An Optical Aerosol Sensor for Process Monitoring of Aerosol-Jet Printing

Aerosol-Jet Printing (AJP) is a fabrication technology for 3D printed electronic applications ranging from conformal metallization to system-in-package hybrid devices. The desire to achieve improved control of the AJP process has spurred the development of strategies for in-line estimation of key AJP process variables. This paper describes the design and initial testing of a novel optical light scattering sensor for enabling in-line process monitoring and control of aerosolized ink properties for future process control strategies. The sensor measures optical scattering by the AJP aerosol at nine angular positions using silicon photodetectors and LED light sources, providing a path for a compact, cost-effective in-line monitoring solution. An experimental evaluation of the sensor response to varied AJP ink compositions and process parameters was performed. Finally, the optical response of the system was modeled, and a general framework for extracting quantitative measurements of the scattering and absorption coefficients of the AJP transport mist was provided.


I. INTRODUCTION
Aerosol-jet printing (AJP) is a direct-write technique for manufacturing electronically functional structures and devices with particular promise in next-generation advanced packaging, sensors, interconnects, and 3D printed hybrid electronics (PHE) [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11].The AJP process involves the generation of a dense aerosol from a liquid functional ink, transport of the ink by a carrier gas flow to a deposition head where the aerosolized ink is aerodynamically focused, enveloped by a secondary flow of sheath gas, and deposited in a collimated laminar flow onto a targeted substrate to form electronically functional patterns with resolution as fine as 10 µm.A conceptual diagram of the AJP process is presented in Fig. 1.Detailed descriptions of the AJP process and its physics can be found in [1], [12], and [13].
The associate editor coordinating the review of this manuscript and approving it for publication was Yiqi Liu .
Although the potential for complex, high-precision fabrication with AJP has been demonstrated, there remains a need for improved predictability and control of the AJP process to facilitate the widespread application of AJP technology [14], [15], [16].Due to the lack of in-situ feedback of deposition variables available on commercial AJP systems, efforts to address AJP process predictability have focused primarily on identifying ranges of AJP material sets and process settings (i.e.''process windows'') associated with certain structure and property outcomes, such as printed line width, morphology, or resistance [17], [18], [19].However, the need for timeconsuming post-deposition sample characterization of these approaches limits the practicality and generalizability of such open-loop approaches.Efforts to characterize the AJP deposition process in near real time have been pursued in recent years, including the use of calibrated, micro-machined inkwells providing an on-demand measurement of the deposition rate [20], and estimation of AJP-deposited microstructures using post-deposition imaging [21].While these approaches have been used to implement feedback for AJP control and improve process predictability, they do not provide continuous real-time monitoring of the deposition process.
More recently, a strategy relying on optical scattering measurements of aerosolized ink immediately prior to deposition has been demonstrated to be predictive of the AJP deposition rate [22], [23], [24], [25].The technique takes advantage of the strong Mie-type coupling of visible and near-infrared light to liquid aerosols with diameters in the 1-5-µm range of AJP inks.Tafoya et al. demonstrated the sensitivity of an extinction-type measurement to the aerosolized ink volume fraction and showed its predictive power for the AJP deposition rate, validated through conductivity and thickness measurements, despite the challenges of drift associated with the extinction measurement approach [22].Later, Secor developed a variation of this approach incorporating a scattering-type measurement along with an extinction measurement, demonstrating reduced measurement drift and improved sensitivity as well as basic closed-loop control capability for both transparent and opaque functional inks [23].Rurup and Secor subsequently matured the scattering measurement approach by integrating automated closed-loop PID control of ultrasonic atomizer power, ink carrier gas flow rate (CGFR), and linear feed rate, demonstrating real-time closed-loop control of the AJP deposition rate [25].This work demonstrated the utility of optical scattering-based inference of aerosol density for controlling the aerosol density itself, and hence controlling the AJP deposition rate.However, as noted by the authors of these studies [22], [23], [24], [25], although these optical configurations are sensitive to aerosol density, they are also sensitive to other key parameters, including ink composition and droplet size distribution, both of which are known to affect the scattering efficiency in the Mie regime.The sensitivity of optical scattering to variations in multiple separate ink aerosol parameters without differentiation between parameters may lead to ambiguity in the measurement, requiring assumptions on the stability of the droplet size distribution as well as the ink composition.While relying on well-engineered inks with highly stable compositions (e.g., solid loading) is likely an effective strategy for mitigating uncertainty in aerosol ink composition, it adds complexity to the ink development process and limits the choice of ink selection to inks with exceptional stability.Similarly, the droplet size distribution strongly influences the overspray and impaction efficiency [26], [27].Thus, without differentiation between droplet size distribution changes and changes in the total aerosol density, the controllability of this key print parameter is inhibited.The three parameters of aerosolized ink -composition, aerosol density, and droplet size distribution -all result in variations in optical scattering and absorption.The inability to discriminate between sources of variation in scattering intensity results from measuring only a single scattering angle, without a wavelength-dependent optical scattering detection scheme.The angle and wavelength dependencies of optical scattering in the Mie regime are both generally nonlinear and nonmonotonic; thus, a measurement scheme that can infer these angular or wavelength dependencies is required to differentiate between changes in the particle number density, droplet size distribution, and material composition [28].
Here, the design and initial testing of an optical aerosol sensor for AJP ink monitoring that detects scattering at multiple scattering angles and wavelengths, enabling the fitting of measured scattering profiles to optical scattering and absorption properties, are described.The sensor provides measurements that may be used to infer the optical scattering and absorption properties of the ink aerosol, which can in turn be mapped to aerosol physical properties, including aerosol density, size distribution, and material composition.The location of the optical aerosol sensor within the AJP process for this angle-and wavelength-sensitive approach is shown in Fig. 1.The sensor measures the optical scattering of the aerosolized ink immediately prior to the entry of the ink into the deposition head, enabling the inference of changes to the ink-stream composition occurring before the focusing and deposition stages.

