1 Introduction

A major advantage of additive manufacturing (AM) is the capability of producing devices and components at small prototyping scales with a short turnaround time so that materials and processes can be quickly evaluated. Considerable advances have been made with the AM of ceramic materials in a diverse set of applications including passive structural components, transducer devices, and dielectric materials [1,2,3,4]. In the direct-write deposition of ceramics called “Robocasting” a slurry containing ceramic particles is extruded from a computer controlled printer to form a desired structure, dried, and then sintered [5]. The additive manufacturing of metal components has also been demonstrated for biomedical, [6, 7] mechanical, [8] and electronics applications [9]. A variety of techniques are available including powder bed fusion, extrusion, and jetting of the material [8, 10]. A number of groups have used AM in the context of sensors to deposit organic or polymer components for sensing volatile organics, [11] charged molecules and surfactants, [12] and H2S [13]. However, the additive manufacturing of ceramic components for mixed potential sensors is an area which has not been previously investigated.

Mixed potential sensors consist of two dissimilar electrodes bridged by a solid-state electrolyte. The operating principle of a mixed potential sensor is the establishment of a difference in the mixed potential at each electrode due to the differences in catalytic activity towards oxidation and reduction of species in the gas stream [14, 15]. Mixed potential sensors are a promising technology for pollution monitoring because of their demonstrated durability in exhaust gas environments, rapid response time, and tunable selectivity to various gases of interest including NOx (NO2 or NO), hydrocarbons, CO, and NH3 [16,17,18]. The selection of materials is key to targeting the selectivity of the devices; for example, NOx and hydrocarbons are most readily sensed by La0.87Sr0.13CrO3 (LSCO)/Pt electrodes while Au and Au alloy electrodes are most selective towards CO [16, 19,20,21]. Three-electrode devices have been demonstrated for use in the detection and quantification of EPA regulated automotive exhaust gases in sensors with a planar geometry similar to commercially available lambda O2 sensors [17, 22]. These devices were produced with a screen printing technique that scales well for mass production, but the materials waste for precious metals like Pt, the additional effort needed to create screen printing patterns for each new sensor geometry, and the turnaround time of this process makes this process suboptimal for rapidly prototyping small numbers of devices. In principle, additive manufacturing substantially decreases both cost and time for rapidly prototyping of custom sensors with a variety of monitoring and feedback control applications.

Variability in the performance of sensors may arise due to inconsistencies across the extrusion process and heat treatment process of each device. There is a need to develop an understanding of how materials parameters including the diffusivity through the porous electrode and the roughness of the electrodes affect the performance of the mixed potential sensors. Aqueous electrochemical tests can easily detect differences in electrode surface area and diffusivity in commercially screen printed and additively manufactured sensor devices. Double layer capacitance is used to measure the surface area of Pt electrodes and the diffusion of ferri/ferrocyanide species is used as a probe for differences in diffusion through the porous layer [23]. In this paper, we report on the manufacturing of LSCO/YSZ/Pt sensors by direct-write deposition of electrode components on a laminated substrate of YSZ. The sensitivity to NOx, C3H8, and NH3 and its variability were assessed and correlated with impedance spectroscopy, Pt electrode surface area, and diffusivity through the YSZ electrode.

2 Experimental section

2.1 Fabrication of mixed potential sensors

Mixed potential sensor electrodes and electrolytes were deposited on a laminated substrate of green YSZ (3 mol% Y2O3) tape (ESL 42000) using a computer controlled extrusion system. The Pt electrodes and leads were deposited as an ink (ESL 5570). The ink was extruded through a nozzle with a 125 µm tip at a table speed of 2 mm/s. The substrate and Pt were co-fired at 1450 °C. The LSCO electrode was deposited as a paste of LSCO powder (Praxair) suspended in a vehicle and thinner (ESL 473 and 401). The LSCO paste was 65 wt% solids and had a vehicle:thinner ratio of 4:1. The LSCO powder, vehicle, and thinner were mixed in a Thinky planetary centrifugal vacuum mixer at 2000 rpm and 85 kPa for 1 min. The paste was then 3-roll milled to break up any agglomerates. The paste was extruded through a nozzle with a 125 µm tip at a table speed of 3 mm/sec. The LSCO electrode was then sintered at 1200 °C. Finally, the porous YSZ electrolyte was formed by extrusion of a YSZ (Tosoh, TZ-3YS) paste. The YSZ paste was 65 wt% solids, had a vehicle:thinner of 4:1, and was processed in the same manner as the LSCO paste. The YSZ paste was extruded using the same parameters as the LSCO paste and was then sintered at 1100 °C. Sintering schedules are available in the Electronic Supplementary Information, Table S1. A schematic and photograph of a sensor are shown in Fig. 1. Ten devices were constructed on a single substrate and sectioned with a diamond wafer cutting saw. Device performance is also compared with a commercially prepared 3-electrode LSCO/AuPd/Pt sensor from ESL ElectroScience.

