Prevention through Design: Insights from computational fluid dynamics modeling to predict exposure to ultrafine particles from 3D printing

Abstract

Fused filament fabrication (FFF) 3D printers are increasingly used in industrial, academic, military, and residential sectors, yet their emissions and associated user exposure scenarios are not fully described. Characterization of potential user exposure and environmental releases requires robust investigation. During operation, common FFF 3D printers emit varying amounts of ultrafine particles (UFPs) depending upon feedstock material and operation procedures. Volatile organic compounds associated with these emissions exhibit distinct odors; however, the UFP portion is largely imperceptible by humans. This investigation presents straightforward computational modeling as well as experimental validation to provide actionable insights for proactive design of lower exposure spaces where 3D printers may be used. Specifically, data suggest that forced clean airflows may create lower exposure spaces, and that computational modeling might be employed to predict these spaces with reasonable accuracy to assist with room design. The configuration and positioning of room air ventilation diffusers may be a key factor identifying lower exposure spaces. A workflow of measuring emissions during a printing process in an ANSI/CAN/UL 2904 environmental chamber was used to provide data for computational fluid dynamics (CFD) modeling of a 6 m² room. Measurements of the particle concentrations in a Class 1000 clean room of identical geometry were found to pass the Hanna test for agreement between model and experimental data, validating the findings.

Introduction

Additive manufacturing technologies, including household 3D printers, continue to experience tremendous market growth. A recent market analysis by the Congressional Research Service reported 591,079 consumer 3D printers were sold in 2018, after annualized growth rates of 49.4% in 2015 (Sargent Jr and Schwartz 2019). As 3D printers become more widely distributed, the potential for human exposure to emissions from 3D printer filaments for both workers (Chan et al. 2018) and the general public, including susceptible populations such as children in classrooms that incorporate ‘making’ (Berman et al. 2018; Ventä-Olkkonen et al. 2019), and individuals with pre-existing respiratory conditions such as asthma (House et al 2017), creates a need to evaluate the potential health risks of new use cases.

Driven by increased availability and decreased cost (Petretta et al. 2019), consumer-grade 3D printers are becoming increasingly popular in homes, schools and businesses, which are settings not traditionally designed for exposure control (Yi et al. 2016), and are locations where children may be at greater risk from particle exposures than adults (Bennett et al 2007). The application of 3D printers for the creation of onsite manufactured parts is continually expanding, including in austere environments (Pepi et al 2018) where parts (1) may be difficult to find or are obsolete, (2) not practical to mass produce, or (3) completely customized parts are required. Such agile manufacturing is particularly valuable in military applications, as evidenced by the U.S. Army’s recent proposals to develop containerized mobile 3D-printing operations for onsite fabrication of replacement parts, enabling the concept of the Soldier As Manufacturer (SAM) (U.S. Army Research 2017). These emerging use cases have the potential to place 3D printers and their operators in small, poorly ventilated rooms where proper use of personal protective equipment (PPE) may not be emphasized and safety requirements may be difficult to enforce (Kim et al. 2017; MakersEmpire 2020; Larson and Liverman 2011). In aAddition, cohorts of new users may not be informed of the potential adverse health risks resulting from exposure to fused filament fabrication (FFF) 3D printer emissions which may contain particulate matter (PM) in concentrations exceeding 300,000 particles/cm³ (Wojityla, Klama, and Bara 2017; Zhou et al. 2015; Stephens et al. 2013; Mendes et al. 2017; Kwon et al. 2017; Byrley et al. 2019; Stefaniak et al. 2019), even when common polymer filaments such as acrylonitrile butadiene styrene (ABS) are used without additives such as dyes or fillers. Consumers have limited awareness of the exposure potential presented by consumer-grade 3D printers, many of which lack controls to prevent exposures to ultrafine particles (UFPs), i.e., those with diameters < 100 nm, sometimes called PM_0.1 or aerosolized nanoparticles (NP). Novices and hobbyists are also more likely to ignore ventilation recommendations and instructions on safe product use than industrial workers, given that education attainment increases workplace safety compliance (Gyekye and Salminen 2009). Further, indoor air quality for home and recreational use is not regulated (Slezakova et al. 2019), elevating potential exposures from release of PM, engineered nanomaterials (ENM) and volatile organic compounds (VOCs) (Stefaniak et al. 2017a).