II. AEROSOLIZED INK OPTICAL PROPERTIES
The ink aerosol in AJP tools is generally understood to be composed of droplets with a 1-5 µm log-normal diameter distribution, with a median distribution near 3.1 µm [1], [12].In this size regime, Mie coupling dominates the observed measurements in the visible and NIR spectra.The intensity of observed optical scattering and attenuation can be understood in terms of the scattering and absorption coefficients of the aerosol, which are the probabilities of scattering and absorption events per unit distance of the optical path.For the AJP transport mist, the scattering and absorption probabilities of the aerosol are dependent on the Mie cross sections of the droplets as well as the number density of droplets.These optical coefficients for a given aerosolized ink are the sum of the contributions from the number density and cross-section for each diameter represented because the cross-sections are both wavelength-and diameter-dependent.For example, the scattering coefficient µ s , is calculated as follows: for a distribution of discrete diameters d where σ s,d is the scattering cross-section, and ρ d is the number density of droplets with diameter d [29].Similarly, the absorption coefficient µ a , can be calculated as where σ a,d is the absorption cross section of a droplet with diameter d.The optical cross-sections of the droplets were calculated using the optical scattering and absorption efficiencies scaled by the cross-sectional area of the droplets.For scattering this is where is the scattering efficiency of a single droplet, which is a function of the wavelength of light λ, and the complex refractive index of the ink m (λ).In this study, Q SCA was calculated numerically, considering unpolarized light, using an open-source Python module [30].Because the complex refractive index is a material property of the liquid droplet, its estimation provides a measurement of the ink composition (e.g., solid fraction and solvent content).Analogous expressions for the absorption cross-section and absorption efficiency, Q ABS , also depend on the droplet diameter and the complex refractive index.
In addition to the wavelength-and droplet-size-dependent scattering and absorption probabilities, the angular direction of scattering can be estimated in the Mie regime.In this study, the effects of droplet size and material composition were calculated by assuming the Henyey-Greenstein angular scattering dependence, g, a commonly used Mie phase function [28], [29], [30].
Together, µ s , µ a , and g can be used to model the scattering characteristics of the AJP ink aerosol using a Monte Carlobased ray-tracing method.In this study, a commercial Zemax optical modeling software was used.In summary, the estimation of the optical coefficients needed to describe the optical scattering of the AJP aerosolized ink requires three physical parameters: m (λ), the complex refractive index of the ink liquid; d, aerosolized ink droplet radius distribution; and ρ d , the number density of droplets as a function of diameter.
To understand the relationship between aerosolized characteristics such as median droplet diameter and ink composition on optical properties, as well as to inform the design of a sensor for monitoring such properties, modeling of wavelength-dependent Mie scattering and absorption cross-sections was performed.Here, the optical properties of a model AJP ink consisting of Ag and water, the primary components in commonly used Ag nanoparticle conductive inks for AJP, were calculated using the reported complex refractive indices for water [31] and silver [32] using a linear mixing rule [33].This approach was adopted because the complex refractive index for common silver nanoparticle (Ag-NP)-based conductive AJP inks is unavailable.While the approximation of Ag-NP inks by linear mixing of bulk Ag and water properties was expected to be imprecise, the purpose of this modeling was to obtain qualitative, representative estimates of conductive ink aerosol optical properties.The Mie scattering coefficients calculated for the 350-1800 nm wavelength range for several Ag-water droplet diameter distributions and compositions are shown in Fig. 2, where a log-normal distribution of the droplet diameters for the ink aerosol was assumed.In Fig. 2A, the cross-sections of the model ink are shown for aerosols with median diameters of 2.91, 3.01, and 3.11 µm.The scattering and absorption cross-sections both increased with increasing median droplet diameter, although the effect on scattering was more pronounced.This effect results in increased scattering of light by the aerosol, especially for wavelengths longer than 750 nm.In Fig. 2B, cross-sections of the model ink are shown for aerosols with fixed droplet size distributions and varied compositions, with the percentage of water in the model aqueous ink varying from 45% to 55%.These results show that the compositional variation has a complex influence on scattering and absorption cross-sections, with increased water content resulting in lower absorption across the wavelength spectrum.However, the scattering profiles exhibit an inflection point around 1250 nm, where the increasing water content results in decreased scattering of longer wavelengths of light and increased scattering of shorter wavelengths.
The results of Fig. 2 highlight the sensitivity of optical scattering and absorption to aerosol composition and droplet distribution.From Eqs. ( 1)-( 3), the strength of scattering and absorption observed is a combination of the scattering and absorption cross-sections, as well as the number density.Achieving specificity in measurement inference to avoid misattribution of a change in scattering measurement to, for instance, aerosol number density, requires a method that can distinguish between linear changes in scattering and absorption strengths due to aerosol density, and nonlinear changes in the optical scattering and absorption cross-sections due to ink composition and droplet size distribution.Based on work on related problems in other fields of property extraction from turbid media, it is likely that an approach using both spatial and spectral variation will yield the most robust outcomes [28], [34].In this study, spatial variation is achieved by implementing an optical measurement cell with scattering measurements at multiple angular positions relative to an illumination source, and spectral variation is achieved by multiplexing two LEDs with distinct emission wavelengths.

