Conceptual Validation of a Near-Infrared Multi-Wavelength Scattering-Absorption Hybrid Sensor for Water Quality Monitoring

2.1 Wavelength selection and physical rationale

Fig. 1 shows the absorption spectrum of pure water over the wavelength range of 400-1900 nm, together with the three wavelengths selected in this study. According to the experimental data reported by Pegau et al. and Segelstein et al., the water absorption coefficient, a(λ), varies by more than four orders of magnitude across this range and exhibits distinct absorption bands associated with overtones and combination bands of the O–H vibrational modes [14,15]. Table 2 summarizes the water absorption coefficient at each selected wavelength and its corresponding optical role in the proposed sensor system.

Fig. 1.

NIR absorption spectrum of pure water and the three wavelengths (660 nm, 850 nm, 1450 nm) selected in this study.

Table 2.

Water absorption coefficient and optical role of each selected wavelength.

Notably, the absorption coefficient of water at 1450 nm is approximately 700 times greater than that at 850 nm. For an optical path length of 10 mm, the transmittance of pure water at 1450 nm is about 5.5 × 10⁻¹³, indicating that the incident light is almost completely absorbed; absorption by air over the same path length is negligible. This substantial difference in absorption forms the physical basis for the wavelength-selective sensitivity to changes in the medium composition along the optical path. Because this principle relies on the intrinsic absorption characteristics of water, it may be broadly applicable to water-based monitoring systems, including those involving ultrapure water, drinking water, wastewater, and process water.

Fig. 2 illustrates this principle using three representative scenarios. In Panel (a), representing pure water, the optical path is completely filled with water, so light at 1450 nm is strongly absorbed and the signal reaching the detector is very low. In Panel (b), representing particle-laden water, the optical path is still occupied by water, so the 1450 nm absorption is maintained, whereas the scattering signal at 850 nm increases due to the presence of suspended particles. In Panel (c), representing bubble-containing water, part of the water in the optical path is replaced by air, thereby reducing absorption at 1450 nm and substantially increasing the corresponding detector signal. Thus, an increase in the 1450 nm signal relative to the water-filled condition can serve as an optical indicator of the presence of bubbles.

Fig. 2.

Conceptual illustration of absorption and scattering behavior along the optical path for three scenarios: (a) pure water. (b) particle-laden water. and (c) bubble-containing water.

2.2 Optical system and signal processing architecture

Fig. 3 presents an integrated view of the optical system layout and the signal processing flow of the proposed sensor. Three LED sources (850 nm, 1450 nm, and 660 nm) are driven sequentially by time-division multiplexing (TDM) and combined into a single optical path through a beam combiner. The combined beam is then divided by a beam splitter into two paths: one is guided directly to the reference detector without passing through the sample, while the other enters an in-line flow chamber.

Fig. 3.

Integrated optical layout and signal-flow block diagram of the proposed three-wavelength time-division LED system with a sample/reference ratiometric dual-detector architecture.

Inside the sample chamber, light scattered by suspended particles or bubbles is collected by the sample detector positioned at 90° with respect to the incident beam. To ensure sufficient sensitivity at 1450 nm, extended-range InGaAs photodiodes covering the 900–1700 nm spectral range are employed. Sapphire or quartz optical windows are used to provide sufficient transmission at 1450 nm. The in-line flow chamber is designed with a standardized flange interface to facilitate integration into various water-industry systems, including drinking- and drainage-water pipelines, aeration tank outlets, bottling lines, ultrapure-water piping, and pipes located upstream of membrane modules.

To compensate in real time for measurement errors arising from source-intensity fluctuations, optical-window fouling, and temperature drift, parallel sample and reference channels are configured for each wavelength. This sample/reference configuration is based on an established architecture commonly employed in NDIR gas sensors and spectrophotometers [16,17]. The measured signal is normalized as

This configuration is designed to compensate for three major sources of measurement error. First, variations in LED output caused by temperature and aging are common to both channels and thus cancel in the ratio. Second, attenuation caused by fouling of the sample-side optical window is partially cancelled, thereby extending maintenance intervals; this is particularly beneficial in wastewater treatment facilities and raw-water intake sections, where optical-window fouling can be significant. Third, because both detectors share the same thermal environment and drift together, temperature-dependent drift can also be reduced.