Exposure to UFPs shows strong correlation with adverse pulmonary, cardiovascular, and inflammation responses (Ohlwein et al. 2019; Heinzerling et al 2016). Recently Davis et al (2019) found that 3D printers sold with enclosures that are designed to reduce ambient air flows which negatively impacts prints did little to reduce human exposure to emissions, and that consumer units with HEPA-filters were noted to actually increase NP and VOC emissions. Previous investigators described measurements of emissions from 3D printers in detail (Stefaniak et al. 2017a; Mendes et al. 2017; Zhang et al. 2017; Stephens et al. 2013). Filaments now include ENMs such as carbon nanotubes, graphene, and metal or metal oxide NP (Alberts et al. 2020), thus increasing the likelihood of consumer inhalation exposure to ENMs and ENM/PM mixtures that may induce respiratory inflammation (Silva et al. 2013). As both acute- and long-term health impacts of FFF emissions become increasingly relevant (Chan et al. 2018; House et al 2017), it is important to better characterize the chemical composition and size distribution of such emissions to inform human health risk assessment. Notably, prolonged exposure to fine dust (PM_2.5) is associated with elevated risk of elevated fasting blood glucose and low-density lipoprotein cholesterol levels (Shin et al. 2020), lipid changes associated with hypertension were recorded as responses to PM_2.5 exposure (Lee et al. 2020), and yet UFPs or PM_0.1 are considered more hazardous than PM_2.5 (Schraufnagel 2020). Some of the common mechanisms of NP-mediated toxicity – where NP are often key constituents of PM_2.5 – include (1) direct impact on the organism’s cell surface, where the membrane might be damaged or initiate internal signaling pathways that damage the cell; (2) dissolution of the material in the cellular environment, releasing toxic ions, affecting the organism by impairing important enzyme functions or through direct interaction with cell DNA; and (3) the generation of reactive oxygen species (ROS) and subsequent oxidative stress, which also destroys important enzymes or an organism’s genetic material (Buchman et al. 2019). Computational models might now describe the behavior of particles in the trachea (Phuong and Ito 2015), supplementing literature showing exposure to high levels of NP result in particle deposition in the lungs and tracheobronchial lymph nodes (Mueller et al. 1990). Subsequently the toxicity attributed to NP results in adverse respiratory effects and systemic reaction of the total organism, including immunosuppression and immunomodulation (Dobrovolskaia et al 2016; Zolnik et al. 2010). Carcinogenicity of several types of NM such as carbon nanotubes and nanoscale titanium dioxide was confirmed based upon sensitive animal models, although epidemiological information is still far from being conclusive (Becker et al. 2011). Only preliminary toxicological information is available for possible adverse health effects related to 3D printing particle emissions. In vitro cellular assays and in vivo mouse exposures demonstrated toxicity to both PLA- and ABS-emitted particles (Zhang et al. 2019). In vitro cellular assays also reported concentration-dependent toxicity to polycarbonate and ABS (Farcas et al. 2019), with in vivo studies of ABS-exposure suggested minimal transient pulmonary and systemic toxicity (Farcas et al. 2020). In humans, Gümperlein et al (2018) found that one hr exposure of healthy volunteers to 3D printer emissions exerted no acute effect on inflammatory markers in nasal secretions and urine. Youn et al (2019) characterized the NP formation during 3D printer operations and suggested that a large number of produced NP between 10 and 30 nm are deposited in the lower human respiratory tract that might lead to significant adverse chronic health effects. Evidence also indicates inhalation of 3D printer emissions may pose adverse cardiovascular health risks (Stefaniak et al. 2017a; Chan et al. 2018; House et al 2017).

Exposures to the particle emissions from a 3D printer are complex and may not be adequately described by simple diffusion models. One experimental report used an indoor diffusion model with assumptions on ventilation rates and zero particle deposition rates to simplify the calculations (Poikkimäki et al. 2019). Other investigators employed a lognormal moment modeling approach, which does not provide spatial distribution information (Zhang et al. 2018), used Monte Carlo simulations to examine industrial processes (Zontek et al. 2019), and looked at the impact on nearby rooms using multi-zone airflow modeling (Azimi et al 2017). To characterize the many variables in modeling the exposures from 3D printers, an appropriate approach is to use a multiphysics simulation like Computational Fluid Dynamics (CFD) to capture both the unique geometry of the space and many additional factors created by the 3D printer and room ventilation. In addition, the influence of ventilation levels, air currents, and mixing in the room are largely unexplored, especially in regards to how the distribution of UFPs within a space may be affected. The spatial distribution afforded by CFD modeling is both novel and critical for understanding potential exposure levels based upon where operators are positioned relative to the exposure source.

