Dual-wavelength optical sensor for measuring the surface area concentration and the volume concentration of aerosols

Sensors and Actuators B 236 (2016) 334–342

Contents lists available at ScienceDirect

Sensors and Actuators B: Chemical journal homepage: www.elsevier.com/locate/snb

Dual-wavelength optical sensor for measuring the surface area concentration and the volume concentration of aerosols Tian Deng, Shu Wang, Ming Zhu ∗ School of Electronic Information and Communications, Huazhong University of Science and Technology, Wuhan 430074, Hubei, China

a r t i c l e

i n f o

Article history: Received 11 January 2016 Received in revised form 2 June 2016 Accepted 3 June 2016 Available online 4 June 2016 Keywords: Dual-wavelength aerosol sensor Surface area concentration Volume concentration Fire smoke sensing

a b s t r a c t The surface area concentration and the volume concentration of aerosols can be used to detect fire smoke. However, existing optical sensors with a single wavelength source cannot measure these two concentrations accurately. In this paper, we design a sensor to measure these two concentrations with dual wavelength sources, based on the phenomenon that the surface area concentration and the volume concentration of aerosols are proportional to the scattering intensity of the different wavelength incident lights. The prototype sensor is tested by monodisperse aerosols with particle sizes ranging from 200 nm to 2000 nm. The standard measurement deviations of the surface area concentration, volume concentration and Sauter mean diameter are 11.6%, 21.9% and 9.4%, respectively, which are better than the results of sensors with a single wavelength source. Moreover, the Sauter mean diameter of fire smokes measured by our sensor are consistent with calibration instruments, which is helpful to resist false alarm in fire detection. With simple mechanical structure and high accuracy, this sensor is promising in fire detection and atmosphere environment monitoring. © 2016 Elsevier B.V. All rights reserved.

1. Introduction Optical sensors are widely used in fire smoke detection and industries due to their simplicity in production and low cost. Existing fire sensors trigger an alarm when the smoke density exceeds a threshold value. However, the density of non-fire aerosols may also exceed this threshold and trigger a false fire alarm. Studies show that the particle size of smoke aerosol is usually smaller than non-fire aerosols [1–5], thus, it is possible to distinguish fire and non-fire smoke to verify fire alarm by measuring particle size of the aerosols. Most existing instruments measure the particle size by dividing particle size range into different channels with electrical mobility, aerodynamics or other methods [6–10]. However, these channel-dividing technologies require precise instruments with complicated structure and high-cost, which cannot be embedded in a simple and low-cost sensor. Therefore, to detect fire smoke with sensors, it is more practical to obtain the statistical particle size of the aerosols (Sauter mean diameter) by measuring the surface area concentration and the volume concentration [11]. According to Mie scattering theory [12], the intensity of light scattered by a particle with different size is related to the wave-

∗ Corresponding author. E-mail address: [email protected] (M. Zhu). http://dx.doi.org/10.1016/j.snb.2016.06.031 0925-4005/© 2016 Elsevier B.V. All rights reserved.

length of incident light, the observing angle and the refractive index. Cole et al. [13] found that the ratio between infrared and blue scattering signals can be used to determine whether the particle sizes of the aerosols are larger than 1 ␮m. However, they did not explain the mechanism by which the infrared and blue scattering signals were influenced by the particle size. Greenberg et al. [11,14,15] established a paraxial system with a single-wavelength incident laser source and dual observing angles named MPASS (Multi-Parameter Aerosol Scattering Sensor) for fire smoke detection in spacecraft. MPASS measured the surface area concentration and the volume concentration of the aerosols with different observing angles. However, according to the general relationship between scattering intensity and particle size [6], the surface area concentration and the volume concentration of the aerosols should be measured in the regions with different ratios of particle size and wavelength of incident light; hence, the wavelengths of the incident light sources must be different to measure these two concentrations. Because MPASS adopts only a single-wavelength laser source, it cannot measure the surface area concentration and the volume concentration accurately. In this paper, we design and produce an optical sensor based on dual-wavelength light sources to measure the surface area concentration and the volume concentration of the aerosols. The longer wavelength light source is used to measure the volume concentration, while the shorter wavelength source is used for the surface

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

335

Fig. 1. The general relationship of light scattered intensity by unit volume of sphere particle qv and particle size x with wavelength of incident light .