2.3 Definition of the discrimination index

By extending Beer–Lambert law to account for scattering, the intensity of the signal reaching the 90°-scattering detector after propagation through an optical path of length L can be expressed as

where a_w is the absorption coefficient of water, b_p is the scattering attenuation coefficient associated with particles, b_b is the scattering attenuation coefficient associated with bubbles, and f_w is the volume fraction of water in the optical path (f_w < 1 when bubbles are present).

Using the normalized signals of the three wavelengths, the bubble–particle discrimination index R is defined as

When bubbles are present, f_w becomes less than unity, thereby reducing the water-absorption contribution at 1450 nm in Eq. (2) decreases sharply; S(1450) therefore rises relatively and R increases. When only particles are present, f_w remains approximately unity and both wavelengths undergo similar increases in the scattering component, so R remains close to its baseline value. This wavelength ratio is robust against variations in source intensity, optical path length, and absolute turbidity level of the medium, which supports the expectation of consistent discrimination performance across the wide range background turbidity levels encountered in different applications (for example, ultrapure water < 0.1 NTU, raw water < 100 NTU, and sewage < 1000 NTU).

3.1 Simulation setup

Based on Eqs. (2) and (3), wavelength-resolved signal responses were calculated for the three simulated scenarios. Table 3 summarizes the simulation parameters and conditions. A turbidity range of 1–500 NTU was considered to represent a broad range of water quality conditions relevant to water-industry applications.

Table 3.

Optical Modeling scenarios and conditions.

The wavelength-dependent water absorption coefficients were obtained by interpolation of the experimental data reported by Pegau et al. [14]. The particle scattering coefficient was derived from a Mie-theory approximation [18] over a log-normal particle-size distribution (0.5–10 μm). An optical path length of L = 10 mm and a scattering angle of 90° were assumed.

3.2 Data processing

For each scenario, N = 80 repeated calculations were performed, with simulated measurement noise corresponding to a standard deviation of 5% added to each calculated signal to approximate hardware-related variability. The discrimination index R was normalized by the median value obtained for the pure-water scenario, so that R₀ = 1. The threshold of R_th = 2.0 was selected based on the following rationale. First, as described in Section 2.1, the substantial difference in water absorption between 1450 and 850 nm causes R to increase as the water fraction, f_w, decreases in the presence of bubbles. Accordingly, R_th = 2.0 was selected as an intermediate threshold above the particle-response baseline(R ≈ 1) and below the expected bubble-response range. Second, a simple sensitivity analysis was performed by varying R_th between 1.5 and 3.0 in steps of 0.25, and the resulting classification accuracies are summarized in Table 4. Across this range, the overall accuracy varied by less than 2 percentage points, indicating that the classification performance is not overly sensitive to the exact choice of R_th, and that R_th = 2.0 provides a balanced operating point between false-positive and false-negative rates under the assumed 5% noise level. In field validation, R_th is expected to be fine-tuned according to the background absorption level of the specific water matrix.

Table 4.

Sensitivity of classification accuracy to the choice of Rth.

4.1 Wavelength-resolved response for each scenario

Fig. 4 shows the wavelength-resolved normalized signals for the three simulated scenarios. In Panel (a), representing pure water, all three wavelengths exhibit low baseline signals, with the signal at 1450 nm being the lowest because of the strong absorption by water. In Panel (b), representing particle-laden water, the 850 nm signal increases with increasing turbidity, whereas the 1450 nm signal also increases due to particle scattering but remains lower than the 850 nm signal because of the strong water absorption at this wavelength. In Panel (c), representing bubble-containing water, the decrease in f_w with increasing void fraction reduces water absorption at 1450 nm, resulting in a pronounced increase in the corresponding signal and a wavelength-response pattern distinct from that observed for particle-laden water.

Fig. 4.

Simulated wavelength-resolved normalized scattering signals for the three scenarios (pure water, particle-laden water 1–500 NTU, and bubble-containing water 5–30% void fraction).

The observed wavelength-dependent responses are consistent with the physical principle described in Section 2.1 (Fig. 2). Although Scenarios B and C are difficult to distinguish based on any single-wavelength signal alone, the relative signal ratio between wavelengths provides a basis for discriminating between the two scenarios.