While challenging, estimating the potential exposure level is a key component for estimating risk and implementing Prevention through Design (PtD) (Geraci et al. 2015). Computational modeling of the release of UFPs from 3D printers in small enclosed spaces, simulating use in a small home-based shop or deployed military environments such as portable shipping containers (U.S. Army Research 2017), is an important step to improving understanding of 3D printer emissions exposures and risk profiles. This investigation provides simulations of 3D printer particle release and diffusion throughout an enclosed space with minimal ventilation to model the concentration changes over time in the space. This information might be utilized to provide improved exposure and risk guidance to the research community, industrial hygienists, and the general public and inform optimal engineering controls and enclosure design

Materials and Methods

FFF 3D Printer

A Taz 6 FFF printer (Lulzbot, Fargo, ND) using filaments of “natural” (unpigmented) ABS (IC3D, Columbus, OH), printed a 40mm cube using the parameters defined in Table 1. This commonly used “neat” ABS polymer contains no additives or dyes, which might contribute to higher concentrations of UFPs (Stefaniak et al. 2017a).

Table 1.

The standard print settings for IC3D ABS were used as specified by the Cura slicing software version 3.6.20.

Parameter	Value
Base nozzle temperature	240 °C
Bed temperature	105 °C
Print speed (head movement)	60 mm/sec
Nozzle diameter	2.85 mm
Layer height	0.25 mm
Line thickness	0.5 mm
Wall thickness	1 mm
Infill	100%, lines
Print duration	3 hr, 30 min

FFF 3D Printer Particle Emission Rate Data Collection

To determine the unique FFF 3D printer particle emission rate profile, a critical parameter for the CFD model, the number, concentration and diameters of particles emitted from a commercially available 3D printer were measured within a 0.91 m by 0.91 m by 1.22 m electropolished stainless steel environmental chamber per the ANSI/CAN/UL 2904 standard located within in a Class 1000 clean room, as illustrated in Figure 1 (ANSI/CAN/UL 2019). The chamber was designed such that heterogeneity effects in the particle distribution were sufficiently minimized and controlled for as defined by current standards.

Figure 1.

(A) Illustration of testing setup to measure the particle emission rate from a 3D printer, utilizing an environmental chamber with prescribed dimensions, air exchange and sampling rates following ANSI/CAN/UL 2904. LPM = liters per minute; OPC = optical particle counter; SMPS = scanning mobility particle sizer. (B) Photo of environmental chamber testing setup. Note: The initial particle release measurements were performed inside the chamber to comply with the ANSI/CAN/UL 2904 standard when obtaining the data for use in the CFD modeling. Subsequent measurements for validation of the CFD modeling was performed openly in the cleanroom, as shown in Figure 2.

Printing was conducted within the electropolished SS304 chamber outfitted with intake ports for sampling and air exchange at a rate of one exchange/hr for the 0.91 m by 0.91 m by 1.22 m chamber, with 5 replicate experiments conducted. Sampling intake ports were located at the top center of the chamber, approximately 0.76 m from the median location of the print head, which was approximately 0.15 m above the chamber floor. A scanning mobility particle sizer (SMPS) (NanoScan 3910, TSI Instruments, Shoreview, MN) measured the diameter and concentration of particles generated during printing, scanning from 10 nm to 420 nm every min. Particle emission rate measurements were iteratively collected. Cleanroom air was forced into the chamber at a rate of 17 liters per min (LPM) to enable one full air exchange/hr (intake rates of sampling equipment are accommodated, including those of the Dorr-Oliver cyclone and Aerotrak 9306 OPC which were engaged to generate the standard air exchange rate, with data omitted from this study). A passive exhaust vent was located on the opposite side of the chamber. Power cords were fed through an access hole near the bottom of the back of the chamber, sealed with a polyethylene bag and closure ties.

Particle Emission Rate Calculations

From the particle concentration data, maximum and average particle emission rates PER(t) were determined according to the following formula presented in equation 1, as described by UL/ANSI/CAN 2904:

$$PER(t)=V_c\left(\frac{C_p\left(t\right)-C_p(t-\mathrm{\Delta }t)\exp (-\beta \cdot \mathrm{\Delta }t)}{\mathrm{\Delta }t\exp (-\beta \cdot \mathrm{\Delta }t)}\right)$$ (1)

Where V_c is the volume of the chamber, C_p is the measured concentration, Δt is the time between measurements (i.e., a sampling resolution, typically one min to allow for a full scan by the SMPS), and β is the particle loss rate, calculated to account for particulate adsorption onto surfaces and removal by sampling devices using equation 2:

$$\beta =\frac{l_n\left(\frac{C_p\left(t_1\right)}{C_p\left(t_2\right)}\right)}{t_2-t_1}$$ (2)

Where t₁ is a time at least 15 min after the print has completed and t₂ is at least 25 min after t₁.