III. CELL DESIGN AND CONSTRUCTION
A prototype light scattering sensor was designed that relied on silicon photodiode light collection and visible (Vis) spectrum LED light generation.While the NIR spectrum provides useful features and enhanced scattering of aerosols with a 1-5 µm diameter distribution relevant to AJP, the higher responsivity (A/W) of commercial Si photodiodes compensates for the generally reduced scattering efficiency and enables low-level light collection without bulky and expensive cooled photodetection.Direct collection of scattered light by photodiodes, rather than collection via waveguides (fibers), was used to further enhance photo collection, based on the larger physical apertures as well as numerical apertures afforded by photodiodes compared with optical fibers.
The design of the optical cell geometry was informed by modeling a small volume of water aerosols with varied droplet sizes.Simulations were performed using Zemax optical modeling software, with droplet optical properties calculated by the Mie approximations described in Section II.A representative modeled ray trace from these simulations is shown in Fig. 3 (left).The black discs in the wireframe model represent the collection apertures used to estimate the light collected at a given aperture size.The design goal for the scattering cell was to maximize the number of detectors surrounding the sensing cavity using commercially available components to maximize the number of angular positions available for correlating optical scattering at a given wavelength to aerosol properties.A design tradeoff to maximizing the number of detectors which may surround a sensing cavity is the size of each detector aperture, since a given radius of cavity may accommodate a larger number of smaller apertures, or a smaller number of larger ones.While more detector apertures were expected to improve optical property measurement capability, a minimum detection aperture is needed to provide sufficient illumination at each detection position.As shown in Fig. 3 (right), the intensity of the scattered light is expected to decrease exponentially as a function of the scattering angle.A pixelated cylindrical detector was used to assess the minimum detector apertures expected to provide sufficient signal collection by integrating portions of the irradiance-vs-angle for a varied solid angle.This approach was validated by comparing the integrated cylindrical detector quantities with those of the modeled radial detectors at varied locations.To compensate for the far higher intensity of the scattered light at angles closer to 0 • , smaller collection apertures and photodiodes were selected for collection at angles closer to 180 • .A larger collection aperture was used for the 0 • transmission location to trap unscattered light.Based on the integration of the modeled scattering profiles, apertures of 1 mm diameter for lower angles, and 2 mm diameter for higher angles were deemed reasonable for the range of optical properties expected for common AJP inks.
The annotated CAD renderings of the prototype optical cell are shown in Fig. 4. Here, the configuration of photodiodes with respect to the aerosol transport direction is shown, along with a description of the locations of the photodiodes with respect to the aerosol sensing cavity and input illumination.Collimation of the input illumination was desired to restrict the path of unscattered light to 0 • , and was achieved with a commercial fiber-coupling collimator from Thorlabs.
The body of the optical cell was machined from black polyacetal plastic (Delrin) with apertures for photodiodes, optical windows, and threaded gas inlet and outlet ports designed for compatibility with commercial off-the-shelf fittings.The sensing cavity was a cylinder of approximately 9 mm in diameter and 15 mm in height.This size of sensing cavity was selected to accommodate integration with commercially available transport tube connections.The sensing cavity size, in turn, resulted in a maximum number of nine optical collection apertures surrounding the cavity, given the combination of 1 mm and 2 mm sensing apertures for low and high angle scattering, respectively, along with the need for an input illumination collimation component.The top and bottom of the cavity were threaded for 1/8-27 NPSM quick-connects, which were chosen to facilitate connection to the AJP ink transport line and for disassembly, inspection, and cleaning of the sensing cavity when needed.The Si photodiode detectors were separated from the ink-stream aerosol using 5 mm diameter sapphire windows with recesses, allowing press-fit sealing of these windows.The windows were set back from the sensing cavity by cylindrical apertures that were 1 cm long and 1 mm or 2 mm in diameter.The 1 cm offset of the windows from the sensing cavity effectively mitigated fouling of the optical windows by the aerosol mist.The commercial Si photodiodes used for each collection position are listed in Table 1.The wavelength dependent responsivity of the Si photodiodes, as well as device-to-device responsivity variation, was compensated for by a calibration procedure described in Section IV.
A system was constructed to deliver light and monitor optical signals from primary cell photodiodes.The block diagram of the system is shown in Fig. 5.The illumination for the optical scattering cell was provided by LEDs coupled with optical fibers.Light from each LED was combined with a 2 × 1, 1:1 fiber bundle into a single 1 × 2, 90:10 fiber coupler.The 10% arm of the 1 × 2, 90:10 fiber coupler was coupled to a Si photodiode, which was converted to a digital voltage via transimpedance amplifier (TIA) and  analog-to-digital converter (ADC) units.The reference arm intensity was then used to correct for fluctuations in LED source intensity over time.The 90% arm of the 1 × 2 coupler is sent to the scattering cell collimated through the SMA fiber collimator, providing an illumination beam diameter of approximately 1.5 mm.The LEDs were controlled electronically via TTL signals from the data-acquisition (DAQ) unit digital-to-analog converter (DAC) ports.The two LEDs were enabled in an alternating fashion, with measurements from the cell photodiodes acquired after a short (∼ 1 s) waiting period for each LED to stabilize.The DAQ unit used for the development was Labjack U6-Pro.The code for controlling the LED sources and measuring the voltages of each TIA was written in Python.The control and monitoring software provided options for varying the number of integrations to perform for each time point measurement and for adjusting the gain and resolution of each channel of the ADCs.