area concentration. Next, the Sauter mean diameter of the aerosols is calculated by the ratio of the volume to the surface area [16]. This sensor is a non-paraxial optical system with two LEDs and one PD (photodiode). Compared with the widely used optical fire smoke sensor [17], only one LED is added in our sensor, which is much simpler in mechanical structure and lower cost than MPASS. Thus, it is very suitable for fire detection in large-scale production. The prototype sensor shows good performance in the tests with monodisperse aerosols ranging from 200 nm to 2000 nm. Moreover, we tested the sensor with real smokes generated by smoldering fires and open fires, the sensor was found to measure the smoke particle sizes accurately. Because our sensor obtains the Sauter mean diameter with high accuracy, it can be used to verify a fire alarm in fire detection. The following paper is organized into four sections. Section 2 introduces the method using sources of two different wavelengths to measure the surface area concentration and the volume concentration of aerosols. Section 3 describes the scattering model of aerosol in the sensor and gives the strategy to optimize the observing angles. The prototype sensor is designed based on the angle optimization principle. Section 4 shows the test results of the prototype sensor for monodisperse aerosols and real smoke aerosols. Section 5 concludes the whole paper. 2. The surface area concentration and the volume concentration measurement by dual wavelengths 2.1. General relationship of scattering light versus particle size

q(x, m, , )  3 x 6

observing angle . As a continuous curve, we have AI ·bI -II 3 = AII ·bI -II 0 and AII ·bII -III 0 = AIII ·bII -III −1 , where bI -II is the boundary between region I and II, bII -III is the boundary between region II and III, it can be inferred that bI -II = (AII /AI )1/3 and bII -III = AIII /AII . Because an aerosol is composed of particles in different sizes, it can be characterized by particle size distribution f(x) in a particle size range. The intensity of light scattered by the aerosol is given by,



P = CN

(1)

Based on Ref. [6], the general relationship of qv relating to x and  is shown in Fig. 1. The relationship can be described in three regions: I) When x < , qv ≈ AI ·x3 , thus q(x,m,,) ≈TI ·x6 ; II) When x ≈ , qv ≈ AII ·x0 , thus q(x,m,,) ≈ TII ·x3 , q(x,m,,) is proportional to the particle volume; III) When x > , qv ≈ AIII ·x−1 , thus q(x,m,,) ≈ TIII ·x2 , q(x,m,,) is proportional to the particle surface area. where AI , AII and AIII are the conversion factors of x and qv in each region, and TI = ␲·AI /6, TII = ␲·AII /6 and TIII = ␲·AIII /6 are the conversion factors of x and q, which are related to refractive index m and

f (x)q(x, m, , )dx

(2)

where CN is the number concentration of the aerosol. When x varies in region II, q(x,m,,) ≈ T·x3 , the scattering intensity P is proportional to the volume concentration of the aerosol CV , we rewrite P as PV . PV =

6 TII · CN 



f (x)(

 3 x )dx = TV · CV 6

(3)

where TV is the conversion factor of the volume concentration of the aerosol, TV = 6·TII /. TV is the scattering intensity by unit volume concentration of the aerosol. Similarly, when x is located in region III, P is proportional to the surface area concentration of the aerosol CS , and we rewrite P as PS . PS =

Illustrated by Baron et al. [6], the scattering intensity versus particle size can be approximated to a simple function of particle size in the statistical measurement of aerosols. The scattering intensity q(x,m,,) is defined as the intensity of monochromatic light scattered by a single particle into a receiving aperture, where x is the particle size, m is the refractive index,  is the wavelength of incident light, and  is the observing angle from emitter to receiver. The intensity of light scattered by unit volume of sphere particle qv can be expressed by: qv =

Fig. 2. Dual wavelength measurement technology, where the longer 1 (solid line) is used to measure the volume concentration and the shorter 2 (dash line) is used to measure the surface area concentration.