Experimental Clean Room Setup

To represent the 3D printing process in a small space, a clean room was used as a simulated printing environment, with room dimensions of approximately 2.4 m by 2.4 m by 2.3 m depicted in Figure SI-1. Sample collection tubing placed within the clean room terminated at 8 selected locations and connected to the SMPS located outside the clean room. A series of ports within the clean room enabled serial sampling of multiple locations during the 4 hr total print duration (including warmup and cooldown) without disruption of airflows, which might arisen due to human movement during sample collection within the clean room.

Computational Fluid Dynamics Modeling

CFD is a computational method of solving fluid flow equations for mass, momentum and energy within a three-dimensional domain divided into thousands or millions of discrete elements that may be utilized to predict the characteristics of flow and contaminant concentrations within that domain. Moving beyond traditional applications such as automobile and aircraft design, CFD is becoming a widely employed modeling tool in defining the dispersion of contaminants in buildings, offices, and occupied spaces (Xu and Wang 2017). Over the past few decades, many studies were conducted to assess CFD-modeled concentrations and with increased utilization, continual research is required to adequately validate results for new applications. CFD possesses the capability to quantify the distribution and magnitude of a contaminant’s dispersion throughout a space, enabling computational modeling to estimate short-term and long-term worker contaminant exposures with significant time and resource savings afforded by not having to conduct numerous extensive experiments in the effort to protect worker health (Anthony 2009; Mele et al. 2016).

The CFD model domain geometry consisted of the clean room described above and the slightly larger room which housed the clean room. The air flowed through the clean room and under the clean room curtain walls which terminated approximately 6 in from the floor allowing air to flow up between the exterior of the clean room walls and larger room walls. The air exited the room either through the outlets (HEPA inlets) mounted on the top of the clean room or through the open door in the room (pressure outlet). The fluid domain was divided into approximately 4,150,000 tetrahedral mesh cells with a size range of approximately 0.05 cm to 10 cm, and included mesh refinement surrounding the 3D printer head particle source, a small printer head cooling fan, air inlets and air outlets.

The fluid, air, was introduced into the simulation via two 2 × 4 foot rectangles representing the ceiling mounted HEPA supply grilles in the clean room. A 120 and 115 fpm inlet velocity with 15 and 5% turbulent intensity, respectively, was applied to the left and right half of the inlets, respectively, at angles shown in Figure 5A. Each half of these two HEPA inlets was modeled with a uniform air velocity across the surface. A flow rate of 356 fpm was applied to a fixed fan zone with a 4-cm diameter in the approximate area of the printer head cooling fan. The HEPA outlet flow rate (outlet to the domain above the cleanroom and below the room ceiling) was 41 fpm from a 33.8 square foot outlet was equivalent to the inlet flow rate, calculated by mass. Spherical nanoparticles, 36 nm in diameter, were introduced into the model at an effective rate of 4.85 × 10¹⁰ particles/min from the print head location.

Figure 5.

A) photo of air inlets, and angle of air flow; B) and C) air current vectors, in a vertical slice centered on one of the HEPA-filtered air inlets; D) vertical plane of particle concentration distributions illustrating the heterogenous distribution. Color scale in B&C represents velocity magnitude in ft/min, and color scale in D represents particles per cm³.

A pressure based steady state simulation with second order upwind spatial discretization for momentum was used and the k-ω turbulence model was selected. Induced air flow was assumed to dominate flows, such that conditions were assumed to be iso-thermal (energy off). Mesh independence was not evaluated, convergence was assumed when all residuals to dropped below 1 × 10⁻³ within about 1,000 iterations and the particle concentration, at a monitoring point, was allowed to stabilize.

The CFD model of the simulated printing environment is illustrated in Figure 2A. Details of the software and processing used for computational modeling are provided in Table 2. UFP emission assumptions are shown in Table 3, based upon data collected and depicted in Figure 3, including particle release/generation rates, particle diameter and density.

Figure 2.

(A) Test Facility Geometry as built in the modeling software, and (B) workflow of 1) Measuring print emissions in standard chamber for values to use in 2) CFD modeling of the room, then validate by 3) measuring print emissions in clean room.

Table 2.

Modeling Software Details

Software and Processing	Description
CFD software	ANSYS Fluids v. 2019 R2
CAD software	SpaceClaim™ R19.2
CFD meshing software	ANSYS Mesher
CFD Solver	ANSYS Fluent 3D, double
software	precision, K-ε realizable, coupled
Processing	12 Xeon cores 96 GB RAM 8+ hr processing

Table 3.

Data inputs used in model

Parameter	Value
Particle emission rate (experimental steady state)	4.85 × 1010 particles/ min
Particle material	ABS
Particle shape	Spherical
Particle diameter	36 nanometers (nm)
Particle density	1.05 grams per cubic centimeter (g/cc)
Particle-Wall interception	Capture
Test facility room temperature	21 °C
HEPA supply average face velocity	0.610 m/sec for angled inlets, and 0.584 m/sec for straight inlets
Left and right HEPA Air Inlet Angle	70° toward center of room, and normal to ceiling surface, respectively.

