housing various mechanical components to allow cooling and heating of the mold, the “ejector pins” to remove the part from the mold after the mold opens, guide and alignment features to keep the two halves of the mold aligned properly with each other, as well as the actual steel cavity into which molten plastic is injected under high pressure to form the part. After the mold maker receives the CAD files, drawings, and other requirements, a tooling engineer takes over and starts the mold design process. Tooling design is a very specialized art, and the folks that work on these are truly master craftsmen in their own right. Part geometry, tolerance requirements, resin selected (and the shrink rates of the resin as it cools) are all considerations that the tooling engineer must take into account when designing the mold. Today, advanced simulation software can help simulate the flow of hot plastic resin under pressure into the mold to help predict and prevent issues such as part warpage, sink marks on part, and filling problems with a mold. The tooling engineer will then create computerized programs to drive the robotic machines involved in cutting the steel pieces into all the components of a mold.
The majority of the metal cutting takes place on a computer numerically controlled (CNC) milling machine. At Western Tools, this work is predominately done by Mori Seki machining centers.
These machines are so accurate, they are subjected to export restrictions by the Japanese government under International Traffic of Arms Regulations (ITAR). The reason being is that these machines have the accuracy required for machining the complex 3D curvatures required by modern nuclear weapons.
Before each program segment for the CNC machine is executed, a laser scanning probe verifies the tool geometry and compensates for wear and tear of the tool. After each program segment, the laser reinspects the tool for breakage and alerts the operator if a breakage occurs. A hot-spare for each tool exists in the tool changer magazine – so without human intervention, the machine can rerun that particular segment of program code. This allows for “lights out” operation and frees up a highly trained tool maker from “babysitting” a machine.
Not all parts can be cut by a CNC mill. sharp inside corners and awkward geometry may make material removal by an end mill difficult, if not impossible. In this situation, a different trick is used.
This is an EDM machine – and any mold shop will have a few of them. It uses a high voltage electric discharge to burn away metal. The nice thing about EDM, is that it doesn’t really care how hard the material is. Normally, your cutting tool has to be harder than the material you are cutting – and for hardened tool steel, only specialized ceramic materials called carbides are hard enough to cut them. But here, a relatively soft copper rod is slicing through a piece of tool steel with relative ease.
EDM machine comes in 2 shapes – sink EDM and wire EDM. In the example above – a sink EDM is using an electrode to make a straight cut through the steel. The cross section cut takes on the exact same shape as the electrode.
A wire EDM machine, on the other hand, allows for free-form cutting. In this machine the wire is threaded through automatically, and the top and bottom position of the wire can be individually driven. The wire itself acts as the cutter, or an impossibly thin band saw. (Imagine a hot-wire foam cutting machine – same principle, except this is cutting hardened steel. Keyways for slides, ejector pins and coolant lines can be cut this way. When the machine is done with one hole, it can automatically cut the wire, move to the next position and rethread the wire through the hole – again allowing for “lights out” manufacturing.
For the finishing touches, part numbers, ISO material identification codes and recycling information are engraved into the part as part of good engineering practice. Finally, the mold is textured by a chemical etching process. This does two things; it makes the part less glossy and plastic looking, and in the case of parts that are glued / labeled, it also increases the surface roughness of the plastic to allow for a better adhesion.
Simultaneously with all the cutting, polishing and assembly work, the operations team is hard at work, sourcing the plastic resins. Often, a color chip is provided to the molder and color mixing and matching occurs. Larger annual usage (one metric ton of resin and above) may have custom compounded color resin to ensure lot-to-lot consistency. Saudi Basic Industres (Sabic, formerly GE Plastics), for example, have patented formulation for Apple products. This is how every Macbook power supply brick and every Magic Mouse have the same glossy white look. For smaller players like myself, lot codes are important if we care about color matching, due to the tendency for color to vary from batch to batch.
This ought to give you an appreciation of how much work goes into the fabrication of an injection mold. Again, it is no small feat to finish a mold like this in five weeks, from CAD file upload to first shots. It takes an incredibly well tuned organization (and fantastic client-vendor relationships) to make this happen.
Here are closeup pictures of the OpenBeam Mold cavities:
You can see the ejector pin guides here, corresponding to the CAD model screenshot:
And finally, here are the first shots off the molds!
