Low-Cost Resistive Microfluidic Salinity Sensor for High-Precision Detection of Drinking Water Salt Levels.

Rapid, inexpensive, and precise water salinity testing remains indispensable in water quality monitoring applications. Despite many sensors and commercialized devices to monitor seawater salinity, salt detection and quantification at very low levels of drinking water (below 120 ppm) have been overlooked. In this paper, we report on optimization of a low-cost microfluidic sensor to measure water salinity in the range of 1-120 ppm. The proposed design employs two copper microbridge wires suspended orthogonally in a PDMS microchannel to measure salinity based on the electrical resistance between the wires. The preliminary design of the sensor microchannel with a rectangular cross-section width (w) of 900 μm and height (h) of 500 μm could measure the water salinity in the range of 1-20 ppm in less than 1 min with detection sensitivity, limit of detection (LOD), and limit of quantification (LOQ) of 17.1 ohm/ohm·cm, 0.31 ppm, and 0.37 ppm, respectively. Data from the preliminary design was used for developing and validating a numerical model which was subsequently used for parametric studies and optimization to improve the sensor's performance. The optimized design demonstrated an order of magnitude increase in sensitivity (385 ohm/ohm·cm), a 6-fold wider detection range (1-120 ppm), and a 15-fold enhancement in miniaturization of the microfluidic channel (w = 200 μm and h = 150 μm) with LOD and LOQ of 0.39 and 0.44 ppm, respectively. In the future, the sensor can be integrated into a hand-held device to remove present impediments for low-cost and ubiquitous salinity surveillance of drinking water.

Introduction

Water salinity measurement is critical in water quality monitoring and surveillance to preserve water safety.[1] Uncontrolled salinity levels can have detrimental effects on human health, agriculture, industry, and the environment.[2] The salt concentrations of water sources are continually changing due to various natural and human-induced environmental phenomena.[3] Exceeding certain salt thresholds could pose health threats to humans, primarily through drinking water.[4]

Among 14 different salt types,[5] sodium chloride (NaCl) is the dominant dissolved salt leading to water salinity.[5] Sodium plays a vital role in maintaining the osmotic pressure of body fluid and preventing excessive fluid loss. However, high sodium levels in the body could cause hypertension, cardiovascular diseases, renal diseases, and Meniere’s disease.[4,6] High blood pressure alone is the leading cause of death and the second major cause of disability in children worldwide.[7,8] According to the World Health Organization (WHO), acceptable salt intake is recognized to be 5 g/day and 500 mg/day for adults on regular and sodium-restricted diets, respectively.[9] Canada’s national health and welfare department requires a salt limit of 20 ppm or mg/L on drinking water to maintain a sodium-restricted diet, assuming a daily water consumption of 1.5 L.[10]

Sodium levels in Canadian drinking water supplies vary seasonally and geographically.[6] The main sources of water include tap water, groundwater, and surface water with sodium levels in the ranges of 0.3–242,[11] 6–130,[12] and 1–300 ppm,[13] respectively. More than 75% of 2100 studied water supply cases in the U.S. had sodium concentrations below 50 ppm.[14] Salinity levels within these low limits must be monitored frequently, rapidly, and preferably specific to the salt type to prevent drastic effects on humans and the ecosystem.

There are several well-established methods for measuring water salinity, among which measuring the electrical conductivity of water remains the most prevalent.[15,16] Most of the current conductivity-based salinity sensing technologies are restricted to measuring high salt levels at the scale of seawater salinity (ranging from 2000 to 42000 ppm, with a typical standard of 35000 ppm) while being expensive, bulky, and not suitable for on-site drinking water salinity monitoring.[16−18] Several fiber optic sensors[19−22] based on interference, fiber Bragg grating, long period grating, and surface plasmon resonance have been proposed for on-field salinity measurement. A key limitation of this technology is its dependence on the change in the refractive index of water which is not sensitive enough for reliable quantification of salinity at drinking water levels.[23]