Figure 3.

(A) The total particulate concentration (10–420nm in diameter, red trace, left axis) has been measured consistently for ABS filaments, allowing for calculation of the particle emission rate PER(t), blue trace, right axis, for modeling input. (B) Particulate concentrations by particle diameter emitted during printing of ABS were largely ultrafine, with the mean diameter of 36 nm.

Model Validation

Experimental measurements from 8 locations were collected to compare to the CFD modeled results. A series of 10 measurements were collected at heights representative of a standing adult’s personal breathing zone (i.e., 1.52 m to 1.68 m above floor) (OSHA).

Volume average concentration data from the CFD outputs were extracted from 8 3-dimensional 0.15 m by 0.15 m half-donut shapes, centered at the same elevation (1.52 m and 1.68 m) and radii from the print head as were the measured data.

A method developed by Chang and Hanna (2004) was utilized to estimate performance of CFD model in comparison with the NP concentration measurements. The following criteria were used:

$$FB=\frac{{\stackrel{-}{C}}_0-{\stackrel{-}{C}}_p}{0.5({\stackrel{-}{C}}_0+{\stackrel{-}{C}}_p)}$$ (3)

$$\text{MG}=\exp \left(\overline{ln{C}_0}-\overline{ln{C}_p}\right)$$ (4)

$$\mathrm{NMSE}=\frac{\overline{{\left(C_0-C_p\right)}^2}}{\overline{C_0}\overline{C_p}}$$ (5)

$$\text{VG}=\exp [\overline{({(ln{C}_0-ln{C}_p)}^2}]$$ (6)

$$\text{FAC}2=\text{fraction}\text{of}\text{data}\text{that}\text{satisfy}\text{the}\text{criteria}0.5\le \frac{c_p}{C_0}\le 2.0$$ (7)

Where, C_p is the model prediction of the concentration, C₀ is the observation of the concentration, overbar ($\stackrel{-}{C}$) denotes the average over the dataset, FB is the fractional bias, MG is the geometric mean bias, NMSE is the normalized mean square error, VG is the geometric variance, and FAC2 is the fraction of predictions within a factor of two of observations.

Chang and Hanna (2004) also utilized Pearson’s correlation coefficient as a test for the model’s performance, and proposed the following “acceptable” ranges of performance criteria: the fraction of predictions within a factor of two of the observations is approximately 50 % (i.e., FAC2 ≥ 0.5). The mean bias is within ± 30 % of the mean (i.e., –0.3 < FB < 0.3 or 0.7 < MG < 1.3). The random scatter is approximately a factor of two of the mean (i.e., NMSE < 4 or VG < 1.6). At the same time, Chang and Hanna (2004) stated that “these are not firm guidelines and it is necessary to consider all performance measures in making a decision concerning model acceptance.”

Results

Particle Emission Rate Measurements and Calculations

Measured particulate concentrations in the diameter range 10 nm to 420 nm are plotted in Figure 3A, including particle emission rates (PER(t)) derived from these measurements. This size range was selected because this is where the maximum particle emission rate of approximately 4.6 × 10¹¹ particles/min was found to be at the brief peak, with an average emission rate of 4.85 × 10¹⁰ particles/min during the majority of the printing phase. The results consistently demondtrate a brief peak of high ultrafine emissions on the SMPS early in the print then the ultrafine concentration tapers off becoming negative briefly approximately 10 min into the print prior to becoming positive again and remaining fairly constant for most of the print. This initial brief peak is commonly reported in similar studies as well as in the ANSI/CAN/UL 2904 standard. The brief period of negative particle emission was observed previously in data from Byrley et al (2019) and Azimi et al (2016) Byrley et al (2019) reported negative particle emission rates as occurring when there are losses in the system that are unaccounted for, leading the slope of the particle number concentration to become negative, and thus driving the entire result of the particle number emission rate calculation to become also negative.

In this case, an initial burst of particles is released, followed by a slower release rate for the remainder of the print. During the transition from the initial burst to the slower release rate, effectively losses of particles are occurring, perhaps being removed due to particle inelastic collisions with the surfaces of the chamber and/or perhaps due to clean air exchange rate prescribed by the standard. Thus, the effective emission rate becomes negative briefly before resuming the new slower release rate.

These experimentally observed PER(t) parameters were used for modeling inputs. The total emitted particulate concentrations varied by sampling location but generally followed the size distribution illustrated in Figure 3B, with emitted concentration peaking around a mean diameter of 36 nm.