(Note the Tam Labs Part number (TL625-000100), revision level number (-002), material code (PA66+GF30) and ISO recycling code molded into the recess of the part as good engineering practice)
At work, virtually all my coworker have product development in their “DNA”. We are all creative types, who takes lots of pride in bringing a product from an idea to life. And so consequently, whenever we read about some new widget or tool, we’ve always tried to convince our boss that it would benefit the company tremendously, if only if we have the latest and greatest Super Machine 2000.
Our boss, who’s rarely wrong, always tell us that as design engineers, our time is the most valuable spent designing. “Whenever we need something, and need something quick, we just toss money at someone and have them bang it out and put it in a Fedex overnight box”.
My buddy Dave’s grandparents owns a laser engraver. They own a trophy engraving shop and laundromat down in Renton. It’s really wierd to think of a sweet old lady at the controls of an Epilog 20W CO2 laser system… but she doesreally good work. So the next time you need something engraved, check out the Puhich Dry Cleaners in Renton on 319 Main Ave. South.
I know what a 20W CO2 laser can do; it can do a lot more than mark plastic and burn through anodize layer on aluminum. There was a discussion thread on making lens cap holders, so I drew one up in Soldiworks real quick:
Here’s the test run in paper. So far so good, right?
Unfortunately, the laser engraver is running on Windows 98. It requires a firmware update before it can talk to anything past Win98, and there is always a risk of bricking a machine doing a firmware update. So we are stuck with a computer that works – abet a very slow one, with a parallel port printer connection. (I bet some of the folks I know had never seen one… they went the way of the dodo after USB became popular).
Complicating the problem is that the printer driver runs as a Corel Draw plugin. Corel Draw 8, to be exact. And even the earliest DXF that Solidworks can save the file in, the simple fillets on the drawings don’t quite come through – let alone the more complicated splines and polylines.
The example above worked okay, because the laser cut it out as a *Raster* art, instead of a vector art. But cutting in raster mode drastically drops the laser’s power output. And it’s not like you can run multiple passes over the same piece of PTFE either – the slow heat transfer of doing so just warps the plastic – and the results looked like someone tried cutting the material with a dull butter knife.
Obviously something like this, out of 3mm PMMA, will be a bit out of the question:
So my options are:
1) Try to mitigate the risk of the current laser cutter’s firmware upgrade (maybe see if I can do a hardware replacement of the logic board, upgrade the computer to something snazzy, then retry the Solidworks -> laser cutter workflow.
2) Pay someone like Pololu online to do the laser cutting for me. Essentially, someone else will be eating part of my lunch if these products go on sale. Might be okay if there’s only a few parts, but I’ll have to rethink the design a little bit.
Turns out, RedWolf airsoft out of Hong Kong will happily sell me a 30mm silencer for about $US10.00. Aluminum barrel, both ends with a machined aluminum plug. They even put a 14mm CW thread on one end and 14mm CCW threads on the other, so out of the box, the dang thing will fit on just about every single airsoft gun out there.
In Hong Kong, we have a saying that “the flour costs more than the bread”. The term originated from the housing bubble days where the value of the land gets bid up so quickly that older apartment buildings prices are being outstripped by the land value of neighbouring lots, but it also applies to engineering and business where by some form of competitive advantage (and economies of scale), someone can build a product cheaper than you can even start sourcing raw materials.
Got a cool little bookmark for all my troubles with the laser cutter though…
This XKCD cartoon sums it up about marketing oneself:
Quick status update on the bullet flight sensor. This is heading into systems integration testing next, where I’ll be firing up each section of the circuit and making sure it all works. Missing is the break beam sensor that I put a air rifle round through by accident
Note the “unusual” arrangement with the pocket wizard. The “hot shoe adapter” is actually plugged into the sensor to simulate a camera’s hotshoe firing the pocket wizard.
Successful night at Tam Labs tonight!
Tonight, the goal was to test the breadboarded prototype of the bullet flight sensor’s electronics. Remember – I am a mechanical engineer; this is a completely new foray into the world of electronics for me, aside from some simple “hook a solid state relay to a microprocessor and bit bang some code to turn on the rice cooker” projects. So even though this may seem like kindergarden EE stuff, it’s a fairly big leap for me, design wise; I’m no longer relying on the ability to clobber code and instead using discrete logic ICs and doing actual calculations and setting RC time constants, etc.
