Turbidity measurement based on a multi-wavelength spectral sensor

SHENG Tao, XU Wenyan, SHI Shengzhe, LIU Sheng

(1. School of Computer Science and Technology， Huaibei Normal University， Huaibei 235000， China；2. School of Physics and Electrical Information， Huaibei Normal University， Huaibei 235000, China)

Abstract： A method of measuring turbidity based on a multi-wavelength spectral sensor is proposed by using SFH4737 broad-band infrared LED, a multi-wavelength spectral sensor and independently developed data processing software. Combining multiple wavelength data from the sensor, the unitary and multivariate fitting models were constructed to investigate the relationship among light intensity information, absorbance and turbidity, respectively. The turbidity of the actual water bodies was measured separately by using proposed method and a commercially visible spectrophotometer. The independent-samples T test (p>0.05) showed that there was no significant difference between the method in this paper and the standard assay method. The method is simple and inexpensive, and can be applied to the rapid detection of water turbidity, providing a new way of industrial online measurement.

Key words： turbidity measurement； multi-wavelength； multi-fitting； multichannel spectral sensor； spectrophotometry

0 Introduction

Turbidity refers to the scattering of light by suspended objects and colloidal particles, and the absorption of light by solute molecules, resulting in the reduced transparency of a solution[1-2]. Turbidity measurements are essential to water quality detection, as they are a measure of the quantity of suspended objects in a solution, and are the most direct indicator of water quality[3]. Turbidity is most commonly quantified by the nephelometric turbidity unit (NTU)[4-5]. Standard turbidity measurement methods include transmission turbidimetry and scattering turbidimetry.

In transmission turbidimetry, the relationship between light intensity and turbidity follows the Beer-Lambert law[6], which describes the relationship between the intensity of light absorption by a substance at a certain wavelength and the concentration of the substance and its liquid layer thickness[7]. In scattering turbidimetry, turbidity is measured by detecting scattered light at 90° according to Rayleigh’s scattering law. Scattering turbidimetry is a standard international measurement method, and the scattered light is less affected by the size of suspended particles[8].

Various methods for measuring turbidity that utilize the photoelectric principle of measurement have been developed, including the scattering light[9-10], transmission light[11], ratio methods[12-13]and visual turbidity[14]. However, there are two problems with these turbidity measurement methods. First, all of them make measurements by utilizing a single-channel mode, which is susceptible to interference from the chromaticity of the water sample. It limits the application of turbidity measurements. Zhao proposed a camera combined with a multivariate fitting algorithm for solution content determination[15]. Cao developed a digital camera method combined with a multivariate fitting method and neural network to measure the water turbidity accurately and efficiently[16]. Second, the high cost of measurement poses a massive burden for continuous online monitoring of water quality, and thus cannot meet the basic needs of water quality testing[17-18].

To address the limitations of current turbidity measurement methods, a simple method for measuring the turbidity of water was proposed. A multi-wavelength spectral sensor combined with computer data processing software was developed to measure the turbidity of water. Then, the absorbance was obtained according to the Beer-Lambert law, a measurement system based on light intensity and absorbance was established, and a fitting method was constructed. The device’s performance was compared with that of a traditional turbidity measuring device, which can be selected according to requirements of users.

1 Experimental principle

1.1 Transmission measurement principle

The incident beam from the light source passes through lens 1 and then passes through the water sample. The substance absorbs the light, causing the intensity of the light attenuated. The light passes through the filter and lens 2 and is received by the detector. The measuring principle is shown in Fig.1(a). The relationship between the transmitted light intensity and turbidity is following in Eq.(1).

It=I0e-Tl,

(1)

whereItis the transmitted light intensity；I0is the incident light intensity；Tis the attenuation coefficient, andlis the length of the optical path. The relatively simple method of measurement is carried out in the transmission mode, which is also used in this paper to determine the concentration of substances in the solution, requiring only one more mathematical step.

1.2 Scattering measurement principle

When the incident light is emitted from the light source, it passes through lens 1. The particles in the water sample scatter the light. The light then passes through the filter and lens 2. The turbidity can be measured by measuring the intensity of the scattered light in the direction perpendicular to the incident light through the detector. The intensity of the scattered light at 90° to the incident light conforms to the Rayleigh formula (Eq.(2)), and the turbidity of the sample solution can be measured, as shown in Fig.1(b).

(2)

whereICis the intensity of scattered light；Kis the proportionality factor；Nis the number of particles per unit volume；Vis the volume of particles；λis the wavelength of incident light.

(a)

(b)

1.3 Multivariate fitting principle

The multiple linear regression model is an extension of the one-dimensional (1D) linear regression model. Its basic principles are similar to those of the 1D linear regression model except that the computations are more complex. The observed values are used to fit a linear equation describing the relationship between an outcome and two or more characteristics by analysis. The characteristics that have the greatest influence on the predicted outcome are identified, and the correlation between the different variables is obtained.

yi=w0+w1xi1+w2xi2+…+wnxin,

(3)

wherew0,w1, …,wnare model parameters representing the intercept and regression coefficients, respectively；yiis the column vector containing the regression results of allnsamples；xiis the feature matrix.