Modeling

Simulations were run to represent the steady state (long term printing) mean particle concentrations emitted by the 3D printer. Maximum particle concentrations per cubic centimeter (#/cm³) were found to approach values of 1 × 10⁴ #/cm³.

Horizontal slices at different elevations are presented in Figure 4. These elevation planes are referred to as floor (0.5 m), print head (1.1 m), breathing zone (1.6 m), and ceiling (2 m). It is important to note that the ‘floor’ plane is 0.5 m above the floor, and the ‘ceiling’ plane is 0.3 m below the ceiling.

Figure 4.

Results of particle concentration distributions at modeling steady state conditions. Elevation slices representing A) ‘floor’ or 0.5m above the actual floor, B) ‘print head’ or 1.1 m, C) ‘breathing zone’ or 1.6 m, and D) ‘ceiling’ or 2 m above the actual floor and 0.3 m below the actual ceiling. Color scale represents particles per cm³.

A heterogeneous particle distribution was noted at each elevation. Near the floor, small regions with higher particle concentrations developed near three corners of the room, with large areas of extremely low particle concentrations throughout (Figure 4A). At print head height, a larger region of higher particle concentrations accumulated along one of the walls, in addition to a small region of higher particle concentrations in the corner, above the region noted in the floor elevation (Figure 4B). Large areas of low particle concentrations were also found at the print head elevation in the room. At the breathing zone height, a large region of high particle concentrations developed along the same wall where the print head elevation high concentration was found (Figure 4C). The high particle region also stretched over the 3D printer and extended nearly continuously into the corner. At the ceiling height, a region of high particle concentrations extended across the room in the plane of the 3D printer, and along the same wall where elevated particle concentrations are located at the lower elevations (Figure 4D). The corner that had high particle concentrations at lower elevations is relatively particle free at the ceiling elevation, as is a zone near the front of the room.

Particle concentrations varied throughout the model, largely depending upon where clean air was introduced into the room and where mixing of clean and contaminated air occurred. Evidence suggests that in situations where an individual needs to be present in the same room as a 3D printer source, their positioning and elevation such as sitting versus standing might markedly influence their exposure to UFP emissions.

Air inlet conditions used in the model were based upon the actual air inlet measurements collected in the experimental clean room, where half of the ceiling-mounted-HEPA delivered air directly down into the room, and half of the HEPA, covered by a supply diffuser, delivered air at an angle, as depicted in Figure 5A. The positioning of the HEPA units created air currents that were accounted for in the model, as shown in Figure 5B and 5C. In the final CFD simulation, this inlet angle was increased (to approximately 70°) to more appropriately represent actual conditions. Examination of the particle concentration contour on a vertical slice through the center of the room on a plane where the 3D printer is positioned revealed a noticeably asymmetric distribution of particle concentrations (Figure 5D). At higher elevations in the room, greater particle concentrations were observed. Further, comparing the two sides of the room, one side displayed a greater volume that contains high particle concentrations compared to the other. In summary, the rendering illustrates certain locations in the room have higher local particle concentrations, and that the locations within the room of the highest released particle concentrations were generally 1) near the printer head origination source, 2) in the upper corners, and 3) in clouds located vertically in the upper third of the room.

Model Validation

The 8 particle concentration measurement locations are superimposed on the CFD particle concentration contour illustrated in Figure 6A, and the measurements collected throughout a series of 3D prints are presented in Table SI-1. In brief, there was general agreement between experimental observations and model predictions. The results of the application of Hanna’s criteria are shown in Table 4, where values in the suggested “acceptance” range are bolded along with significant correlation. Acceptance is established when at least one acceptance parameter is within the acceptance range for each performance category. The relationship between measured and modeled concentrations is demonstrated in Figure 6B. The results in Table 4 demonstrate that the CFD prediction completely agrees with Hanna’s criteria for airborne dispersion modeling. The importance of this finding is in agreement between experimental and modeling results using Hanna’s criteria that are elaborated in the discussion.

Figure 6.

(A) Plot of particle concentration measurement locations during 3D print before steady state was achieved, overlaid on model of particle concentration distribution in room at steady state. Color scale represents particles per cm³. (B) Plot of CFD modeled vs. measured particle concentrations.

Table 4.