IV. PROTOTYPE TESTING
Testing was performed to assess the ability of the system to monitor the changes in aerosols transported from the atomizer to the AJP deposition head.For this study, an AJ200 AJP system (Optomec Inc., NM, USA) with an ultrasonic atomizer (UA) was used for aerosol generation.The ink compositions tested included commercial Ag ink (Novacentrix JS-A426), Ag ink diluted with deionized (DI) water (the primary solvent for this ink), and pure DI water.Testing with pure DI water as a model ink was the focus of this stage of cell characterization because the refractive index of DI water is well characterized, and because the material composition of the water-as-ink is not sensitive to the effects of solvent evaporation, which facilitates comparison with optical modeling of the cell behavior.The goal of this prototype testing was to establish whether each detector channel on the optical cell had sufficient collection efficiency to respond to known changes in the transport aerosol and whether the magnitude of the response for each channel was on the order of that expected based on optical modeling.The ability of each detector channel to respond to changes in the ink aerosol was tested by varying the carrier gas flow rate (CGFR), ultrasonic atomizer (UA) current, and ink composition in the UA cartridge.Fig. 6 shows the uncalibrated response of the cell over time as the CGFR was varied for a water ink, where for each time point, the measurements are the mean of 20 ADC measurements of the TIA voltage outputs, with the standard deviation of the 20 measurements plotted as error bars.The standard deviation for the scattering detector measurements ranged between 0.5% and 2.0% of the measurement magnitude for the time points at which the CGFR aerosol was present.Prior work has demonstrated that the CGFR is positively correlated with the aerosol droplet density and droplet distribution, which, for a transparent ink such as water results in increased scattering and decreased transmission measurement magnitudes [12], [22], [25].
To understand whether the cell detector readings were close to the expected response for the cell's designed geometry, calibration measurements were performed of the cell response to incident light along the axis of aerosol transport (orthogonal to the illumination direction depicted in Fig. 4).Because of the multiple scattering of light by the aerosol in the sensing cavity, light entering the cell along the aerosol transport axis scatters isotopically along the axis of the array of the detector apertures.Therefore, the measurement of such light can be used to establish the ratios of the response magnitudes between detector channels.This calibration measurement accounts for differences in detector reading owing to factors including photodiode responsivity, optical window transmission efficiency, and collection efficiency of each detector aperture (owing to geometry and fabrication variations).The measurement was performed by removing the aerosol transport tube from the aerosol outlet quick-connect, inserting the input illumination fiber into the port, and flowing aerosolized DI water into the sensing cavity.Specular reflections were removed by subtracting the measurements with zero flow for this configuration.To mitigate the dependence of the measurement on the launch condition of the input illumination in this configuration (e.g., owing to precise alignment), a high-density water aerosol was generated with a UA current of 0.4 mA and a CGFR of 62 ccm.In this configuration, the light incident along the axis of the aerosol flow is scattered