1 TIII · CN 



f (x)(x2 )dx = TS · CS

(4)

where TS is the conversion factor of the surface area concentration, TS = TIII /. TS is the scattering intensity by unit surface area concentration of the aerosol. 2.2. Dual-wavelength selection principle According to Section 2.1, the volume concentration CV and the surface area concentration CS are proportional to the scattering intensity in regions II and III, respectively. The boundaries of the regions are determined by the ratio of particle size x and wavelength of incident light . To measure CV and CS , the wavelengths of incident lights should be selected to make the ratio x/ in region II or III. To measure CV , the wavelength should be selected as the middle value of particle size range, such as 1 (solid line) shown in Fig. 2, setting x/ into region II. To measure CS , the wavelength should be smaller than the minimum value of x, e.g., 2 (dashed line) in Fig. 2, setting x/ into region III. Consequently, the wavelength to measure CV must be different from the wavelength to measure CS for the same aerosol. In fire smoke detection, the range of particle sizes is mainly from 200 nm to 2000 nm. This range includes most of fire aerosols and some of non-fire aerosols that float in the air. Thus, an infrared incident light with wavelength of 1100 nm can be used to measure the volume concentration, and an ultraviolet incident light with

336

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

Fig. 4. Non-paraxial optical scattering system.

Fig. 3. Boundary of region II and III (the shaded area) varies by adjusting the observing angle. (a) Expand region II at  1 to measure the volume concentration. (b) Expand region III at  2 to measure the surface area concentration. (c) Measurement mechanism of MPASS with single incident light by adjusting the observing angle.

wavelength slightly less than 200 nm is suitable to measure the surface area concentration. 2.3. Measurement deviation dependence on the observing angle Although the scattering intensity q(x,m,,) can be approximated to a simple function curve of x/ described in Fig. 1, there are still irregular deviations between the approximation curve and the practical measurement. According to Mie theory [12], the deviations result from the refractive index m, the wavelength of incident light  and the observing angle . In designing an optical sensor, the refractive index m is considered as known, and the wavelength of the incident light can be selected by the principle discussed in Section 2.2. When the refractive index m and the wavelength of incident light  are determined, the deviation of measurement mainly comes from the observing angle . According to the approximation relationship discussed in Section 2.1, the change of observing angle would change the values of AI , AII and AIII . The boundary value bII -III = AIII /AII will change correspondingly, so that the change of boundary value yields a band with the observing angle from 0◦ to 180◦ , as shown as the shaded area in Fig. 3. From the discussions in Sections 2.1 and 2.2, we know that the wavelength and observing angle should be selected to measure the volume concentration when the x/ of the particles is in region II, and they should be selected to measure the surface concentration when the x/ of particle is in region III. For a range of particle sizes to be measured, we select 1 and adjust the observing angle to  1 to set the whole range of particle size into region II for the measurement of the volume concentration, as shown in Fig. 3(a). Similarly, we select 2 and adjust observing angle to  2 to set the whole range of particle size into region III for the measurement of the surface area concentration, as shown in Fig. 3(b). If the

observing angle is not properly adjusted, then some particle sizes are located in region III when measuring the volume concentration, i.e., the measurement results of the volume concentration will be contaminated with the surface area concentration, and measurement deviations will occur. A similar case will occur when some particle sizes are located in region II when measuring the surface area concentration. As a sensor using a single wavelength of 650 nm of incident light, MPASS measures the aerosol ranging from 100 nm to 1000 nm with two observing angles [11]. To measure the surface area concentration and the volume concentration, the particle size range of MPASS must contain the band of changing bII -III , as shown in Fig. 3(c). The optimized observing angles of MPASS are searched from 0◦ to 180◦ to maximize the part of particle size locating in region II to measure the volume concentration and in region III to measure surface concentration. At the observing angle  1 , the majority of particles are proportional to the volume concentration, while the particles are more inclined to be proportional to the surface area concentration at the observing angle  2 . However, regardless of the angle selected as the observing angle, the measurement deviations cannot be eliminated when a single wavelength is used. Using two wavelength light sources, the wavelength of incident light  could be selected to match the range of particle sizes and obtain better measurement results for the volume concentration CV and surface area concentration CS , respectively. To eliminate the deviations caused by observing angles, we select the proper observing angle on each wavelength. Thus, it is necessary to optimize the observing angle to improve the measurement accuracy in the sensor design.