Microfluidics allows for low-cost mass production of compact and field-deployable salinity sensors that would require a short time and small sample volume for measurement.[24] Moreover, the surface to volume ratio in microfluidic devices increases substantially due to the scaling laws of physics, which can lead to attaining high levels of sensitivity and low levels of detection. Sun et al.[25] used grayscale readouts from a microscope to detect salinity concentration changes in water on a microfluidic device. This method was restricted to seawater samples as changes at lower salt concentrations did not change the visual readouts. Kim et al.[26] developed an integrated temperature and conductivity sensor for monitoring the product water of reverse osmosis desalination. Their hand-held microfluidic device was calibrated to detect conductivities in the range of 0.33–14.64 mS/cm, corresponding to relatively high salt levels of 165–7320 ppm, which is not suitable for drinking water monitoring. In another effort, Xie et al.[27] presented a salinity sensor based on integrating microfluidics and optical fiber-assisted Mach–Zehnder interferometry. The study showed a high sensitivity detection for a 40–31000 ppm range of NaCl but was inept below 40 ppm, which is needed in the drinking water application. As such, the need to measure low water salinity ranges of 1–120 ppm, which is more relevant to drinking waters, is not fully addressed by existing technologies.

In this paper, we introduce a low-cost and sensitive microfluidic sensor, in two primary and optimized design versions, to detect water salinity levels in the range of 1–120 ppm. Water samples with various NaCl concentrations were infused into a microfluidic channel. The electric resistance of the sample was measured using two copper microwires suspended in the middle of the channel and orthogonal to the flow direction (called microbridges).[28] The measured resistance values were correlated to NaCl concentrations, and the sensor calibration curve was established. Subsequently, a numerical model of the salinity sensor was developed, verified, and validated using preliminary experimental findings. A validated model was then employed for a parametric study of geometrical and physical properties to reach an optimized sensor design with 1 order of magnitude higher sensitivity and 6-fold wider detection range, which was experimentally evaluated. We envision that the integration of this sensing paradigm into a hand-held device would allow for addressing the existing unmet need of low-cost and on-field salinity surveillance of drinking waters.

Materials and Methods

Experimental Study

The experimental setup is shown in Figure a. It consisted of a microfluidic sensor, a DC electrical source-meter (Model 2410, Keithley Instruments Inc., USA), a syringe pump (Legato 110, KD Scientific Inc., USA), a DMIL LED conventional inverted fluorescence microscope (Leica, Germany), and a PC to control and record electrical signals of the source-meter using the manufacturer’s software (KickStart, Keithley Instruments Inc., USA).

Figure 1

Experimental setup and microfluidic salinity sensor. (a) Various components of the experimental setup consisting of the microfluidic sensor, electric source-meter, syringe pump, microscope, and PC. (b) Microfluidic device consisting of a sample flow microchannel with an inlet and an outlet and two copper microwire bridges acting as terminal and ground electrodes. (c) Schematic of the microfluidic device and the geometrical parameters contributing to the salinity sensor’s response, including the channel height (h), channel length (L), channel width (w), interwire distance (g), and microwire diameter (dw). (d) The two wires were connected to the source meter. Samples with different NaCl concentrations dissolved in DI water were flown in the channel, DC electric current was swept between the wires from 10 nA to 1 μA, and the corresponding voltages were recorded to calculate the electric resistances based on ohm’s law. The resistance consists of the solution resistance (Rsol.) and the electrode–electrolyte interface resistances (Rint.).

The microfluidic sensor shown in Figure b,c comprised two polydimethylsiloxanes (PDMS) layers, with mirrored channel designs, cast on 3D printed molds. PDMS prepolymer (Sylgard 184, Dow Corning Co., USA) was prepared with a 1:10 cross-linker to monomer ratio, poured on 3D printed molds containing the microchannel design, and maintained at 80 °C for 2 h. After peeling off the cured PDMS layers, two copper microwires (also called microbridges) with a diameter of 90 μm were placed in the middle of the channel. In the primary and optimized designs of the microfluidic sensor, the device geometries such as channel width (w), height (h), and interwire spacing (g) were changed. Oxygen plasma bonding was used to seal the two PDMS layers and then to bond them to a glass substrate. Afterward, the device was placed on the inverted microscope to monitor the microchannel while conducting the experiments.

Four and three replicates of the primary and optimized microfluidic sensors were manufactured and tested, respectively. Experiments were repeated for each replicate of the primary and optimized devices seven and five times, respectively, to measure NaCl concentration levels at a statistical significance of 0.05. Samples with salinities ranging from 1 to 20 ppm and 1 to 120 ppm of NaCl dissolved in DI water were used in the primary and optimized sensor studies, respectively. The 20 and 120 ppm samples were prepared by dissolving 40 and 240 mg of NaCl, from Sigma-Aldrich Inc. (cat. no. S7653), in 2 L of DI water, respectively. Through serial dilution with deionized water (DI), concentrations of 15, 10, 7.5, 5, 3, 2, and 1 ppm of NaCl were produced and used in the primary device experiments. For the optimized design, serial dilution was performed to obtain 100, 80, and 60 ppm of NaCl in addition to the above concentrations. The samples were infused into the microchannels with the syringe pump at a flow rate of 1 and 0.2 mL/min in the primary and optimized phases, respectively. As the optimized device had a smaller cross-section channel, the sample influx rate was reduced to maintain a similar Reynolds number for both devices.

