Table 1 summarizes the five desktop vat polymerization printers evaluated in this study. Types of vat polymerization technology include, but are not limited to, stereolithography (SLA) and digital light processing (DLP). SLA printers scan a laser beam across the print area to selectively cure resin in a vat. This process cures the resin as series of points and rounded lines to build objects layer-by-layer. DLP printers use a high-resolution projector to flash black and white image slices of each object layer across the entire vat surface at once. The projector is a digital screen that forms white areas of the projected image made of square pixels that are cured using multi-wavelength light from a lamp to build an object. In this study, three were SLA type and two were DLP type. All printers used gray liquid feedstock resins. Elemental content from photoinitiators in the bulk feedstock resins and pieces of printed objects was determined by microwave assisted acid extraction following EPA Method 3051A, and subsequent analysis by inductively coupled plasma mass spectroscopy (ICP-MS). Briefly, 0.1 g of bulk resin or shavings from a printed object was placed into a pre-cleaned 50mL Teflon digestion vessel along with 9mL of trace metal grade nitric acid and 3mL of trace metal grade hydrochloric acid. The vessels were capped and placed in a microwave (MARS 6 Xpress, CEM, Matthews, NC). The samples were heated to 175 °C and extracted for 10 min at pressure. Samples were cooled and diluted to a final volume of 50 mL. A single sample of bulk liquid resin was digested whereas triplicate samples obtained from different locations of a printed object were digested to account for potential spatial variability. Elemental analysis was conducted on an ICP-MS (Agilent 7900, Agilent Technologies, Santa Clara, CA). Sample analyses were conducted in both no gas and collision mode using helium (5.0 mL/min).

To characterize the properties of particles in the feedstock, resin was drop-cast onto 3-μm pore size track-etched polycarbonate (TEPC) filters, allowed to air dry, and analyzed using a field emission scanning electron microscope (FE-SEM, S-4800, Hitachi, Tokyo, Japan) to determine particle morphology and energy dispersive X-ray analysis (EDX, Quantax, Bruker Scientific Instruments, Berlin, Germany) to identify elemental constituents. Additionally, pieces of printed objects were mounted on aluminum sample stubs using double-sided carbon tape and evaluated using FE-SEM-EDX.

All printing was performed in a temperature (21 ± 1°C) and humidity (50 ± 5%) controlled 12.85 m³ stainless steel chamber meeting international guidelines for emissions testing.^[25] Using SF₆, the calculated mixing level (η) was 92% (a level above 80% is considered satisfactory), the leak rate was 0.024 air changes per hour, and the air exchange rate was 1 per hour.^[26] Air entering the chamber was passed through a carbon filter and high efficiency particulate air (HEPA) filter to remove organic vapors and particles, respectively.

Chamber air was sampled for at least 30 min prior to printing to establish background using the real-time instrumentation and time-integrated sampling approaches for particles and organic chemicals (described below). All print jobs were of an artifact from the National Institute of Standards and Technology (NIST).^[27] The artifact has a 10 cm × 10 cm square base and is 1 cm thick with various features to test printer performance, including fine features (depressions, holes, and protruding pins) as well as larger features (holes, protruding pins, elevated staircases and a ramp). The same artifact was printed five times per printer at 50% scale of full size (5 cm × 5 cm × 1 cm). Print times ranged from 78–192 min, depending on the machine.

Real-time instruments and time-integrated sampling approaches were used to measure particle and volatile organic chemical levels in the chamber (Figure S1 in the Supplemental File). Separate samples were collected during the background phase and printing phase (including the decay period, which lasted for three chamber air exchanges after completion of printing).