Criteria of model’s performance

Performance Category	Parameter and level of acceptability	Criteria for CFD study	Acceptance Range
Mean bias	FB	−0.41	−0.3 < FB < 0.3
	MG	0.75	0.7 < MG < 1.3
Scatter	NMSE	0.68	NMSE < 4
	VG	3.55	VG < 1.6
Fraction of prediction	FAC2	0.5	FAC2 ≥ 0.5
Correlation	Pearson correlation	0.72	N/A

Discussion

Prevention through Design principles include strategies to reduce both hazards and exposures in the design and manufacturing of products and materials (Geraci et al. 2015). Data demonstrated progress towards reducing inhalation exposures to 3D-printer emissions through the combination of computational modeling validated with experimental data. Modeling enables rapid design iteration when screening potential PtD interventions without requiring safety professionals to physically build and test each option, thereby saving significant time and resources. These models might predict and optimize the impact of clean airflow filtration and directional characteristics on the particulate distribution in the specific room geometries of anticipated user locations in ways that might be useful for diminishing particle exposures for operators when being in the room is unavoidable during FFF 3D printing operations. The models might also inform the recommended type and extent of engineering controls and PPE that the operator needs to employ based upon additional air filtration requirements, keeping in mind within the industrial hygiene hierarchy of controls that the use of PPE is a last resort and primary emphasis needs to be placed on approaches of elimination, substitution, and engineering controls. Modeling data might potentially be utilized to inform elimination strategies, such as avoiding the use of high emitting filaments, and substitution strategies, such as identifying a high emitting filament that may be employed under proper conditions in a limited capacity but needs to be substituted with lower emitting filaments during routine work.

The objective of studying the PtD scenario described in this investigation was to develop a better understanding of UFP distributions within a small enclosed room where FFF 3D printing occurs, using the approach of combining modeling with experimental data collection in a Class 1000 clean room to validate the modeling. The results presented in Figure 4 indicate that assumptions of homogenous particle distributions within an enclosed space cannot be made for FFF printer emissions, particularly when air flows are present. While simple point diffusion mathematical models yielding homogenous particle distributions are a useful starting point, these make many assumptions that may not fully reflect the conditions found in real-world use cases. Therefore, a combination of CFD modeling with initial standardized measurements of the particle generation rates and sizes might be used to provide more robust insights regarding the nature of particulate exposures within work areas used for FFF 3D printing. To keep simulation costs at reasonable levels and more broadly applicable to other scenarios, a focus on using the most representative air inlet angle, clean air delivery, measured particle generation rate and sizes, and using a roughly cubic room size and shape were made to simplify the computational intensity requirements of this modeling effort.

To compare the model to real world conditions, a series of observations were made in a Class 1000 clean room of similar dimensions to the cubic room being modeled. Closely pairing the modeling and experimental conditions is important because CFD results are specific to one set of conditions. Further, for purposes of measuring primarily the emissions from the 3D printer, use of the clean room provides a low background particle concentration, thus improving the sensitivity and quality of the measurements. Reliable quantitative agreement was found between measured data and predicted CFD results, shown by Hanna’s criteria that were met for the 8 measurements locations depicted in Figure 6B and Table SI-1. The CFD results were taken in donut shapes rather than from a single data point in the simulation to increase the confidence in the comparison. The measurements were collected after several hr of run time, which is the time frame recommended by the ANSI/CAN/UL 2904 standard. This time frame is also on the same order of magnitude as the real-world time anticipated to achieve the steady state results modeled by CFD for a room of this size.

The heterogeneous particle distribution might primarily be attributed to the role of air flow and mixing of inlet air, as shown in Figure 5B. The angled side of the HEPA diffuser appears to create an air flow that generally concentrates particles more to one side of the facility, along the plane of the printer, and towards higher elevations within the room. The orientation and angle of air inlets can create a heterogeneous distribution of particle concentrations that leads to regions within the room that have a lower than predicted exposure and regions that have a higher than predicted exposure as compared to the homogenous particle distribution assumption. The practical application of this observation leads to understanding the physical locations within the room where users may be at greater risk, based upon the CFD model results, which might be used to inform a lower cost approach to designing and testing engineering solutions such as optimal air inlet angles for directing particles away from the breathing zone and toward other zones.

The emission rate of 4.85 × 10¹⁰ particles/min determined from the present study is within the range of 2 × 10⁹ to 9 × 10¹⁰ particles/min reported by Byrley, et al (2019) in a recent literature meta-analysis. Therefore, it is possible to extend the methods developed and presented in the current investigation to other desktop FFF 3D printers examined in the literature with a reasonable degree of certainty for successful implementation.

On the side of the room with high particle concentrations, the breathing zone concentration generally is found to be between 1.0 × 10² and 1.0 × 10³ #/cm³. A mean particle concentration of 2.87 × 10⁵ ± 5.41 × 10⁵ #/cm³ was reported for ABS filaments in the meta-analysis (Byrley et al. 2019). Byrley et al (2019) states high variability in reports arises from publications included in their analysis often omitting details such as number of print runs, as well as limited number of papers being suitable for inclusion.