The terminal and ground copper microbridge wires were connected to the DC source meter. As this study aims to pave the way for developing a hand-held sensor that runs by a battery in the future, a DC supply was used in our tests. The source-meter swept the DC electric current from I = 10 nA to I = 1 μA in 56 s during each experiment while measuring the electric voltage (V) across the copper microwires (Figure d). The corresponding voltages were recorded using the KickStart software. The measured current and voltage values were used to calculate the NaCl sample resistance, R = 2Rint + Rsol in Figure d, using Ohm’s law (R = V/I). Here, 2Rint is the electrode–electrolyte interface resistances of the two wires due to their electric double layers in series with the nonuniform solution resistances near the wires.[29]Rsol. is the bulk solution resistance which is relatively high for the cases in this study with samples containing low salt concentrations, i.e., high resistivities.[30,31] All of the measurements were performed at standard ambient temperature and pressure (approximately 25 °C and 100 kPa, respectively), and the results would be valid for these conditions.

To eliminate the variability between devices, each set of results was normalized with a measured baseline resistance, R0, at 0 ppm of NaCl (blank DI water). The fold change method (R/R0) was used for normalization, and data points at each concentration were divided to the mean resistance value of the baseline.

All of the statistical analyses in this study were conducted using Minitab 20. Levene’s test was applied to examine the homogeneity of variances prior to conventional one-way ANOVA. If the test showed a significant result (p < 0.05), then Welch’s ANOVA was conducted instead of one-way ANOVA. Also, when ANOVA showed a significant difference between the mean values, a post hoc pairwise analysis was performed to identify which pairs of means are significantly different. Tukey HSD posthoc pairwise comparison was conducted for data sets with homogeneous variances, and a Games–Howell post hoc test was adapted for the ones with inhomogeneous variances.

The limits of detection (LOD) and quantification (LOQ) were determined as salinity concentrations with resistances equal to DI water resistance (RDI) minus 3 and 10 times its standard deviation (SD), respectively (i.e., RLOD = RDI – 3SDDI and RLOQ = RDI −10SDDI). The NaCl concentrations at which these resistances would be obtained were found using the calibration curves established for the primary and optimized sensors.

Numerical Study

A numerical model was developed in COMSOL Multiphysics based on the experimental configuration of the primary sensor (Figure c,d) which was then verified and validated against the experimental measurements. Figure c depicts the geometrical parameters of the salinity sensor, such as the channel length (L), height (h), width (w), wire diameter (dw), and interwire distance (g) that were investigated numerically. After validation, the model was employed to find an optimized sensor geometry to increase its sensitivity for salinity measurement applications. The parameters used in this study and their units are summarized in Table .

Table 1

Parameters Used in the Numerical Model with Their Symbols, Descriptions, And Units

symbol	description	units	symbol	description	units
ρ	density	kg/m3	E	electric field	V/m
u	mean velocity	m/s	ρel	electrical resistivity	ohm. m
p	pressure	N/m2	J	current density	A/m2
μ	viscosity	Pa·s	V	electrical potential	V
Re	Reynolds number		σ	electrical conductivity	S/m
Dh	hydraulic diameter	m	Qj	current source	A/m3
A	microchannel cross-section area	m2	Je	external current density	A/m2
P	microchannel wetted perimeter	m	zi	charge number of ions (e.g., +1 for Na+)	C
dw	wire diameter	m	T	temperature	K
w	width (along the z-axis in Figure 1)	m	Cp	specific heat capacity	J/kg·K
h	height (along the y-axis in Figure 1)	m	k	thermal conductivity	W/m·K
L	microchannel length (along the x-axis in Figure 1)	m	Qe	resistive dissipation	W/m3
g	interwire distance (along the x-axis in Figure 1)	m	Qted	thermoelastic dissipation	W/m3
Ji	diffusive flux	mol/cm2.s	Qvd	viscous dissipation	W/m3
Fc	Faraday’s constant	C/mol1	Qp	pressure work	W/m3
um,i	ionic mobility	m2/s·volt	Q	heat source	W/m3
Λ	molar conductivity	S/cm2.mol	R	molar gas constant	J/mol.K
c	molar concentration	mol/L	Ri	reaction rate	M/s
α	correlation factor		F	other forces	N/m3
Λ0	the infinite dilution	S/cm2·mol	I	identity matrix
κ	specific conductance	S/m