The loss function can be expressed as

(4)

2 Instrument design

2.1 Structure of instrument system

The measurement instrument consists of a constant light source and its driving circuit, a sample tank, a multi-wavelength spectral sensor and a computer connected to the device via a USB cable. The data acquisition device is placed in a dark box that avoids the bias of experimental data acquisition due to external light sources. The SFH4737 broad-band infrared LED is used as the light source and powered by a driving circuit board.

The cuvette with the sample solution is placed in the sample tank. The multi-wavelength spectral sensor acquires the data of the sample solution. It transmits the data stream to the computer software through the USB serial port. Finally, a piece of software processes the acquired information, and fits the curve.

2.2 Constant light intensity driving circuit

The light source is Oslon P1616 SFH4737 wideband infrared LED with an operating current of about 0.3 A. To make it emit constant infrared light, the device is driven by a constant light intensity driving circuit, powered by a computer connected via USB. The board consists of reference voltage, phototriode and adjusting potentiometer. The LM385 chip provides the reference voltage of the circuit. The phototriode can control the value of the collector current according to the light intensity and generates a voltage signal to control the working current. The adjusting potentiometer can change the current size of the light source. The driver circuit provides a stable current output for light so that a constant light intensity is generated.

2.3 Multi-wavelength spectroscopic sensor

The light intensity was acquired with a multi-wavelength spectral sensor, which is small and has low power consumption. The sensor uses a new manufacturing technology. The nano-optical interference filter is attached to the CMOS silicon wafer with filter-width is 20 nm. This interference filter technology provides highly accurate and repeatable filtering characteristics, stability over the life cycle and at different temperatures, and a smaller size than the standard component solutions required for multichannel spectroscopy instruments, reducing component and manufacturing costs. It also uses LGA packaging technology to provide a built-in aperture to control the light entering the sensor array, achieving the light intensity collection of 6 infrared channels, spanning wavelengths from 610 nm to 860 nm. It can be used in food-safety inspections with high accuracy and stability.

2.4 Design of upper computer software

The software was designed based on the Microsoft Visual Studio platform and developed by using C# language. The software is shown in Fig.2. It obtains information from the data received by the multi-wavelength spectral sensor and converts it into information about the light intensity and absorbance of the water sample, and plots a fitting curve. The measurement device was connected to the computer via a USB cable. The port number and baud rate required for the experiment were selected in the “Serial Port Parameter Setting” column. In the “LED light source selection settings” of the three light sources (incandescent, infrared and ultraviolet), infrared (I.R.) was chosen to meet the experimental requirements. The “LED current settings” can determine the current size of the light source. When the parameters were set, “Read DATA” was clicked to read the light intensity of 0 NTU and save it to “BLANK DATA” so that the absorbance could be calculated and saved in the “ABSORBANCE” list for fitting the curve when measuring other turbidity levels.

Fig.2 Upper computer data acquisition software interface

3 Results and discussion

This section outlines the implementation of the experimental models, a comparison of the device with a commercial spectrophotometer and the error analysis among the experimental models.

3.1 Single-channel mode results

3.1.1 Results of transmission mode

The light intensity information of the water samples in the 0 NTU-1 000 NTU range in the transmission mode was obtained by the experimental equipment. Each standard solution was measured more than five times to reduce errors. As the calculated standard deviations obtained with Eq.(5) were relatively small, the average of the measurements was used as the experimental data for this experiment.

(5)

whereSis the sample standard deviation；siis the measurement value； ¯sis the mean value, andnis the number of measurements.

There is good monotonicity between light intensity and turbidity. The data were fitted by using an exponential fit, and the experimental results and fitting coefficients are shown in Fig.3. Detection models were established based on the equations in Fig.3, and a series of standard turbidity solutions were measured according to each detection model. The measurement results are shown in Table 1.

Table 1 Error analysis of transmission mode

Fig.3 Fitting results of standard turbidity solution

In the transmission mode, the 610 nm wavelength channel had a better fit and contained minor errors compared with the other channels. The detection model errors of the 760 nm, 810 nm and 860 nm channels did not differ considerably.

3.1.2 Results of scattering mode

Data were collected for the scattered light intensity of the water sample from 0 NTU-1 000 NTU. There were almost no significant changes in light intensity for the 610 nm and 680 nm wavelength channels. This could be attributed to the low concentrations of colloidal material and particles in the sample. It could also be attributed to the weak light sensitivity of the sensor for this wavelength channel, which leads to an insignificant turbidity trend for these two channels. Therefore, the data for these two channels were discarded. The other channels are linearly fitted by the least squares, and the fit function and the goodness of fit are shown in Fig.4.