Fig. 1
figure 1

a A schematic of the two electrode LSCO/YSZ/Pt device and b a photograph of the sensing element of a completed device

Full size image

2.2 Gas sensitivity tests of sensors

The sensors were mounted on a holder obtained from a commercial Bosch Lambda O2 automotive sensor unit and contacts were made by pressing Ni wires against the Pt contact leads. The wires were passed through an alumina tube and the device was placed in the center of a 1′′ quartz tube within a Lindberg Blue tube furnace. Each sensor was heated to a temperature of 500 °C and allowed to remain there for at least 1 h prior to testing. Gas mixing and delivery were provided by an Environics 2000 gas mixer controlled with LabVIEW. The sensors were exposed to a base gas mixture containing 10% O2 and 2.5% CO2 with balance N2 at a flow rate of 180 SCCM. Test gases of 200 ppm of NO, NO2, C3H8, and NH3 were then introduced into the gas stream. The measurement started with a 30-min stabilization period under base gas, and each test gas was introduced in 10-min intervals, followed by a 10-min purge with base gas. Voltages were recorded between the electrodes using a Keithley 2400 digital sourcemeter under open circuit conditions or with an applied bias of − 5.0 μA. The polarity convention used is LSCO as the positive electrode and Pt as the negative electrode.

2.3 Other electrochemical testing of sensors

Electrochemical impedance spectroscopy (EIS) and cyclic voltammetry (CV) were performed using a PAR 2273 potentiostat in two-electrode configuration with the LSCO as working electrode and Pt as the counter and reference electrode. CVs were obtained between ± 0.3 V at a scanrate of 25 mV/s. EIS was performed with a 10 mV perturbation and frequency range from 100 mHz to 2 MHz. The EIS data were fit to the equivalent circuit model shown in Fig. 2. Impedance measurements were acquired in base gas mixtures and each of the test gas mixtures at 500 °C. One sample was also tested at temperatures between 300 and 525 °C and fit using a modified form of the impedance model where the C2 capacitor is replaced with a constant phase element. The resistance of R2 was converted to conductivity using a rectangular channel geometry. The relationships between the conductivity (σ), Boltzmann’s constant (k), absolute temperature (T), an empirical pre-exponential constant (A), and the activation energy (Ea) are given in Eq. 1 [24]. The rearrangement of Eq. 1 into logarithmic form in Eq. 2 can be used to extract the activation energy from the slope of an ln(σ × T) plot vs. reciprocal temperature.

Fig. 2
figure 2

Electrochemical impedance spectroscopy equivalent circuit model used to fit the impedance data collected on the printed sensors. R represents a resistor, C represents a capacitor, and Q represents a constant phase element

Full size image
$$\sigma =\frac{A}{T}\exp \left( { - \frac{{{E_a}}}{{kT}}} \right)$$
(1)
$$\ln \left( {\sigma \times T} \right)=\ln \left( A \right) - \frac{{{E_a}}}{k}\left( {\frac{1}{T}} \right)$$
(2)

Aqueous electrochemical tests were performed after all gas sensing experiments were completed to ensure no contamination by these experiments. The sensors were prepared by first contacting Ni wires to the Pt electrode of the sensor using an Ag conductive epoxy (Creative Materials 251) and insulating the sample with at least four coats of clear nail polish so that only the Pt electrode and the YSZ on top is exposed to the solution. Electrochemical measurements were taken in a three-electrode configuration with the PAR 2273, a Pt coil counter electrode, and an Ag/AgCl (saturated KCl) reference electrode. The sensors were tested first in a solution of 100 mM H2SO4 to obtain the double layer capacitance of the Pt electrodes. A set of 20 CVs were collected between − 0.25 and 1.0 V vs. Ag/AgCl to clean the surface and then 20 CVs were collected at varying scan rates between 0.5 and 0.7 V vs. Ag/AgCl and 50–300 mV/s. The double layer capacitance is given by Eq. 3 where I is the current, CDL is the double layer capacitance, and \(\nu\) is the scanrate.

$$I={C_{DL}}\nu$$
(3)

The double layer capacitance of Pt with a conversion factor [25] of 20 μF/cm2 was used to assess a roughness factor, the ratio between the real surface area and the geometric surface area. The sensors were then cleaned with DI water, and placed in a solution of 100 mM K3Fe(CN)6, 100 mM K4Fe(CN)6, and 1M Na2SO4. 20 scans were collected with a potential window between − 0.25 and 0.75 V vs. Ag/AgCl at scan rates between 50 and 300 mV/s. The peak in oxidation and reduction of the ferri/ferro-cyanide species were fit to the Randles–Sevcik equation (Eq. 4) [26] to obtain the effective diffusion coefficient through the porous layer.

$${j_p}=2.68~x~{10^5}~{n^{\frac{3}{2}}}{D^{\frac{1}{2}}}C{v^{\frac{1}{2}}}$$
(4)

In Eq. 2, n is the number of electrons transferred, D is the diffusion coefficient, C is the concentration of the redox species in solution, and ν is the scanrate.