immediately after its entry into the sensing cavity, and the relative intensities measured at each detector port are weakly dependent on the precise alignment of the input illumination.To verify this calibration approach, the process was repeated multiple times with varied input fiber positioning, with negligible differences in the resulting calibration profiles.The cell responses shown in Fig. 6 divided by these calibration measurements, as well as by the normalized reference PD readings, provided the normalized measurements shown in Fig. 7 which enabled comparison with the modeled results.The profiles in Fig. 7A are plotted along with the CGFR recorded by the AJP mass flow controller (MFC) at the time of the tests, exhibiting expected correspondence between detector magnitude changes and the CGFR setting.The delay between the detector response and the rising edge of the CGFR MFC readings, most notably for the first setpoint, is believed to be due to ''seasoning'' of the initially dry transport tubes.The expected positive correlation between CGFR and measured scattering, and negative correlation between CGFR and transmission, is clearer in Fig. 7B, where the portions of same data in Fig. 7A are plotted with averaged detector signals corresponding to steady-states of CGFR values applied.
These profiles generally show an exponential decrease in detector collection efficiency as the angle from the normal increased from 30 • to 135 • (PD2 to PD9).However, for detectors at angles greater than 100 • (PD7 to PD9), the higherangle detectors exhibit a slightly increased calibrated signal magnitude, which is unexpected based on the pre-design cell modeling shown in Fig. 3 (right inset).This effect is clearly shown in Fig. 8, where the calibrated signal magnitudes are plotted as a function of the detector angular position, rather than as a function of the measurement time point, for four ink compositions.The ink compositions for which the results are presented in Fig. 8 consisted of stock Ag-NP ink (Novacentrix JS-A426), the same ink was diluted at an ink-to-water ratio of 3:1 (dilution 1), the same ink diluted at an ink-to-water ratio of 3:2 (dilution 2), and pure DI water.Despite the unexpected increase in higher angle scattering, the expected relationships between the ink composition, optical scattering wavelength, and measured sensor response, based on optical cross-section calculations, such as those shown in Fig. 2, are supported by the results shown in Fig. 8.In Fig. 8, a higher scattering signal is observed for aerosols with a higher water content, along with a higher scattering signal for the 810 nm LED compared to the 470 nm LED.The increase in the detector magnitude was most likely due to the imperfect collimation of the input illumination.Specular reflection off the edges of the optical cell PD0 aperture was observed, indicating that the cell did not sufficiently trap input illumination left unscattered by the aerosol, resulting in high-angle reflection directed primarily to the higherangle PDs.Modeling in Zemax with the divergence of the input collimation increased to 9 • was performed, resulting in a qualitatively equivalent higher-angle scattering signal.The results of this model are presented in Fig. 9, where the optical properties input to the Zemax model are µ a = 0 mm −1 , µ s = 0.015 mm −1 , and g = 0.65.
These absorption and scattering coefficients and the anisotropy factor were calculated for pure DI water aerosol, a CGFR of 40 ccm, and a deposition rate of 0.01 mm 3 /sec, illuminated by the 810 nm center wavelength LED used in this prototype cell.