4.2 Distribution of the discrimination index R and classification performance

Fig. 5 shows the distribution of the discrimination index, R, for the three simulated scenarios. Pure water (Scenario A) and particle-laden water (Scenario B) are concentrated near R ≈ 1, whereas bubble-containing water (Scenario C) is distinctly separated in the range R = 5–50. Table 5 summarized the classification performance obtained using a threshold R_th = 2.0.

Fig. 5.

Distribution of the discrimination index R = S(1450) / S(850) for each scenario and the corresponding classification accuracies at the threshold Rth = 2.0.

Table 5.

Simulated classification accuracy at the threshold Rth = 2.0.

The 1.2% misclassification rate observed for Scenario B occurred primarily at high turbidity levels near 500 NTU, where the combination of strong particle scattering and with the assumed 5% measurement noise shifted some R values toward the classification threshold. It should be emphasized that the accuracy values reported here are obtained from simulations using literature-based optical constants and an extended Beer-Lambert model, and therefore represent a conceptual, simulation-based validation rather than a measured performance of an actual sensor. Most water industry applications operate at much lower turbidity than the boundary case (for example, treated water in drinking-water plants < 1 NTU, bottled water < 0.5 NTU, and ultrapure water < 0.1 NTU), so such boundary cases are unlikely to occur in real operation; however, this expectation itself remains to be verified experimentally.

4.3 Discussion and limitations

These simulation results support the conceptual feasibility of distinguish particles from bubbles by incorporating the water-absorption response at 1450 nm into the discrimination index, R, whereas the 850 nm signal alone does not provide sufficient discrimination between the two scenarios. In particular, the fact that R maintains a constant baseline value independent of the absolute turbidity level supports the universality argument, that is, that a single discrimination logic can be applied consistently across the wide range of background turbidities encountered in different water industry applications, from ultrapure water to sewage. However, these findings are based on proof-of-concept simulations using simplified optical models and should not be interpreted as validation of universal applicability or actual sensor performance. Experimental verification using a physical prototype and representative water matrices is therefore required.

Nevertheless, the present modeling has the following limitations. First, and most importantly with respect to the discrimination principle, particles and bubbles were considered as independent scenarios rather than as coexisting scatterers in a single optical path. In real water-industry applications, particles and bubbles may coexist, particularly in aeration tanks, coagulation basins, and bottling lines. Under such conditions, their combined effects on scattering and water absorption may alter the discrimination index, R, relative to the independently simulated scenarios. The present study therefore characterizes only the endpoints of this response space, and the transition behavior under coexistence conditions must be evaluated in dedicated follow-up experiments before the method can be regarded as validated for practical deployment. Second, pure water was assumed as the background medium, but in real applications trace organics, minerals, and color-forming constituents can be present, and their background absorption may affect the discrimination index R and, in turn, the optimal threshold R_th. The expected effects can be summarized as follows. (i) Colored dissolved organic matter, such as humic and fulvic acids in raw or wastewater, exhibits residual absorption tailing into the visible and near-visible NIR, so background absorption at 660 nm and, to a lesser extent, 850 nm may be increased. This tends to reduce the numerator-relative baseline of R and could shift the pure-water and particle-laden distributions slightly downward, effectively increasing the safety margin against the bubble class. (ii) At very high concentrations of organic matter, weak C-H overtone absorption in bands adjacent to 1450 nm may raise the background of S(1450) and thereby elevate the baseline R, which could reduce the effective margin to R_th. (iii) Dissolved minerals such as Ca²⁺, Mg²⁺, and Fe species typically show negligible NIR absorption at the selected wavelengths and are not expected to have a first-order effect on R, although Fe-related color may contribute weakly to (i). For these reasons, R_th should be treated as a matrix-dependent parameter that is calibrated in situ for each application (for example, a higher R_th for organic-rich matrices such as wastewater, and R_th close to the present value of 2.0 for clean process water and ultrapure water). Determination of matrix-specific R_th values will be one of the primary tasks in the follow-up field validation. Third, the water absorption coefficient at 1450 nm depends on temperature at approximately 0.5%/°C [14], and a temperature-compensation algorithm will therefore be required during field validation. Fourth, hardware-related factors including the spectral responsivity and dark current of the InGaAs detector and the spectral bandwidth of the LEDs, were not fully incorporated into the present model. Finally, changes in the Mie-scattering characteristics of bubbles across the size distribution (from several μm to several mm) may influence the choice of R_th, and application-specific quantification through field data will be required.