3 Results and discussion

3.1 Distribution of sensor response

Figure 3 shows a plot of the distribution of sensor signals for the ten sensors. While the polarity of the measured signals are consistent with LSCO/Pt devices we have previously studied, the signals observed are weaker by a factor of 2–5× compared with the commercially prepared ESL devices such as the 3-electrode LSCO/AuPd/Pt device [17]. The sensors under current bias also show neither the enhancement of NOx sensitivity nor the suppression of C3H8 sensitivity. Furthermore, there is a broad distribution in the responses with the NO2 test producing a response between 5 and 15 mV response and NH3 signals spanning − 10 to + 10 mV.

Fig. 3
figure 3

The distribution of sensor signals in a unbiased mode and b under a − 5.0 muA bias. Signals are approximately 2–5× smaller than planar ESL sensors

Full size image

3.2 Electrochemical impedance spectroscopy

To gain insight into the difference between the ESL sensors and our manufactured devices as well as study the influence of impedance on sensitivity, electrochemical impedance spectroscopy and cyclic voltammetry were performed. Figure 4 shows the impedance spectra taken on one of the sensor samples. There are three time constants observed: a semicircular arc at high frequency situated between 1 and 5 kΩ, a depressed arc at intermediate frequency between 5 and 7 kΩ, and a large depressed arc at low frequency. In comparison to the impedance taken on ESL’s sensor (Supplementary Information, Fig. S1), the impedance of our printed devices is an order of magnitude smaller. A key difference between the printed sensors and ESL’s devices is the presence of an insulating Mg–Al–O spinel beneath the electrodes which blocks the transport of oxygen ions through the substrate as illustrated in Fig. 4b. The impact of the higher conductivity is found in the CV measurements shown in Supplementary Information, Fig. S2 where under the same applied bias, an increase in current by a factor of 20× is observed in our printed sensors. The current enhancement by introduction of NOx is only on the order of 0.1–0.3 μA and results in poor signal to noise ratio under bias in the absence of the insulating layer. We also observe an Ohmic resistance absent in the ESL sensors since the intercept of the high frequency arc with the real axis by approximately 1 kΩ. This may indicate a smaller particle size of the Pt contributing a greater grain boundary resistance.

Fig. 4
figure 4

a Impedance spectra of an additively manufactured sensor and b a schematic showing potential conduction paths in the absence of (top) and in the presence of (bottom) an insulating layer between the electrodes and the YSZ substrate. In the former, oxygen ions can conduct through both the substrate and the top porous layer. In the latter, only transport through the top porous layer is permitted

Full size image

The physical interpretation of the electrochemical impedance analysis is constructed from the equivalent circuit model in Fig. 2. Table 1 shows the results of fitting the same equivalent circuit model to all ten manufactured sensors in base gas and with the addition of 200 ppm of each test gas. The R2/C2 and R3/Q3 elements are not affected by the selection of gases, so these are likely to be associated with ionic transport through the YSZ electrolyte. R4/Q4, however, decreases in the presence of test gas and is smallest in the presence of NH3. The R4/Q4 equivalent circuit elements are assigned to the impedance associated with electrochemical reactions in the gas phase. The R2/C2 circuit element contains the smallest variability and is smaller on average in magnitude than R3/C3. The R2/C2 and R3/C3 elements are associated with the conduction through the substrate and the porous YSZ electrolyte, respectively, since a smaller resistance is expected for transport through a dense structure compared with a porous one. Measurements of the resistance of the R2 element were performed over a temperature range of 300–525 °C, and the conductivity was plotted in Fig. 5. The conversion from resistance to conductivity was performed by approximating a rectangular channel formed by the two Pt contact leads. The slope of the line (R2 > 0.999) in Fig. 5b yields an activation energy of 1.04 eV. This value is close to the value reported by Tuller (0.93 eV) for polycrystalline YSZ and is in support of the assignment of the R2/C2 circuit element with conduction through the substrate [27].

Table 1 Summary of parameters (minimum, maximum, average, and standard deviation) obtained by fitting the equivalent circuit in Fig. 1 to impedance spectroscopy on the sensors
Full size table
Fig. 5
figure 5

a Impedance spectra collected on a printed sensor at temperatures from 300 to 525 °C. b The natural log of conductivity associated with R2 is plotted with respect to reciprocal temperature

Full size image

The signal strength for each test gas as a function of the R4 resistance in base gas is plotted in Fig. 6. As the R4 resistance increases, the magnitude of the signal also increases. This indicates that the impedance of this circuit element can be used as a quality control parameter for the signal response of additively manufactured sensors.