We begin with the breadboarded model:
And our setup in the lab:
On the bench is a trusty oscilloscope to look at the signals at different lines, a DC power supply set to 5V, 100mA current limit, and a signal generator. The signal generator won’t be used for this project here.
First, I verified that the *new* sensor is working – the last one had a round put through it by accident:
Next, I verified that the 555IC is getting the power that it needs. Turns out that the power rails aren’t fully connected all the way. A bit of poking with an ohm-meter fixed that. Now I am ready to insert my test points:
And trip the break-beam sensor, with my gimpy fingers:
Orange line, or Ch1, is my sensor’s output. It goes from High to Low when the beam is broken. The turquoise line, or Ch2, is my 555′s trigger output. It goes from low to high when the input pulse is received. That’s a VERY promising sign.
My fat butter fingers can only move so fast through a 10mm opening, so the event scrolls off the oscilloscope’s screen. I tried dropping a small machine screw through the opening, but that actually prove to be much harder than expected (don’t laugh!). Frustrated, I finally came up with the following idea:
By flexing the rubber ducky antenna on one of my pocket wizards and getting it to spring through the break-beam sensor gap, I can generate a quick enough blip from the sensor:
Each major division is 5 milli-second on this setting, so the rubber ducky antenna is only in the beam’s path for about 10mS. My fingers can’t move *that* fast, for sure
Repeating the test again a few times, got me the same result:
Note that irregardless of the pulse length from the sensor, the 555′s output always sits at about 35ms. This is a litte bit off from the design goal of 40ms (1/250 second shutter speed, or sync speed on a 1.6x crop camera), but close enough for government work. I attribute the difference in component value tolerances on setting the RC constant.
Now the final test – does the output from the 555 trip the SCR to fire the strobe and pocket wizard?
(I selected an SCR instead of a cheaper / more common transistor. The SCR is rated to 400V, so even an older, high voltage “digital camera killer” flash will work on this sensor. )
*drum roll please*
Turns out the same bug that bit me on the 555 timer bit me again. The top and bottom half of the power bus on this breadboard is not connected, and the SCR wasn’t grounded properly because of that. Now, plugging in a pocket wizard, this is what I get (with Ch2 now monitoring the anode of the SCR):
Interesting, it seems to add a bit of noise to the sensor output line. But the characteristic beep of the PW firing can be heard as the beam is broken. (Note that the sync voltage of the pocket wizard is only 3V or so).
Plugging in the 580EXii:
Again, some electronic noise on the sensor line, but we got what we need out of it – the clean voltage drop that triggers the monostable multivibrator.
And here’s the happy camera dork with his new toy (click link for video:)
Now that the circuit is verified working, I am okay with releasing the resources to order the acrylic for laser cutting to form the chassis, as well as starting PCB layout. Stay tuned…
Now here’s something you don’t see everyday: an SLA machine in action.
The machine in question is a 3D Systems Viper high resolution unit. I used a long exposure on a tripod to capture the actual laser beam tracing the part; the laser is a diode pumped solid state Nd:YVO4 100mW laser. For those of you unfamiliar with how SLA process works, it works like this:
1) A computer solid model is sliced into thin slices, between 0.002″ to 0.003″ thick. All the newer machines are 0.003″ or better in Z-resolution.
2) A laser then traces the cross section in a vat of photopolymer. Okay, by photo-polymer, I mean, a vat of very expensive goo – where the laser touches the goo, the goo turns into plastic. How expensive? I’ve heard that when it first came onto the market, resins were about $5000 – $6000 per gallon. They have since came down in price, but suffice to say it’s a very expensive proposition.
3) After each layer is formed, the platform in the tank drops by a build layer thickness and the goo flows over the part, readying the next layer to be drawn.
4) At the end of the build the platform gets raised again and the goo drains off the part, leaving you with a solid object.
These are very expensive machines – starting price of one of these is in the quarter-million dollar range, and the laser only lasts so many hours before it has to be replaced. Needless to say, unless you’re keeping the machine *very* busy, just the depreciation would eat you alive.