The model of the primary sensor consisted of a 2D rectangular microfluidic channel, cross-sectioned across the x–y plane of Figure c at the middle of the channel, with two 90 μm diameter circles representing the wire electrodes in the middle. The microchannel height (along the y-axis) was h = 500 μm. The width of the microchannel along the z-axis was set to w = 900 μm using COMSOL’s out-of-plane thickness feature, which allowed achieving similar results to 3D simulation in a 2D setup. Interwire distance was set to g = 1500 μm. Simulation runtime was minimized by reducing the microchannel length along the x-axis while maintaining the model domain independence (see Figure S1). For the optimized sensor simulation, the geometrical characteristics were h = 150 μm, w = 200 μm, and g = 2000 μm.

The boundary conditions were selected based on the experimental conditions. A fully developed flow profile with a constant flow rate of 1 or 0.2 mL/min was imposed in the x-direction for the primary and optimized sensors, respectively. The actual devices had a longer channel length in the x-direction than the numerical study that allowed for the safe assumption of a fully developed flow. The velocity in the y-axis direction was zero, and a no-slip boundary condition was applied throughout the microchannel and wire walls. A static gage pressure of zero (p = 0) was considered at the channel outlet. The two copper microwires were considered as the electrodes that probe and measure the resistance in the salinity sensor. A ramping DC current of 10 nA to 1 μA was maintained in the terminal while the ground was assumed to have no voltage (V = 0). The PDMS walls were insulated (J = 0).

Two important parameters of the numerical model were the dynamic viscosity and density of the sample. According to previous studies,[32,33] upon adding 10000 ppm of salt to water at 25 °C, an increase of 0.7% and 2% would be observed in the dynamic viscosity and density of water, respectively. These slight changes were safe to be neglected in our models with 1–120 ppm salt concentrations. Therefore, the water dynamic viscosity of μ = 8.90 × 10–4 Pa·s and density of ρ = 1000 kg/m3 were considered in the numerical analyses.

The simulation model was governed by the physics of fluid flow, electric field, transport of diluted species, and heat transfer (discussed in the Supporting Information Section 1). To obtain a better convergence in our simulation, we first computed the fluid flow and electric potential profiles in the microchannel, and then the electrochemical and heat transfer behaviors were developed.

The flow regime in this study would fall into the laminar flow category with the Reynolds number, Re in equation , being 16 < Re < 64. Thereby, the simplified incompressible non-Newtonian continuity equation and Navier–Stokes equation were employed to solve for the electrolyte flow under steady-state conditions.

The hydraulic diameter of the channel was obtained from eq .

The electrolyte sample containing sodium and chloride ions flows at a constant flow rate in the microchannel. The applied electric field between the wires causes the ions to migrate toward the electrodes with the opposite charge, and an electric double layer would be shaped. Steady-state charge conservation eq governs the electric field in the microchannel and electric current passing from the terminal to the ground electrode through the electrically conductive electrolyte sample.

The electrical resistance, R, between the wires was assumed to be related to the sample electrical resistivity, ρel, and other geometrical properties of the channel as defined in eq .

Mass conservation equation was solved to investigate the diffusion, convection, and ionic migration of the dissolved NaCl ions for one or more solutes (i).[34] Fick’s law governs the diffusion term.

The reaction rate is for chemical reactions in the model, which does not apply in our study. The diffusive flux vector in the case of applying the electric field is expressed below.

For a very dilute solution, the ionic mobility of solute i can be obtained from the Nernst–Einstein (eq ).

The Debye–Hückel–Onsager (eq ) was used to calculate the electrolyte conductivity at different concentrations of NaCl below 0.001 mol/L (∼60 ppm).[35] This equation is suitable for very dilute solutions of strong electrolytes (e.g., NaCl) with high solubility in their solvent, as expressed below.

The values of the A and B constants depend on the viscosity and dielectric constant of the solvent, temperature, and charge. In the case of NaCl and water at 25 °C, A = 60.20, B = 0.229, and Λ0 = 126.39 × 10–14.[35] For salinity concentrations between 60 and 120 ppm, a correlation factor of α = 0.55 was used to correlate the salt concentration to electrical conductivity.[36,37] The electrical resistivity for the salty water samples was then calculated from eq .