3. Design of a prototype sensor 3.1. Non-paraxial optical measurement system To obtain the parameters of the sensor to measure the surface area concentration and the volume concentration of an aerosol, we simulate the scattering intensity in the optical measurement system based on Mie theory. The sensor designed in this paper is a non-paraxial optical system composed by LEDs and PD for low cost, high stability and ease of assembly. In the non-paraxial system, the scattering zone ˝ is a scattering space rather than a focus point as in a paraxial optical system such as MPASS. When a particle is at a certain position in zone ˝, the parameters to describe the scattering intensity are the luminous efficiency of emitter wE , the light receiving efficiency of collector wC and the scattering angle , as shown in Fig. 4. When the emitting angle and receiving angle change, wE is a function of emitting angle  E (˝) and wC is related to the receiving angle  C (˝), which are determined by the properties of the selected LED and PD. We assume that a group of particles of size x is distributed uniformly in the scattering space. In this case,

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

337

3.2. Optimization of the observing angles for the selected wavelengths Based on the optical scattering model in Section 3.1, we obtain the observing angles at the selected wavelengths of the prototype sensor. From Eqs. (3), (4) and (7), the conversion factors of the volume concentration TV and of the surface area concentration TS are given by:



CN ˝ f (x)qw (x, m, , )dx P˝  = CV CN f (x) 6 x3 dx

(8)

CN ˝ f (x)qw (x, m, , )dx P  TS () = ˝ = CS CN f (x)x2 dx

(9)

TV () =



Fig. 5. Light scattered by a single particle into a round receiving aperture.

the average equivalent scattering power of one particle is described as follows: qw (x, m, , ) =

1 CN ˝

· wE (E (˝)) · wC (C (˝)) · qp (x, m, , )d˝

˝ CN

According to Section 2.1, TV and TS can be considered as constant. However, there are deviations between the approximation curve and practical measurement. Taking N different channels of the particle size range of estimated aerosol into consideration, N pairs of TVi and TSi can be calculated by Eqs. (8) and (9), respectively, i.e., TVi () and TSi (), i = 1,. . .,N. Here, we define the average values of TVi () and TSi () for the estimated aerosol as: 1 TVi () N N

TAV () =

(5)

(10)

i=0

1 TSi () N N

Eq. (5) is an integration of ˝, where qp (x,m,,) is the power of the light scattered by a particle into a receiving aperture of collector shown in Fig. 5. Wang [18] gives the expression of qp (x,m,,) as follows: qp (x, m, , ) = I0

 arccos

cos

2 42

low −up 2

sin



up

i(x, m, , ) low

− cos  cos low +up 2

low +up 2

sin d

(6)

sin 

 f (x)qw (x, m, , )dx

(7)

Eq. (7) indicates the relationship between the scattering power P˝ and the aerosol with particle size distribution f(x) in a nonparaxial optical system based on Mie theory. The approximation relationship shown in Fig. 1 must be consistent with results from Eq. (7). Thus, we use Eq. (7) to simulate the scattering power of the non-paraxial optical system and verify the general relationship of the particle size and the scattering intensity. In the following sections, we also use Eq. (7) to optimize the observing angles on the selected incident light wavelengths to obtain the design parameters of the prototype sensor.

(11)

i=0

The relative standard deviations of the conversion factors can be described as: RSDTV () =

std(TV i ()) × 100% TAV ()

(12)

RSDTS () =

std(TS i ()) × 100% TAS ()

(13)



where I0 is the luminous intensity of the incident light against zaxis, i(x,m,,) is the scattering function of the spatial distribution of the light scattered by the single particle into the observing angle , which is described by Mie theory.  up and  low are the upper and lower bounds, respectively, of the receiving aperture to determine the values of observing angle  in a solid angle, which are decided by the photo sensitive area of the receiving aperture S and the distance from the particle to receiving aperture r. ϕ is the polarizing angle, and i(x,m,,) is unrelated to ϕ when the incident light is unpolarized. Eqs. (5) and (6) describe the equivalent scattering power of one particle in a non-paraxial optical system. Substituting Eqs. (5) and (6) into Eq. (2), the equivalent scattering power of an aerosol in zone ˝ with particle size distribution f(x) is described as: P˝ = CN ˝

TAS () =

It is convenient to select the wavelengths by the principle discussed in Section 2.2; thus, we focus on optimizing observing angles  TS and  TV from the minimum values of RSDTV and RSDTS , respectively: TV = {| min(RSDTV ())}

(14)