Fig.4 Fitting results of standard turbidity solution in the scattering mode

Detection models were established according to the equations of light intensity and turbidity in Fig.4. The turbidity solutions shown in Table 2 were measured by using the detection models, and the measurement errors were calculated. As shown in Table 2, the detection model errors in the scattering mode were all relatively small.

Table 2 Error analysis of detection model in scattering mode

3.1.3 Transmission-scattering ratio results

The transmission-scattering measurement method simultaneously measures the transmitted light intensity and scattered light intensity of the water sample. Measuring the turbidity value of a water sample by utilizing the ratio of the light intensity can eliminate the influence of the aging of the light source on the measurement accuracy. This measurement only has an approximately linear relationship in a low turbidity range, which places some limitations on the measurement range of high turbidity. In this study, the range of the turbidity measurements was 0 NTU-1 000 NTU； therefore, the fitting equation did not use a linear relationship, and the fitting results were shown in Fig.5. Through the fitting equation in Fig.5, the detection model of the transmission-scattering ratio was established. The turbidity of the solution was calculated and compared with the actual results for verification. The results are shown in Table 3.

The detection model error in the transmission-scattering ratio mode is more significant owing to the light source selection and the large measurement range. By combining the data of the above tables, it can be seen that the measurement results of the scattering model are smaller and more stable. The average error is consistently within 0.3 NTU-1.2 NTU.

Fig.5 Results of transmission-scattering ratio fitting of a standard turbidity solution

Table 3 Experimental errors of transmission-scattering ratios

3.1.4 Comparison of proposed device and transmission spectrophotometry

The light intensity characteristics were further processed, and the absorbance of the water sample was calculated by

(6)

whereAis the absorbance of the channel；I0is the light intensity of the blank solution, andItis the light intensity.

The processed data were modeled for detection. The fitting equations and fitting coefficients are shown in Table 4. To verify the accuracy of the method, the results obtained were compared with those of spectrophotometer. The wavelength of the comparison experiment was set as 680 nm, and the absorbances of different turbidity standard solutions were measured. The least-squares linear fit was performed to obtain the fitted equationy=-40.306 93+1 087.563 21x；R2=0.995 43.

Table 4 Fitting expressions for 610 nm, 680 nm, 730 nm, 760 nm, 810 nm, and 860 nm channels

The processed light intensity data of the sample solution were input into the detection model to obtain the predicted values, and error analysis was performed with the actual results, as shown inFig.6. It is clear that the measurement results of the 860 nm channel have a larger error, whereas those of the other channels have a smaller error with the commercially visible spectrophotometer (721 G), and the actual values are closer under the 610 nm channel.

Fig.6 Predicted results for seven test channels

3.2 Multi-channel model results

3.2.1 Results of optical intensity under multiple channels

A multivariate linear fit on the acquired light intensity data and the fitted equations andR2for transmission and scattering are shown in Table 5.

Table 5 Multivariate fitting expressions for sample turbidity and optical intensity

Detection models were established based on the fitted equations for transmission and scattering in Table 5. The turbidities of a series of standard turbidity solutions were measured by using each detection model and the corresponding measurement device. The actual turbidity values of the sample solutions were compared with the predicted values of the relational curves as shown in Fig.7. The results revealed that the values obtained with the multivariate detection models in the scattering model were closer to the actual values.

Fig.7 Comparison of predicted values and actual values

3.2.2 Absorbance results under multiple channels

The transmitted light intensity characteristics were further processed for multivariate linear fitting. The fitted equation wasy=-18.494+126.873x1+5 848.062x2+4 077.835x3-12 458.453x4+11 470.541x5-4 161.061x6.

The fitted curve correlation coefficientR2was 0.994 23, which was close to 1, indicating that the fitting equation fitted the data well. The results measured by the designed device were input into the detection model. The actual turbidity values were compared with the predicted turbidity values, as shown in Fig.8. It was found that the turbidity measurements in the multichannel mode were closer to the actual values, indicating that the multichannel method was the best method for measuring the turbidity of water quality solutions.

Fig.8 Comparison of results between spectrophotometric method and proposed method

4 Conclusions

In this study, sample solutions were measured by using near-infrared light and a multispectral sensor measurement system to obtain the light intensity and absorbance corresponding to the turbidity of the solutions. Furthermore, the turbidities of the solutions were measured by unitary and multivariate detection models. The measurement error of the scattering mode detection model was smaller than that of the transmission mode detection model and the transmission-scattering ratio detection model for the same wavelength band of the unit channel. The measurement error of the single-channel detection mode did not differ by much from that of a commercial spectrophotometer, and the measurements obtained by the two devices were similar. The multichannel detection model measurement error is generally smaller than that of the commercial visible spectrophotometer detection model. In this study, the multivariate fitting method used by the near-infrared multi-wavelength sensor can be applied to turbidity solution measurement and can better detect water quality compared to other sensors. It can replace the optical detection system, which is costly and complicated to use. The method can also be used in other near-infrared measurement fields and can be employed in a wide range of applications in the future.