Results and Discussion

Primary Sensor Experimental Evaluation

We measured the primary sensor’s electric resistances for salty water samples with NaCl concentrations of 1–20 ppm flown at a flow rate of 1 mL/min in the channel, as shown in Figure (four replicate devices, each tested 7 times). The sensor had a transient response at the beginning of the experiment, followed by a steady-state plateau. The transient effect is mainly a result of sudden implementation of an external electric field which is intrinsic to measurement systems.[38] It usually dies out over time and the device would reach a steady state that is suitable for measurement.[38] Another reason for the transient mode presumably resulted from changes in the ionic composition of the medium around the wires when the current was applied. Samples with NaCl concentrations lower than 5 ppm exhibited shorter transient modes (∼10 s), and the duration was increased at higher concentrations (∼25 s). A relatively stable plateau was observed for all of the concentrations once a steady-state regime was reached (after 30 s; see Figure insets). The resistance values in the range of 30–56 s, corresponding to the current range of 0.5 to 1 μA, showed a gradual decrease. Analysis of this data showed that the resistance changes in the last 30 s of the experiments were smaller than the standard deviations (SD) in the same duration (Figure S2). Thus, the recorded data in the 30–56 s of each experiment was inferred as the steady state plateau and used to obtain the resistance means and standard deviations for correlation to NaCl concentration in the samples.

Figure 2

Dose–response measurements of the primary salinity sensor. Electrical resistances were recorded as the current was swept from 10 nA to 1 μA during 56 s. Results are shown for samples with (a) 1, (b) 2, (c) 3, (d) 5, (e) 7.5, (f) 10, (g) 15, and (h) 20 ppm of NaCl in DI water. Samples were flown in the microchannel at a flow rate of 1 mL/min. The insets depict the 30–56 s intervals of the resistance measurements. Each panel consists of 28 recorded measurements from four replicates and seven measurements. Repetition experiments in each plot are represented with different colors.

Figure a illustrates the electrical resistance means and SDs from the four replicated devices at different NaCl concentrations. The results show that different replicates had similar readouts at each concentration as the error bars of replicates overlapped. Results qualitatively demonstrate that measurements were reproducible and reliable with no overlap between the error bars of any two concentrations. However, there are wider gaps between resistances of the samples with lower salinities (e.g., ΔR = ∼1.6 MΩ between 2 and 3 ppm of NaCl vs ΔR = ∼300 kΩ between 15 and 20 ppm of NaCl). To compare the device readings quantitatively, we performed a one-way analysis of variance (ANOVA) test to examine if the sensor could statistically distinguish the difference among the samples. Statistically significant differences between the resistance mean values of different concentrations were observed (p < 0.0001). Games–Howell post hoc test was adopted to conduct a pairwise comparison for different concentrations. An effect size of 3.58 with a statistical power of 0.9 was obtained at a significance level of 0.05 and a sample size of 7 for each concentration. It was then established that the sensor could distinguish different tested concentrations, and the statistical difference for every two concentrations would be sufficiently large with at least seven repetitions.

Figure 3

Dose–response measurements of the primary salinity sensor. (a) Averaged electric resistances of four replicate devices at different NaCl concentrations. To eliminate interdevice variability, we normalized the results of each replicate in (a) with their baseline values at 0 ppm of NaCl, and the normalized mean values and standard deviations are shown versus samples NaCl concentrations and resistivities in (b). The calibration curve fits the experimental data. The numerical simulation results are shown in (c). The error bars represent standard deviations

The raw resistances in Figure a were normalized with respect to measured DI water resistances to eliminate interdevice variabilities, as depicted in Figure b. As shown, the normalized resistance decreases with an increase in salt concentration due to the rise in the number of ions between the two wires. Similar statistical analyses to the ones described for Figure a were performed with the normalized data in Figure b, and identical results were achieved in terms of distinguishing between various salt concentrations (p < 0.0001).

Figure b inset depicts the normalized mean resistances correlated to the electrical resistivities of the samples, calculated according to the Debye–Hückel–Onsager equation (eq ). The samples with lower NaCl concentrations would have higher resistivities. A linear relationship between the resistance and resistivity in eq allows the sensor’s calibration curve to be established as

A goodness of linear fit of 97.84% was achieved for eq . Disregarding the 1 ppm concentration in the curve-fitting process (since it demonstrated more noise than the other concentrations) would increase the goodness of the fit to 99.26%. Thus, the sensitivity of the sensor based on the primary design was 17.1 ohm/ohm·cm. The blank DI water resistance in the primary sensor was RDI = (30.1 ± 7.04) × 106. Accordingly, the limit of detection (LOD) and limit of quantification (LOQ) were calculated to be 0.31 and 0.37 ppm according to 3SD and 10SD methods, respectively.