TS = {| min(RSDTS ())}

(15)

To design the prototype sensor with optimized observing angles, we obtain TAV ( TV ) as the conversion factor of the volume concentration and TAS ( TS ) as the conversion factor of the surface area concentration. 3.3. Simulation of multiple light wavelengths with different observing angles To guide the design of prototype sensor, we simulate the scattering optical system at multiple wavelengths with different observing angles. In the simulations, the incident light wavelengths  are selected by the principle discussed in Section 2.2. The measurement deviations (relative standard deviations) of the volume concentration and the surface area concentration are calculated by Eqs. (12) and (13), respectively. Finally,  are optimized by Eqs. (14) and (15). As illustrated in Section 2.2, V should be selected as the middle value of the particle size range to measure the volume concentration, and S should be selected as the minimum value of the particle size range to measure the surface area concentration. The particle size ranging from 200 nm to 2000 nm includes most of fire aerosols and some of non-fire aerosols that float in the air. Here, we consider

338

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

Fig. 7. The prototype sensor, where the observing angles are shown as plane angles in (a) and are shown as space angles in (b). (a) Schematic plot of prototype sensor. (b) Pictures of real prototype sensor. Fig. 6. Measurement relative standard deviations of different wavelengths of incident lights versus observing angles , with particles size ranging from 200 nm to 2000 nm in the step-size of 10 nm. (a) RSDTV , the relative standard deviations of volume concentration of different wavelengths of the incident light. (b) RSDTS , the relative standard deviations of surface area concentration of different wavelengths of the incident light.

1100 nm infrared light and 100 nm ultraviolet light as the most suitable means to sense CV and CS , respectively. Because fire aerosols are small particles, a slightly shorter infrared wavelength of 860 nm is chosen to measure the volume concentration CV . Because blue LEDs are accessible at low prices, the 460 nm blue incident light is chosen to measure the surface area concentration CS . Both of 860 nm and 460 nm LEDs use one PD for the collector. A shorter ultraviolet light of 380 nm can be applied to improve the measurement accuracy, but it is not worth adding one more PD working in the ultraviolet for a practical fire sensor. Consequently, 860 nm infrared light and 460 nm blue light are selected to be the incident lights of a prototype sensor with dual-wavelength technology. The refractive index in the simulation is set as 1.45, which is the value of the refractive index of the aerosols generated by DEHS (di-ethylhexyl sebacate) in the experiments in section 4. For fire aerosols, the refractive index is set 1.6 in Greenberg et al. [13]. The observing angle is restricted from 30◦ to 150◦ because of the limitation of the mechanical design and the measurement errors caused by stray light at the angles lower than 30◦ and higher than 150◦ . Based on Eqs. (12) and (13), we obtain the relative standard deviations RSDTV and RSDTS of different wavelengths of incident lights versus observing angles , with particles size ranging from 200 nm to 2000 nm in the steps of 10 nm. The measurement results of 460 nm blue light, 860 nm infrared light and 650 nm infrared light (which is used by MPASS) are compared in Fig. 6. As shown in Fig. 6, the relative standard deviations of the volume concentration RSDTV decrease when the wavelength increases from 460 nm (solid line) to 860 nm (dash line) in Fig. 6 (a), while the relative standard deviations of the surface area concentration RSDTS

decrease when the wavelength decreases from 860 nm to 460 nm in Fig. 6 (b). Thus, the simulation results validate the dual wavelength selection principle introduced in section 2. The wavelength selected as the middle value of particle size range is suitable to measure the volume concentration of the aerosols, and the wavelength selected as the minimum value of the particle size range is suitable to measure the surface area concentration. The dotted lines in Fig. 6 show the measurement deviations of 650 nm incident light. From Fig. 6, we can see that the measurement deviation of CV of 650 nm light is larger than that of 860 nm light, and the measurement deviation of CS of 650 nm light is larger than that of 460 nm light. From this comparison, we can conclude that the measurement accuracy of our dual wavelength technology is better than that of the single wavelength technology.