Total particle number concentration from 20 nm to 1 μm (Model 8525, P-Trak, TSI Inc., Shoreview, MN) and size distribution of particles from 5.6–560 nm (Model 3091, fast mobility particle sizer [FMPS], TSI Inc.) were measured inside the chamber. The P-trak counts ultrafine and sub-micron particles by sampling air at 0.1 L/min (0.7 L/min total flow). As particles are aspirated into the instrument, they are passed through a saturated isopropyl alcohol vapor, which condenses on the particles and grows their size large enough to scatter laser light and be detected by the instrument. Particle counts were logged at a 10-sec interval. The FMPS samples air at 10 L/min. As particle-laden air is aspirated into the instrument, the particles are positively charged using a corona charger. The charged particles are detected using multiple electrometers and enumerated based on their mobility size in 32 size channels. Particle counts and sizes were logged at a 1-sec interval. Airborne particles were collected on TEPC filters during printing by drawing chamber air through filters at up to 10 L/min using a calibrated sampling pump (SG10–2, GSA Messgerätebau GmbH, Ratingen, Germany). All filters were analyzed using FE-SEM-EDX.

Total volatile organic compound (TVOC) concentration was monitored using a real-time photoionization detector (PID) (RAE Systems, San Jose, CA or Ion Science Inc., Stafford, TX) with a 10.6 eV lamp. This instrument uses an ultraviolet light to ionize organic molecules (with an ionization potential below 10.6 eV); the positively charged ions are detected as changes in electrical current by the instrument. Measurements were logged on a 1-sec basis. The potential for chemical emissions from the printers to react with ozone to form secondary chemical compounds such as carbonyl compounds was evaluated by drawing air through liquid impingers at 4 L/min for three of the five trials per printer, followed by derivatization and analysis using GC-MS.^[28]

Particle emission rates (ER) were calculated from the real-time measurement data collected during printing using a mass-balance model^[8]. From the real-time particle concentration data logged by the P-Trak or FMPS, the ER was calculated using Equation (1): (1)ER(t)=VΔt[C(t+Δt)−C(t)+R⋅C(mean)⋅Δt], where t = time, sec.

Δt = time difference between two successive data points, sec;

C(t + Δt) = particle number concentration at (t + Δt), particles/cm³;

C(t) = particle number concentration at (t), particles/cm³;

V = chamber volume, cm³;

R = air exchange rate in the chamber, hr⁻¹

C(mean) = mean particle concentration between (t) and (t + Δt), particles/cm³.

Emission rates for TVOCs were calculated in accordance with RAL-UZ-205.^[29] ER values were calculated from the beginning of the print phase (from start to end of print job) until one air exchange occurred in the post-operating phase using Equation (2): (2)ER=cs⋅nd2⋅vc⋅tg⋅SERB⋅nd⋅tgnd⋅td−e−nd(tg−td)+e−nd⋅tg, where c_s = concentration during printing and post-decay phases, n_d = air exchange rate (h⁻¹) during printing and post-decay phases, v_c = chamber volume (m³), t_g = total sampling time (h), SER_B = specific emission rate during background phase (μg/h), and t_d = printing time (h).

To account for differences in print time among printers, the ER values were converted to total particle number or TVOC mass and normalized to the mass of resin used to build an object (determined gravimetrically using a calibrated microbalance) to calculate yield (emission per unit mass printed).^[4]

Temporal changes in particle size distributions (including median and variance) from real-time FMPS measurements were estimated using a two-level Bayesian model with a Markov chain Monte Carlo algorithm developed by Klein Entink et al.^[30] We used 1-min averages of FMPS real-time measurements for two emission tests as examples. A run of the Pegasus Touch SLA printer had 315 one-minute average measurements, of which, the 1^st to 35^th 1-min average measurements were background, 36^th to 191^st were collected during the printing phase, and the 192^nd to 315^th were the post-printing phase. A run of the M-One DLP printer had 214 measurements, of which, the 1^st to 42^nd, 43^rd to 147^th, and 148^th to 214^th 1-min average measurements were collected during the background, printing, and post-printing phases, respectively. Analyses were conducted in R 3.3.1 using the “NanoPSDA” package (R Foundation for Statistical Computing, Vienna, Austria).^[31,32]

Mean yield and particle size distribution values were compared using one-way Analysis of Variance (ANOVA) models. Specific pairwise differences between printers or types (SLA or DLP) were compared using Tukey’s tests. A significance level of α = 0.05 was used for all comparisons. Statistics were computed using JMP software (version 13, SAS Institute Inc., Cary, NC).