Numerical Optimization of the Sensor

To increase the sensor’s sensitivity and its detection range, further optimization studies were warranted. For rapid optimization, a numerical model of the primary sensor based on the experimental results of Figure was developed, verified, and validated and then applied in a parametric study to determine the most contributing device parameters.

We simulated several conditions to ensure the results were domain and mesh independent while achieving the least possible computational load in developing the model. The optimum channel length and mesh properties were found, according to Figures S1 and S3, to be 3 mm and 191026 triangular elements, respectively.

To verify the model, we examined several basic physical phenomena that were expected with our numerical method (results not shown). For example, the fluid flow showed a parabolic pattern through the microfluidic channel. The electrolyte conductivity increased at higher salt concentrations, resulting in lower electric resistance values. The electric potential decreased from its initial value to zero from the terminal to the ground. The ions migration was observed, and Na+ and Cl– ions accumulated around the ground and terminal electrodes due to potential differences, respectively. The accumulation of ions around the wires was increased as the salt concentration was raised in the sample. The fluid flow caused the convection of accumulated ions around the wires and reduced the effective ions in the current transfer. The above observations agreed well with the sensor’s physics and verified the correct performance of the model.

We then validated our numerical model by comparing our results with the obtained experimental resistance values from the primary sensor. Figure c depicts the recorded electrical resistance mean values in the experimental study and the numerical study in the range of 1–20 ppm, corresponding to resistivities of 46.4 × 104 to 4.5 × 104 Ω·cm. As expected, the resulting resistances from the simulation show a perfectly linear behavior (R = 2.52 × 105 ρel; R2 = 1). There was a deviation between the two data sets as the simulation did not consider the experimental errors. As the simulation could not completely mimic experimental conditions, a transfer function was defined so that the user could obtain experimental values using the simulated model and compensate for the observed deviation. We derived the transfer function of system by dividing the line equations of the two data sets.

The numerical optimization study was carried out with a full factorial parametric approach. The effects of six parameters with three levels were examined on the sensor performance as tabulated in Table . The minimum channel height and width levels were selected on the basis of fabrication restrictions using 3D printing. The wire diameters were chosen from several available choices in the market. The effects of changing the fluid properties and flow rate were also studied using the Reynolds number. The study involved 36 = 729 combinations, and each combination was simulated for 1–20 ppm of NaCl concentrations.

Table 2

Numerical Optimization Study Parameters and Levelsa

	levels
parameter	low	medium	high	P value
channel height (h) (μm)	150	250	500	<0.001
channel width (w) (μm)	200	500	900	<0.001
interwire distance (g) (μm)	1000	1500	2000	<0.001
wire diameter (dw) (μm)	90	110	130	0.973
electric current (I) (nA)	1	100	1000	1.00
Reynolds number (Re)	16	32	64	0.998

The P values were achieved through a full factorial design for each parameter.

The results of the optimization study are shown partially in Figure . Our simulations showed that the channel width, channel height, and interwire distance had the most significant effects on the sensor’s performance, determined by examining the increase in resistance compared to the primary sensor; these observations were also confirmed by ANOVA test (Table ). Low levels of the channel width and height caused the highest increase in resistance value as the sensing medium was shrunk (h × w would decrease in eq ). On the other hand, the effect of increasing the interwire distance was not as intense as the channel height and width. The reason is that as g increased, there were two opposing effects on the resistance, i.e., an increase as there was more electrolyte with small conductivity and a decrease as the electric field was weakened when the two electrodes moved further apart.

Figure 4

Effect of the most contributing parameters at different levels on the electrical resistance of the sensor. The increase in resistance compared to the primary sensor is plotted. The optimized design parameter configuration should be composed of low levels of channel width and height and a high level of interwire distance, resulting in g = 2000 μm, w = 200 μm, and h = 150 μm.