3.4. Prototype sensor Based on simulation in Section 3.3, the RSDTV () of 860 nm incident light for the volume concentration CV in Fig. 6(a) has the minimum measurement deviation of 18.1% at the optimized observing angle 45◦ . RSDTS () of 460 nm incident light for the surface area concentration CS has the minimum measurement deviation 13.6% at the optimized observing angle of 112◦ . Based on the simulation results, we can design the prototype sensor. The optical system of the prototype sensor is shown in Fig. 7(a), and the real prototype sensor is shown in Fig. 7(b). The optical maze in Fig. 7(b) is used to attenuate the ambient light from the environment. As a non-paraxial optical system, our sensor has a simple mechanical structure without any optical lens. Compared with existing optical fire smoke sensor, only one LED is added. Therefore, the cost is much lower than the optical system based on a laser light source, such as MPASS.

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

339

Fig. 9. Results of the volume concentration CV measured by the sensor versus the measurement results of SMPS & APS.

Fig. 8. The experimental platform. (a) Schematic diagram of the experimental platform. (b) The experimental platform in the laboratory

4. Experiment and results 4.1. Experimental platform The experimental platform is shown in Fig. 8. A monodisperse aerosol generator (MAG) produces monodisperse DEHS aerosols in a different particle sizes and pumps them into a dilute chamber. The aerosols are stirred well by muffin fans and then enter into the sensors. For convenience, in the experiments, we use a given surface area concentration CS of the aerosols as a reference parameter to control the aerosol concentrations in each experiment. After CS is stable, the outputs of the sensors are recorded. When reference instruments SMPS3936 (Scanning Mobility Particle Sizer Spectrometer 3936, TSI) and APS3321 (Aerodynamic Particle Sizer Spectrometer 3321, TSI) inhale air and aerosol for measuring, the airbag is used to maintain a stable air pressure and a stable aerosol concentration in the chamber. In our experiments, we implemented two groups of experiments. In Section 4.2, the first group of experiments compares the measurement accuracy of our prototype sensor with five types of single wavelength sensors at different observing angles. In Section 4.3, the second group of experiments tests our prototype sensor using smoldering cotton wick and wood smoke aerosols. 4.2. Experimental results for DEHS monodisperse aerosols We compare our prototype sensor (PS) with five types of contrast sensors to measure the surface area concentration and the volume concentration of the sample aerosols. Contrast sensors are represented as CS(I) to CS(V) in Table 1. Sensors with infrared incident lights are compared with the volume concentration measurement results of the prototype sensor, whereas sensors with blue incident lights are compared with the surface area concentration measurement results of the prototype sensor. To evaluate the measure accuracy of different sensors, we consider the measurement results of SMPS & APS as the correct value for reference. The measurement results of these sensors in the experiments are compared with the results of SMPS & APS to evaluate the performances of sensors. The experiments include 19 aerosols with different particle sizes generated by the MAG: 248 nm, 270.1 nm, 297.7 nm, 332 nm, 364.5 nm, 404.5 nm,

465.9 nm, 507.4 nm, 547 nm, 605 nm, 666 nm, 736 nm, 851 nm, 939 nm, 1119 nm, 1405 nm, 1559 nm, 1724 nm, 1906 nm, 2123 nm. In practice, the scattered light is converted into digital signals by a circuit; therefore, conversion factors TV () and TS () are affected by the parameters of the circuit, such as the luminescence efficiency of the incident lights from the LEDs, the receiving efficiency of the PD, the amplification factor and so on. By normalizing these factors via experimental linear calibration, we can obtain the practical TAV () and TAS (). Thus, we obtain the measurements of CV and CS as: CV =

P TAV ()

(16)

CS =

P TAS ()

(17)

By this processing, the signals of the scattering power are translated into the values of the surface area concentration and the volume concentration of the aerosols. Based on Eqs. (16) and (17), CV and CS of the sample aerosols in different particle sizes are obtained and compared with the reference values measured by SMPS & APS. Fig. 9 shows the values of the volume concentration CV measured by different sensors (vertical axis) versus the measurement results of SMPS & APS (horizontal axis). The dashed line is a reference line that indicates the similarity among the measurement results by the sensors and by SMPS & APS. As shown in Fig. 9, the deviations of the measurement results are different with the change of observing angles. The black dot points are the results of our prototype sensor at 45◦ for the volume concentration, which is the most consistent with the measurement results of SMPS & APS compared to the other sensors at the same wavelength. We also compare the surface area concentration CS measured by the sensors with the results of SMPS & APS in Fig. 10. The measurement results of our prototype sensor, represented by the black dot points, have the minimum deviation. The experimental results in Figs. 9 and 10 are consistent with the simulation results in Fig. 6. To analyze the measurement deviations, we define the relative standard deviation of the volume concentration CV as:



RSDCV () =

1 N

 (C

VSi

− CVRi )2 2 CVRi

(18)

where CVSi is the volume concentration measured by the sensor, CVSi = Pi /TAV . CVRi is the reference volume concentration measured by SMPS & APS, and CVRi = Pi /TVi . Pi is the scattering power of the

340

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

Table 1 Prototype sensor and existing sensors in the experiments. Sensor

PS

Angle (◦ ) Wavelength (nm)

45 860

112 460

CS(I)

CS(II)

CS(III)

CS(IV)

CS(V)

65 860

85 860

140 860

135 460

140 460

Fig. 10. Results of the surface area concentration CS measured by the sensor versus the measurement results of SMPS & APS.

Fig. 12. Sauter mean diameter measured by the prototype sensor versus the count median diameter of the aerosols by SMPS & APS.

standard deviation RSDCS 11.6%. As introduced in Ref. [11], RSDCV of MPASS for particles ranging from 100 nm to 1000 nm is approximately 36%, and RSDCS of MPASS is not mentioned at all. Obviously, the results by our prototype sensor are better than MPASS. Moreover, our measurement deviation would be decreased further when the target particle size range is reduced from 200–2000 nm to 100–1000 nm. The Sauter mean diameter of an aerosol [16] is defined as DS = 6·CV /CS . DS is approximately equal to the count median diameter (CMD) when the aerosol is monodisperse. Fig. 12 shows the comparison between DS measured by our prototype sensor and CMD obtained by SMPS & APS; the average relative standard deviation is only 9.4%.

Fig. 11. Measurement deviations of CV and CS in the experiment versus the simulation.

ith experiment. Therefore, the relative standard deviation of the measured volume concentration RSDCV () is equal to the relative standard deviation of the simulation conversion factors RSDTV () in Eq. (12). Similarly, the relative standard deviations of the surface area concentration RSDCS () is defined as:



RSDCS () =

1 N

 (C

SSi

− CSRi )2 2 CSRi

(19)

where CSS is the surface area concentration measured by the sensor and CSR is the reference surface area concentration measured by SMPS & APS. RSDCS () is also equal to RSDTS () described by Eq. (13). RSDCV () and RSDCS () of the sensors are compared with the simulation results of RSDTV () and RSDTS (), respectively, as shown in Fig. 11. As shown in Fig. 11, the experiment results are consistent with the simulations. RSDCV of our prototype sensor is 21.9%, which is the minimum among all of the experimental results. The measurements of CS by our prototype sensor have the minimum relative

4.3. Experimental results for smoke aerosols As described by Eqs. (16) and (17), the conversion functions from scattering power to volume concentration and to surface area concentration are linear. Therefore, our prototype sensor could measure any polydisperse aerosol when the whole particle size range is located in region II or region III. The measurement deviations of polydisperse aerosols will be no larger than the maximum measurement deviation of monodisperse aerosols. As polydisperse aerosols, fire smokes are used to test our prototype sensor. As an example, the tests of smoldering cotton and smoldering wood are shown in Fig. 13, which demonstrates the full time outputs of CV and CS from prototype sensor and the reference values given by SMPS. The prototype sensor can output continuous results in real time, while SMPS needs 2–3 min to obtain the reference results. We employ the measurement results of prototype sensor and SMPS at stable stage to evaluate the accuracy of our sensor. In fire smoke experiments, the prototype sensor is tested by four kinds of smokes from different fire sources: smoldering cotton, smoldering wood, open fire of n-heptane and open fire of polyurethane. The results of CV and CS in each fire smoke test are listed in Tables 2 and Table 3. As shown in Table 2, the prototype sensor measured the volume concentrations of smoldering wood and smoldering cotton correctly, but the responses to open fire of

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

341

Table 4 Test results of Sauter mean diameter by prototype sensor and SMPS. Test fire

Prototype sensor (nm)

SMPS (nm)

smoldering cotton smoldering wood open fire of n-heptane open fire of polyurethane

