Last month Gizmodo posted a story about the “toxic particles spewed by 3D printers.” It was based on two articles from the last 18 months in the journal Aerosol Science and Technology (this one and this one). These are not the first articles on the subject but are carefully done and use standard protocols established earlier for documenting emissions from laser printers. The articles describe tests of six different printers (all FDM printers in which molten plastic is extruded as a bead) and 10 different filament types (all ABS, PLA, or nylon).

It is no secret that you can smell a 3D printer. Many people who use ABS filament vent their printer because the fumes are unpleasant. PLA fumes are generally not bothersome and I work next to my printer and never think about the PLA fumes. I was interested to learn more about what I might be breathing, and what thousands of school kids might be breathing when there is a 3D printer in their school or classroom. The journal Aerosol Science and Technology posted a video abstract of one of the articles which is below.

Video Abstract - Characterization of particle emissions from consumer fused deposition modeling 3D printers from Taylor & Francis on Vimeo.

Previous studies indicate that printing with ABS emits styrene and ethylbenzene, and PLA emits lactide and methyl-methacrylate. The recent articles did not determine what compounds are being emitted from 3D printers. The focus was on the size and number of generated particles whatever they might be (so the title of the Gizmodo article and some of the content was blatantly misleading). They measured the rate at which particles of different size were generated before, during, and after printing operations which took about one to seven hours. The particles measured were all under 1 micron (1µm) and smaller than many of the particles generally included in air pollution studies (up to 10 µm) or measured well by inexpensive laser-scattering dust sensors.

Figure 1. From the article: “Schematic of particle formation, growth, and loss processes. NPF is new particle formation resulting from nucleation of emitted semi-volatile vapors.”

To explain their results, the researchers developed a model of what happens in the air around a 3D printer nozzle (Figure 1). Hot molten plastic is exposed during 3D printing as it exits the hot metal nozzle. In the seconds before the plastic cools, molecules of various volatile and semi-volatile compounds are released into the air. There are enough of these molecules that they collide with each other and form particles. Over time, condensation (gas molecules combine to form a droplet) and coagulation (droplets combine) cause particle size to increase. Particle numbers increase quickly at first then stabilize or decrease. One or two orders of magnitude more particles were generated by printing ABS vs. PLA (in part because ABS is printed hotter than PLA), but there was high variability in these results (Figure 2), and other studies have come to different conclusions.

Figure 2. From the article.

These articles inspired me to set up my new particle and gas sensor near the 3D printer. The PMS7003 particulate sensor reports particles as small as the 300nm (0.3µm) size class which is the upper range of particles reported in the articles. The CCS811 gas sensor reports an estimate of the concentration of total volatile organic compounds (VOCs, in ppb). It has a heated metal plate which changes its resistance when certain types of airborne molecules contact it. It does not necessarily discriminate perfectly between VOCs and many other molecules, but it’s all you get for $15.00.

Figure 3. The CSS811 gas sensor module. This is an I2C module which also includes an HDC1080 temperature and humidity sensor (left).

Figure 4. The CSS811 gas sensor (lower right) and the Plantower PMS7003 particle sensor are controlled by a Nano Data Logger which saves new data every minute.

I suspended two replicate data loggers over a MakerBot Replicator 1 Dual 3D printer using the arm of a floor lamp. That placed the sensors directly above the printer and prevented printer vibrations from affecting the sensors. Each logger operated for nine consecutive days and made 13,000 writes (1 per minute) to the microSD card with date, time, and 15 readings from the sensors. One to three print jobs were run each day for five days, and no print jobs were run for the last four days. Printing was with red PLA filament at 205°C on a heated build plate (50°C) and each print lasted about three to four hours.

Figure 5. The two replicate data loggers secure in their lamp shade above the 3D printer.

Figure 6. PM 2.5 results for two replicate PMS7003 sensors for nine days. Sensor #3 was on top of sensor #1. The 3D printer was operating during the first five days only. The two particle sensors responded very similarly. Sensor #3 (blue) produced readings that were slightly lower than sensor #1 (red). Each dot is the mean of 60 sensor readings taken about once per second and saved about once per minute.

Figure 7. VOC results for two replicate CSS811 gas sensors for nine days. Both sensors responded to the same stimuli although the response was not identical, especially during days when the 3D printer was operating (the first five days). Each point is the mean of 60 sensor readings taken about once per second and saved about once per minute.

Figure 8. VOC (blue) and PM 2.5 (brown) results for the 4.5 days when the 3D printer was not operating. Total VOC readings were higher during periods of each day when the room heater was on (red bars). The heater has a fan to circulate air. The heater being on is highly correlated with me being in the room, and the room is always vacant when the heater is off. In three of the four full days, there is a peak of VOC during the evening period (after I eat dinner and then return to the room). Just sayin’, several things are going on here. Note that the PM 2.5 readings are all very low and the variation might be noise. Each point is the mean of 60 sensor readings taken about once per second and saved about once per minute.

Figure 9. VOC (blue) and PM 2.5 (brown) results for the five days when the 3D printer was operating. One to three peaks in both VOC and PM 2.5 are recorded each day. These peaks are perfectly correlated with times that Aquanet hairspray was applied to the glass build plate used in the 3D printer. It took about 3 to 5 hours for VOC and PM 2.5 to return to “ambient” levels after each spray application. These peaks obscure any evidence of response to the operation of the 3D printer itself with one exception (A.) when a print was started without preparing a build plate (it had been done the previous day). Each point is the mean of 60 sensor readings taken about once per second and saved about once per minute.

Figure 10. Glass plates are clipped to the heated build plate of the 3D printer and the parts are built directly on the glass. To ensure the proper adhesion, the glass plates are lightly coated with hairspray before each print.

Figure 11. Aquanet Extra Super Hold unscented hairspray is widely regarded in the prosumer 3D printing world as a very good solution to the build plate adhesion issue. Except by those who have yet to see the light and still swear by blue painters’ tape or UHU glue stick but really, get with it. Also, there is no substituting other brands--take your Garnier Fructis out of here.

Before each print, I put a fresh layer of hairspray on a glass plate. This was done about 1.5 m (5 feet) from the sensors. I assumed the sensors would notice this so I recorded exactly when it happened and I then waited 10 to 70 minutes before starting the print. To my surprise, it took longer than that for the signal of the hairspray to fade, so any signal of the 3D printer was mostly obscured. This series of observations will have to be repeated with the hairspray operation done in another room so the effect of the printer can be evaluated.

During the first day of this test a glass plate was sprayed with Aquanet four times (Figure 9). The four peaks of PM 2.5 on that day are higher than any peak on the four subsequent days. The difference is because I emptied a can of Aquanet late on the first day and started using a new can on the second day. The almost empty old can sprayed a weak, fine mist but the new can sprayed a strong, wet, splurt.

Both the gas sensor and particle sensor responded strongly to the hairspray (Figure 9). The response was quick and lasted three or more hours. Although the contribution of the 3D printer to VOC and PM 2.5 was obscured, it was obviously much smaller than the contribution of the hairspray.

I hope the health effects of breathing hairspray particles are not serious, but the label on the can is not reassuring. The essential ingredient in this type of hairspray is a water-soluble acrylic. Molten PLA adheres well to the acrylic layer applied to the glass plate. That layer seems harmless because it washes off easily with water (from the plate or your hair). But the aerosol spray-can contains stuff like ether, propanol, sodium benzoate, and cyclohexylamine (Figure 12). How much worse can the emissions from the printer be, especially when they are substantially lower?

Figure 12. The scary looking ingredients in Aquanet.