We found that the electrode diameter, electric current, and Reynolds number had no statistically significant effect on the resistance (P values close to 1 in Table ). Changing the electric current resulted in similar changes in voltage that left the resistance intact. The fluid flow was always in the laminar regime and did not significantly reduce the effective concentration of the ions in the electric circuit to cause changes in the resistance. It was counterintuitive that the wire diameter had no significant impact on sensor performance as increasing the electrode surface was expected to allow for better electron transfer. However, we hypothesize that as the salt concentrations were scarce, the wires surface area was more than enough for the ions to transfer electrons. To investigate this hypothesis, we simulated smaller wire diameters in the range of 1–130 μm and studied the discharged current density on the wire surfaces (Figure S4). There was less than 10% difference in current densities between wire diameters of 90, 110, and 130 μm. This is also in agreement with our findings from Table that wire diameter did not have significant contribution.

According to Table and Figure , the optimized sensor should have the dimensions of g = 2000 μm, w = 200 μm, and h = 150 μm. This finding is in agreement with eq as increasing and decreasing the cross-section area and interwire distance, respectively, would result in a wider range of resistances and, thereby, higher sensitivity. The achieved resistances, calculated through simulation, were plotted against NaCl solution resistivities for the optimized and primary designs in Figure . It is estimated that the sensitivity of the optimized sensor would increase 27-fold compared to the primary sensor. This estimation will be further investigated in experimental analysis in the next section.

Figure 5

Numerically calculated electrical resistance versus saltwater resistivity for the primary and optimized sensors. The numerical model predicts 27-fold higher sensitivities with the optimized sensor compared to the primary one.

Experimental Analysis of the Optimized Sensor

To evaluate the optimized sensor after fabrication, we performed the same experimental procedures as the preliminary device (see Figure S5). Figure a illustrates the electrical resistance means and SDs at different NaCl concentrations in the optimized sensor (dw = 90 μm, g = 2000 μm, w = 200 μm, and h = 150 μm). Similar to the primary sensor, the resistance decreases as the ions population increases, and the electron transfer improves. Also, there are wider resistance variations between replicates of samples with less than 3 ppm of NaCl concentrations. The analysis of variance and Games–Howell post hoc analyses were performed. It was obtained that there is a statistically significant difference between all and each pair of concentrations with a 0.01 significance level for both analyses (p < 0.0001). An effect size of 2.89 with a statistical power of 0.99 was obtained at a significance level of 0.01 and a sample size of 5 for each concentration. The analysis demonstrated that the sensor could distinguish different tested concentrations, and the statistical difference for every two concentrations would be sufficiently large with at least five repetitions.

Figure 6

Dose–response experimental measurements of the optimized salinity sensor. (a) Measured electric resistances of three replicate devices at different NaCl concentrations. To eliminate interdevice variability, the results of each replicate in (a) were normalized relative to their baseline values at 0 ppm of NaCl and the normalized mean values and standard deviations are shown versus samples (b) NaCl concentrations and (inset) resistivities. The calibration curve fit to the experimental data is also shown in the inset of (b). The error bars represent standard deviations.

The raw resistance values (R) in Figure a were normalized to eliminate interdevice variability, as in Figure b. Moreover, the normalized values were plotted against the resistivity to establish the sensor’s linear calibration curve, expressed in Figure b inset and eq .

The goodness of linear regression was found to be R2 = 96.49%. Neglecting 1, 2, and 3 ppm of NaCl concentrations would increase R2 to 99.76%. Comparing the primary and optimized sensors’ calibration curves, we found that the sensitivity increased an order of magnitude to 385 ohm/ohm·cm with averaged data (further discussed in Figure ). The higher sensitivity allowed the sensor’s detection range to be widened by 6-fold, increasing it from 1 to 20 ppm to 1–120 ppm. The blank DI water resistance in the optimized sensor was RDI = (491.2 ± 8) × 106. Accordingly, the LOD and LOQ was calculated to be 0.39 and 0.44 ppm using 3SD and 10SD methods, respectively, which was almost equivalent to the primary design.

Figure 7

Comparison of the dose–response measurements of the primary and optimized sensors in our experimental and numerical analyses. (a) The optimized model has higher theoretical and experimental sensitivities and allows a detection range of 1–120 ppm of NaCl (46.37 – 0.39 × 104 Ω·cm) compared to 1–20 ppm (46.37 – 2.34 × 104 Ω·cm) for the primary sensor. (b) Otimized sensor results are illustrated in the range of 5–120 ppm of NaCl (9.31 – 0.39 × 104 Ω·cm) that has the least deviation between the simulation and experimental results. The inset magnifies the 7.5–120 ppm of NaCl (6.21 – 0.39 × 104 Ω·cm) range for the optimized sensor.