The fraction of emitted particles that could deposit in the lung were calculated using the Multiple-Path Particle Dosimetry model (MPPD, v3.04, ARA).^[33] Dosimetry estimates were made using the stochastic morphometric lung model with 60^th percentile size to represent the majority of the general human population. This model uses morphometric measurements of the tracheobronchial tree (length, diameter, branching angle, cross-sectional area, etc.) supplemented with theoretical reconstructions of the alveolar region to predict particle deposition via the primary mechanisms of impaction, sedimentation, and diffusion. Model parameters were as follows: uniformly expanding flow, upright body position, oronasal-mouth breather with 0.5 inspiratory fraction and no pause fraction. International Commission for Radiological Protection (ICRP) reference human default values breathing parameters for a Caucasian adult male at light level of activity were used: functional residual capacity (3,300 mL), upper respiratory tract volume of 50 mL, tidal volume (1,000 mL), and breath frequency (20/min).^[34]

Table 2 summarizes the average particle emission yield values for each printer (individual yield values and size distribution measurements for each trial for each printer are provided in the Supplemental File, Tables S1–S4). ANOVA analysis indicated a significant difference in mean particle number yield values (P-Trak data) among printers (p < 0.05). Pairwise comparison revealed that the yield for the Titan 1 was significantly higher than all other printers (p < 0.05). The mean particle emission yield was higher for DLP type printers compared to SLA type printers (p < 0.05).

We used all FMPS size channels (5.6–560 nm) to calculate yield values (see Table 2). ANOVA analysis indicated a significant difference in mean number yield among printers (p < 0.05). Pairwise comparison revealed that the yield for the Titan 1 printer was greater than for all other printers and that the yield for the Nobel 1.0A printer was significantly higher than for the Pegasus Touch printer (p < 0.05). The mean yield for particles with sizes 5.6–560 nm was significantly higher for DLP-type printers compared to SLA type printers (p < 0.05).

Table 2 also includes the average geometric mean (GM) and geometric standard deviation (GSD) of particle sizes measured using the FMPS during printing. There were no statistical differences in the average GM mobility diameter for the Form 1+ (41.3 nm), Pegasus Touch (41.1 nm), and Nobel 1.0A (45.1 nm) printers. The average GM diameter of particles emitted by the M-One printer (28.8 nm) was significantly larger than the Titan 1 printer (15.3 nm); (p < 0.05). The mean mobility size for DLP-type printers (22.0 nm) was significantly smaller than the mean size for the SLA printers (42.5 nm) (p < 0.05).

The results of model-fitted particle size distributions using real-time measurements are shown in Figure 1. This logarithmic regression model revealed that particle size changes among the background, printing, and post-printing phases. For the Pegasus Touch printer example, the estimated GM of the particle size distributions decreased from exp(4.04) = 56.8 nm in the background period to exp(3.82) = 45.6 nm in the printing period, and then to exp(3.79) = 44.3 nm in the post-print decay period (Figure 1a). However, the standard deviation gets larger over these three time phases (Figure 1b). For the M-One printer example, the estimated GM particle size distributions also decreased from the background period (33.4 nm) to the printing period (21.3 nm), but slightly rebounded in the decay period (24.3 nm) (Figure 1c). The standard deviation decreased over these three time phases (Figure 1d).

Analysis of bulk resins, emitted particles, and built objects indicates that vat polymerization chemistry is highly complex. Table 3 summarizes the identities of elements present throughout the vat polymerization AM process. Nine different elements were identified in the bulk feedstock resins, 14 different elements were identified in the airborne particles, and 19 elements of health interest were found amongst the printed object samples.

Supplemental Table S5 summarizes the ICP-MS measurements for concentrations of elements in the bulk feedstock resins. Aluminum (Al), chromium (Cr), phosphorous (P), tin (Sn), and titanium (Ti) were detected in all resins, though concentrations varied among brands. Antimony (Sb), Ba, copper (Cu), and lead (Pb) were quantifiable in some but not all of the bulk resins. In addition to ICP analysis of the liquid resin, samples of dried resin were inspected using FE-SEM-EDX. Analysis identified discrete particles and cluster particles (Supplemental Figure S2) that were composed of single and/or multiple elements (Supplemental Figure S3).