V. DISCUSSION
Here, the motivation and design approach for developing an improved aerosol sensor to monitor the optical properties of AJP aerosolized ink, with sensitivity to both the wavelength and angle dependence of the aerosol's optical scattering, is described.The system-level ability of the sensor to respond to changes in the CGFR, which corresponds to changes in the ink aerosol density [25] as well as to changes in the ink composition, was demonstrated.It is also important to evaluate whether the sensor design can be expected to provide useful inputs to future AJP controllers.While a future AJP control system may not require the translation of the optical sensor output to physical variables, if, for instance, a machine learning model ingests raw optical sensor data to generate AJP controller setpoints, it is expected that the sensor output must be capable of differentiating between optical absorption and scattering changes independently, such that an increase in the absorption coefficient is not misinterpreted as a decrease in the scattering coefficient.
To test whether the data generated by this prototype sensor design, consisting of ten measurement points at two wavelength bands, is sufficient for differentiating between scattering and absorption changes, a synthetic dataset was generated in Zemax, with µ a varying from 0 to 0.09 mm −1 , and µ s varied from 0.001 to 0.097 mm −1 , for 1600 simulated combinations.Using this dataset in a 50% train-test split, the ability of the LinearRegression scikit-learn library with default parameters to extract µ a and µ s from the ten-detector, two-LED type data, was tested.The results of this test are presented in Fig. 10 exhibit mean-squared-error (MSE) well below 0.5% for both absorption and scattering.Independent inference of µ a and µ s for the AJP aerosol may provide useful information about aerosol changes, even without directly fitting these values to aerosol ink chemical properties or droplet distribution properties by considering the known optical properties of the ink and the optical property relationships highlighted by Eq. 1-3.For example, consider an aqueous Ag-NP ink and the absorption and scattering cross-section relationships described in Fig. 2. In this case, an increase µ s while µ a increases at a much lower rate, may indicate an upward shift in the droplet diameter distribution.Similarly, a decrease µ s along with increasing µ a is more likely indicative of a decrease in the solvent (water) content.Finally, a case where both µ s and µ a increase proportionally to one another most likely indicates a shift purely in the number density of the aerosol, without a change in the composition or droplet distribution.
While discrepancies between the modeled predictions and experimental results point to the need for further model development and validation, these preliminary results indicate that the approach to modeling the optical sensor is valid, and with improved calibration between the model and experiment, extraction of absorption and scattering properties from measured results is achievable.Future work should include tests on optical property measurements extracted from measured aerosols with known and controlled optical properties across a range of droplet densities and size distributions.One important assumption made in this work to model the sensor using a Mie model relying on µ s , µ a , and g is that the aerosol population follows a log-normal distribution defined by the mean and variance.This sensor simultaneously samples multiple aerosol particles for each measurement, and the measurements may be fitted to accommodate other particle distributions, a priori knowledge of the form of the distribution is required to assign relative weights to the dropletsize-dependent scattering and absorption efficiencies.
The system presented herein evaluated the use of two wavelength bands of illumination and nine detection angles, but future work may benefit from integrating additional wavelength light sources as well as a higher density of detection angles.To evaluate compatibility with commercial manufacturing applications, additional future work may focus on ensuring that the sensor is stable over much longer print times than were investigated as part of this proof-of-concept work.For example, negligible ink condensation within the optical sensing cavity is a requirement for accurate performance of the sensor and while condensation was not observed during the execution of the experiments described herein, these experiments were all completed with print times of less than one hour.Validation of stability over many hours is expected to be a requirement for commercial viability.