Fig. 6
figure 6

The sensor response to 200 ppm of a NO2, b NO, c C3H8, d NH3 is found to be linearly correlated with the log of the R4 resistance in base gas. R2 represents the goodness of fit parameter for a linear fit through the data

Full size image

3.3 Diffusion through the porous electrolyte

While impedance can be readily measured and used as a quality control parameter, an underlying physical reason for the difference in the signal response of our sensors is sought. Aqueous electrochemical measurements were used to assess the Pt electrode surface area by double layer capacitance and the effective diffusion coefficient by obtaining CVs in a solution of 100 mM K3Fe(CN)6, 100 mM K4Fe(CN)6, and 1 M Na2SO4. Typical sets of cyclic voltammograms for scan rates from 50 to 300 mV/s and the fit to the Randles–Sevcik equation (Eq. 4) are provided in Fig. S3 in the Supplementary Information. The effective diffusion coefficient varies from 10−8 to 10−10 cm2/s, though this level of variability is also observed in screen printed devices [23]. The effective roughness factor determined by the double layer capacitance technique in 0.1 M H2SO4 for the Pt exposed through the YSZ varies from 5 to 65, and indicates either that the Pt does not sinter uniformly across these samples or there is a variation in the amount of Pt exposed beneath the YSZ. Figure 7 shows the correlation of signal response to (a–d) the effective diffusion coefficient, and (e–g) the Pt electrode roughness. The signal associated with NOx and C3H8 are found to be positively correlated to lower diffusion coefficient and lower Pt surface area, while the NH3 signal is unaffected by these two parameters.

Fig. 7
figure 7

Increased signal magnitude is observed for NOx and C3H8 with the a–d a decrease in the log of the diffusion coefficient for Fe(CN)63+ oxidation and d, e an increase in the roughness factor of the Pt electrodes

Full size image

The increased roughness factor of Pt could result in greater heterogeneous catalysis of the gas species, for example, decomposing C3H8 into the more easily oxidized CO, which then readily undergo electrochemical reactions at the three-phase interface. Similarly, a more tortuous path through YSZ increases the contact time with the electrolyte, resulting in a higher amount of heterogeneous reactions on the electrolyte prior to reaching the three-phase interface. An alternate explanation for the dependence of the signal on the effective diffusion coefficient is illustrated schematically in Fig. 8, shown for the case of an oxidized gas like C3H8. The concentration of test gas species is 1000× smaller than the O2 concentration in the gas stream, and the electrochemical reaction is expected to be diffusion limited [15]. We assume the oxygen reduction reaction is never diffusion limited and the electrochemical kinetics for Pt are more facile than they are on LSCO. Consider the case of an oxidized gas such as C3H8, where three distinct regions appear in Fig. 8. In the limit of high diffusivity indicated by (a), neither electrode is diffusion limited nor the mixed potentials of both electrodes are determined entirely by the electrochemical kinetics. In regions indicated by (b–e) the LSCO electrode is under kinetic control, while the Pt electrode’s mixed potential is determined by the intersection of the diffusion limiting current of the oxidized gas with the oxygen reduction reaction current. Within this region the magnitude of the difference in mixed potential between the two electrodes increases. Finally, in the region noted by (f), the mixed potential of both electrodes is controlled by diffusion, and the difference in mixed potential tends towards zero.

Fig. 8
figure 8

A qualitative schematic of the mixed potential system showing a region where a decrease in diffusion coefficient results in a increase in the difference between the mixed potential of the two electrodes. a Neither electrode is under kinetic control and the difference in mixed potential is independent of diffusion. b–e The mixed potential of Pt is determined by the interception of the diffusion limiting current of OX with ORR while the mixed potential of LSCO is fixed under kinetic control. In this region slower diffusion implies a larger difference in mixed potential. f Both electrodes are under diffusion control and the difference tends towards zero

Full size image

4 Conclusion

We have successfully prepared Pt/YSZ/LSCO mixed potential sensors by AM and assessed their sensitivity to NOx, C3H8, and NH3. The current bias effectiveness and signal magnitude were diminished in the absence of an insulating protective layer to isolate the electrodes from the substrate. Variability within the batch of sensors was studied using impedance spectroscopy, and the gas reactions impedance (R4) was found to be a useful quality control parameter to predict the signal response. The surface area of Pt and the effective diffusion coefficient through the porous YSZ electrolyte were correlated to signal response, and mechanisms involving heterogeneous catalysis of the electrolyte and electrode components as well as the influence of diffusion limiting current on the difference in mixed potential of the two electrodes were proposed.