Figure 2 shows the morphology and Figure 3 shows the corresponding elemental composition of exemplar airborne particles emitted during operation of the vat polymerization printers. Regardless of the resin brand or printer type all filter samples contained branch chain aggregates of nanoscale spherical primary particles which, from EDX analysis, were composed of carbon and oxygen (Figures 2a and 3a). In addition to these branch chain particles, each printer emitted submicron-scale compact particles with variable elemental composition (Figures 2b–2f, 3b–3f). Only iron (Fe) was identified in the aerosol emissions from all of the printers. Particles emitted by the Form 1+ printer included single-constituent calcium (Ca), Si, and Fe particles, multi-constituent particles containing Fe and nickel (Ni), and particles containing Fe, Ni, and Cr. The M-One printer emitted single-constituent particles that contained Fe, Si, Ti, or P as well as multi-constituent particles. Particles emitted by the Pegasus Touch printer included single-constituent Al particles as well as multi-constituent particles containing Ca, Fe, sodium (Na), sulfur (S), and/or Si. During operation of the Nobel 1.0A printer, multi-constituent particles were emitted that included Cr, Ni, and zinc (Zn). For the Titan 1 printer, emitted particles included Fe or Si particles as well as multi-constituent particles, some of which contained Ba.

Supplemental Table S5 summarizes the ICP-MS measurements for concentrations of elements in the printed objects. ICP analysis of printed objects identified the same elements as in the bulk feedstock resins, as well as several additional metals: arsenic (As), cadmium (Cd), Ca, cobalt (Co), Fe, manganese (Mn), molybdenum (Mo), Ni, vanadium (V), and Zn. With the exception of Sb and P, elements were more concentrated in the printed objects than in the corresponding bulk resins. Table 3 shows that ICP analysis identified a total of 19 different elements in the printed objects; all but Ba were common among samples. Although Formlabs resin was used in both the M-One and Form 1+ printers, concentrations of five elements (Cd, Cr, Co, Ni, and V) were higher in objects printed with the M-One compared with the Form 1+ printer.

Table 4 summarizes the regional deposition fractions for the respiratory tract for particles emitted by each type of vat polymerization printer. Deposition fractions were calculated using the printer-specific GM mobility diameters for particles emitted during printing. Among printers, deposition values were as follows: head region (3–7%), tracheobronchial region (9–21%), and pulmonary region (15–38%). Pulmonary deposition values for the DLP printers were about twice as great as the SLA printers. These estimates indicate potential for high numbers of particles containing various metals to deposit throughout the respiratory tract.

Table 2 summarizes the calculated TVOC emission yield values during printing. Pairwise comparisons revealed that the TVOC yield for the M-One was significantly higher than all other printers (p < 0.05) and the yield for the Titan 1 printer was significantly higher than for the Form1+, Pegasus Touch, and Nobel 1.0A printers (p < 0.05). The mean TVOC emission yield was significantly higher for DLP type printers compared to SLA printers (p < 0.05).

As summarized in Table S6, acetone, benzaldehyde, and 4-oxopentanal (4-OPA) were detectable during the printing and post-printing phases for specific combinations of printer and feedstock resin. There was no clear difference in carbonyl concentrations between types of printers. Acetone concentrations ranged from 0.7–18.0 ppb and from 0.9–26.0 ppb during the print and post-printing phases, respectively. The highest acetone concentrations were observed from the Nobel 1.0A printer using XYZPrinting resin and were 18.1 and 26.0 ppb from the print and post-print sampling, respectively. Benzaldehyde concentrations ranged from 1.0–12.3 ppb and from 0.9–12.7 ppb during the print and post-printing phases, respectively. The highest benzaldehyde concentrations were observed from the M-One printer using Formlabs resin and were 12.3 and 12.7 ppb from the print and post-printing phases, respectively. 4-OPA concentrations were below the limit of detection, except for the Nobel 1.0A printer using XYZPrinting resin during post-print sampling, where a concentration of 0.07 ppb was observed.