The overall performance of the two sensors was replotted under both experimental and numerical conditions in Figure a. The first three concentrations of NaCl at 1, 2, and 3 ppm (46.37 – 15.49 × 104 Ω·cm resistivity range) had a lower signal-to-noise ratio and brought about the highest deviation from the simulation findings. As previously mentioned, this deviation resulted from simulation ideal intrinsic quality that was not counted in the experimental errors. This deviation is observed as the experiment and simulation results are compared for the case of the optimized sensor (Figure a). However, the simulation results are in better agreement with the experimental findings in the range of 5–120 ppm (corresponding to 9.31 – 0.39 × 104 Ω.cm resistivity range), as illustrated in Figure b, with an average deviation of 12%. As the simulation could not completely mimic experimental conditions, a transfer function was calculated and applied to experimental results. The transfer function was established as a result of dividing eq and optimized sensor numerical equation (R = 6.75 × 106ρel) and is expressed in eq .

As shown in Figure a,b, the slope of the primary sensor fitted line is 17.1 ohm/ohm·cm with the prenormalized data, which in comparison to the optimized sensor slope of 385 ohm/ohm·cm (1–120 ppm) or 692.4 ohm/ohm·cm (5–120 ppm) shows a significant improvement in the sensitivity of the optimized sensor. The optimized salinity sensor demonstrated promising potential in differentiating the samples with NaCl concentrations in the range of 1–120 ppm. The sensor exhibited a LOD of 0.39 ppm, lower than the NaCl detectors reported before. Although previously developed optical salinity sensors have reported lowest detected values of 2,[39,40] 4,[41] 6.7,[20] 10,[42] and 40 ppm,[27] they lack the required resolution to quantify salinity within drinking water ranges. This limitation occurs because a 1000 ppm change in salinity normally causes an infinitesimal deviation in the optical path length.[1] In addition, these sensors were complex, laboratory-based, and costly due to their reliance on prisms or delicate fiber optics but were able to detect nonionic salts as well. Reported conductivity-based sensors offered lowest detected values of 12,[43] 16,[44] 165,[26] and even higher (e.g., 7800 ppm),[45−47] which are not suitable for sensitive detection in the drinking water applications. Also, these sensors involved labor-intensive and expensive fabrication procedures. In comparison, the proposed optimized sensor was fabricated with a simple and low-cost technique which allows for sensing in drinking water ranges on a miniaturized and inexpensive platform.

The current design of the developed microfluidic salinity sensor suffers from three main limitations. First, salinity detection is performed through measuring conductivity which can be interfered by nonspecific ionic entities in the sample. As such, the detection of the current version of the sensor is not selective. Second, the discussed characterizations of the sensors were performed in ambient temperature and pressure and the established calibration curves are only accurate in these conditions. Third, the present sensors rely on syringe pump and source meter that impedes in situ measurement. Given the promising results presented in this manuscript, we are carrying out research toward addressing the above limitations.

Conclusions

In this paper, a low-cost microfluidic sensor was developed and optimized to measure water salinity based on the sample’s DC electrical resistance. The sensor consisted of a microchannel and two microbridge wires, with two different configurations resulted from primary studies and numerical optimizations. Using the primary sensor, we detected low NaCl levels in the range of 1–20 ppm in less than 1 min by measuring the resistance between the two wires. Our findings showed that the primary sensor sensitivity, LOD, and LOQ were 17.1 ohm/ohm·cm, 0.31 ppm, and 0.37 ppm, respectively. We developed, verified, and validated a numerical model against the experimental findings to optimize the sensor and increase its sensitivity and detection range. A parametric study was conducted to establish the most contributing parameters of the device, which were the channel width, channel height and interwire spacing among six investigated parameters. Through experimental analysis with the optimized sensor, it was found that the sensitivity and detection range were increased by 1 order of magnitude and 6-fold (to 1–120 ppm), respectively, while the sensor could be miniaturized further by 15-fold. The experimental evidence showed the sensor’s accuracy and repeatability, making it a promising candidate for different applications such as surveillance of consumed water by individuals with salt-restricted diets. Although the proposed device does not specifically detect sodium chloride, it could be used as a preliminary surveillance system to warn the water consumers or inspection officials to perform further tests on water safety. We envision integrating it into a hand-held sensor for on-site and point-of-care surveillance of salinity levels after miniaturization of the DC source-meter with portable alternatives and replacing the syringe pump with a passive on-chip or a battery-operated peristaltic pump.