To place the results of this study in a strict PM_2.5 perspective, exposure levels observed equate to 0.018 μg/m³, which is well below the PM_2.5 air quality standard of 15 μg/m³ set by the U.S. Environmental Protection Agency (USEPA 2019), and below other studies where particles for one printer were found to be 0.3 mg/m³ and for three printers run simultaneously was 0.7 mg/m³ (Chan et al. 2020). However, similar to other nanoscale materials that do not adhere to conventional toxicological mass-based dosimetry (Hull et al. 2012), the hazards of smaller diameter particles such as the UFPs examined in this report, while technically a subset of PM_2.5, are different than those on the larger spectrum of PM_2.5 (Chen et al. 2016). Inhaled UFPs translocate beyond the nose and lung to the brain, circulatory, and gastrointestinal regions, depending upon their physicochemical properties (Oberdörster et al. 2004; 2005; Li et al. 2012).

Our observations are an important step, validating CFD modeling with experimental data, to enable rapid screening of PtD considerations to inform engineering design and controls for exposure reduction. A recent significant work examined modeling 3D printer emissions using Monte Carlo simulations, which enables determining a probability of exceeding a specified threshold concentration without the ability to offer site-specific particle concentration determinations.(Zontek et al. 2019) This investigation enables both site-specific determination of particle concentration, and validation of the model with experimental data. However, there are many exposure parameters that future studies need to explore to determine the sensitivity and impact of each parameter on the results, including: (1) role of the print head motion, (2) print head cooling fans,(3) convection currents from melting the filament and stage materials, (4) return air vents at various room exchange rates, (5) local exhaust ventilation, (6) influence of individuals moving briskly throughout the room, (7) heat sources, (8) kinetics of the clearance of UFPs from the room, and (9) multiple printers in simultaneous operation. As future studies determine which of these variables are most impactful, this may lead to further refinement of appropriately simplified computational models by focusing on first order effects, thereby enabling even lower costs from the decreased computational intensity requirements, and ultimately more widespread deployment of these techniques.

In CFD simulations, simplifications of the real world are necessary to achieve reasonable results in a reasonable amount of time; therefore, in the CFD simulation performed, simplifications of the air flow inlets, particle size, and temperature were applied to a degree that would not significantly affect the results. The inlet velocity and inlet angle for each of the HEPA units was measured and a uniform air velocity across the HEPA inlet (outlet from the HEPA into the fluid domain) surface was applied to the model and deemed acceptable due to high room air exchange rate. The HEPA outlet flow rate (inlet into the HEPA unit above the cleanroom) was 41 fpm from a 33.8 square foot outlet. This large outlet area was applied to reduce unnecessarily large air velocities and consequently accelerate convergence, and would not affect the results due to the outlet being significantly downstream from where the results were extracted. Although the size distribution was measured, a single sized NP was introduced into the model to simplify the computational intensity and further within the measured distribution of particle sizes, NP tend to behave in a similar manner due to their small size. Conditions were assumed to be iso-thermal (energy off) because the convective flow produced by high temperature print head was negligible as compared to the significant room air exchange rate (greater than 250 air changes per hour). Initially the k-Ɛ turbulence model was used, however, due to the large air change rate that generated low velocity swirling flow, the k-ω turbulence model was finally selected and applied.

The United States Centers for Disease Control and Prevention’s (CDC) National Occupational Research Agenda (NORA) for Healthy Work Design and Well-Being states Objective 3 as: “Address the safety and health implications of advancing technology.” The approach of this investigation of combining modeling and experimental data supports PtD principles, which include strategies to reduce potentially harmful exposures (Geraci et al. 2015). The approach presented of quantitatively measuring and modeling the number of UFPs in specific exposure scenarios enables insights to be gained by modeling before necessity to physically build and test as well as to validate measurements after specific design solutions are implemented. The importance of designing the models for a specific set of room conditions is critical for the success of this approach. Understanding the velocity and angle of ventilation air being introduced into the room is an important consideration. The approach presented herein enables modeling to quickly screen PtD options, without needing to build and test, potentially saving significant time, money, and human resources. As an example, this might enable safety professionals to use PtD principals to optimize regions of minimal exposure where operators need to be when presence in the room is unavoidable. Such insights may be especially important in home- or school-based applications of FFF 3D printers where human error may be more likely and the use of PPE or other protective strategies are either impractical or unenforceable.

In conclusion, data demonstrate a workflow by which rapid and efficient computational modeling provides useful insights for predicting potential exposures of users of FFF 3D printers to UFPs, which might inform subsequent human health risk estimations. The insights of heterogeneity of particle concentration distributions within a room might be determined quantitatively and cost-effectively using CFD. The complexity of the distribution of particle concentrations highlights the need to perform simulations to provide visualizations useful for informed and targeted risk mitigation efforts and PtD when installing 3D printer equipment.