FIGURE 1 .
FIGURE 1. Conceptual diagram of the AJP process highlighting the aerosol generation, transport, and deposition stages process.The aerosol sensor developed for this work measures the polydisperse aerosolized ink droplets at the end of the transport stage, immediately prior to the aerosol entering the deposition head.

FIGURE 2 .
FIGURE 2. Absorption and scattering cross-sections for modeled silver AJP ink for varied ink composition and droplet distributions.In the top row (A), the effect of the droplet distribution median diameter varied from 2.91 µm to 3.11 µm is shown.In the bottom row (B) the effect of varied water content in the model aqueous ink is shown for a fixed droplet size distribution.

FIGURE 3 .
FIGURE 3. (Left) Wireframe model of initial design simulations, where the orange wireframed cylinder has optical properties of a water aerosol, and the blue lines are ray traces for collimated input light.(Right) Scattered power detected as a function of angle for the modeled configuration on the left for two LEDs with peak wavelengths at 470 nm and 810 nm, showing the slight but discernable difference in scattering between these wavelengths.

FIGURE 4 .
FIGURE 4. Annotated CAD diagrams of the optical cell.In the top diagram, the outside of the cell is shown with the location of the aerosol transport inlets noted.The bottom diagram shows a cross sectional view of the cell, with each of the photodiodes (PDs), input light, and reference angle θ annotated.

FIGURE 5 .
FIGURE 5. Block diagram for the illumination control and optical reflectance data acquisition system.The inset on the left shows the measured, normalized emission spectrum for LEDs 1 and 2.

FIGURE 6 .
FIGURE 6. Uncalibrated optical cell detector TIA voltage readings.The step changes in detectors are due to step-changes in the CGFR, varying between 20 ccm and 35 ccm, and 0 ccm when response is near zero.

FIGURE 7 .
FIGURE 7. Calibrated responses of the optical sensor's 9 PDs to varied CGFR.In (A), the top two panels show the signals in log-scale as a function time, with the 470 nm and 810 nm responses in blue and red tones, respectively.The CGFR flow applied is shown in linear scale on the bottom pane of (A).In (B), averaged values of the sensor signals corresponding to the four plateaus of CGFR plotted in (A) are shown as functions of the CGFR.

FIGURE 8 .
FIGURE 8. Calibrated responses of the optical sensor to varied ink compositions at a fixed CGFR.Dilutions 1 and 2 are Ag-NP ink with water added at ratios of 3:1 and 3:2 ink-to-water, respectively.

FIGURE 9 .
FIGURE 9. Modeled scattering signal vs detector angular position for optical properties near those expected for water ink at CGFR of 40 ccm and deposition rate of 0.01 mm 3 /sec, illuminated by the 810 nm center wavelength LED used in this prototype scattering cell.

FIGURE 10 .
FIGURE 10.Linear regression model prediction of absorption and scattering coefficients vs true synthetic values for a test of 800 simulated sensor responses to varied AJP aerosolized inks.