225.30 368.54 555.74 339.09

241.89 344.94 543.92 391.39

Fig. 14. Sauter mean diameter measured by the prototype sensor versus SMPS in fire smoke tests. Fig. 13. Full time outputs of prototype sensor versus reference results given by SMPS. (a) Full time outputs of the surface area concentration CS . (b) Full time outputs of the volume concentration CV Table 2 Test results of CV by prototype sensor and SMPS. Test fire

Prototype sensor (E + 12 nm3 /m3 )

SMPS (E + 12 nm3 /m3 )

smoldering cotton smoldering wood open fire of n-heptane open fire of polyurethane

5.45 4.58 14.17 1.96

5.89 4.26 41.90 5.91

Table 3 Test results of CS by prototype sensor and SMPS. Test fire

Prototype sensor (E + 10 nm2 /m3)

SMPS (E + 10 nm2 /m3 )

smoldering cotton smoldering wood open fire of n-heptane open fire of polyurethane

14.51 7.45 15.30 3.47

14.63 7.41 46.22 9.06

n-heptane and open fire of polyurethane reduce to 33.8% and 33.2% of the reference values given by SMPS respectively. The similar situation occurs in the measurements of surface area concentrations, as low as 33.1% and 38.3% respectively, as shown in Table 3. The measurement errors of open fire smokes are caused by using the conversion factors of smoldering fire smokes, from scattering power to volume and to surface area concentrations. The difference of reflective indices between smoldering fire smokes and open fire smokes is the dominant reason that causes the deviations of conversion factors. The two smoldering fires generate gray smokes, while the two open fires generate soot (black smokes). The gray smokes scatter most of incident lights while the soot absorbs part of incident lights, thus the scattering signals of soot are lower than those of gray smokes. Since the conversion factors of smoldering fire smoke are used in fire smoke tests, the CV and CS of soot measured by prototype sensor are lower than the reference values.

While in fire detection, Sauter mean diameter is helpful to resist false alarm. From DS = 6·CV /CS , we can see that the Sauter mean diameters of soot can be approximately constant as the responses of CV and CS just decrease at similar rate (as shown in Tables 2 and 3). Thus, our prototype sensor can still obtain Sauter mean diameters accurately. From Table 4, we can see that the Sauter mean diameters of all four kinds of fire smokes measured by our prototype sensor are similar with the reference results given by SMPS. To further validate this view, we make more experiments in different fuel quantity and aging time, and obtain the fire smokes with different particle size distributions for comparison. As shown in Fig. 14, the relative standard deviation of Sauter mean diameters of our prototype sensor to SMPS is only 6.04%. Hence, the prototype sensor can measure the Sauter mean diameters of fire smokes accurately (Table 4). 5. Conclusion We designed an optical sensor based on the selection of two different wavelength light sources to measure the surface area concentration and the volume concentration of aerosols. The measurement deviations are reduced by optimizing the observing angles at the two wavelengths. Thus, we can obtain the Sauter mean diameter with high accuracy. In our prototype sensor, a simple non-paraxial optical system was built with optimized observing angles at the selected wavelengths: an 860 nm infrared LED at 45◦ is used to measure the volume concentration, and a 460 nm blue LED at 112◦ is used to measure the surface area concentration. Our prototype sensor was tested by DEHS monodisperse aerosols. The relative standard deviations of measurements for the surface area concentration, the volume concentration and the Sauter mean diameter are 11.6%, 21.9% and 9.4%, respectively. Compared with the sensor with single wavelength, our prototype sensor significantly improves the measurement accuracy of the surface area concentration and the volume concentration. We also tested the sensor with fire smokes from smoldering fires and open fires. The measurement results show that the responses of CV and CS to soot are lower than expectations, mostly due to the absorption of incident light. As the responses of CV and CS just decrease at sim-

342

T. Deng et al. / Sensors and Actuators B 236 (2016) 334–342

ilar rate, the Sauter mean diameters of soot can be approximately constant. Therefore, the prototype sensor can still measure Sauter mean diameter accurately, the relative standard deviation in all the fire smoke tests is 6.04%. The sensor proposed in this paper has a simple mechanical structure and is low cost; thus, it can be utilized to measure the surface area concentration and the volume concentration of aerosols ranging from sub-micrometer to micrometer scale in a wide range of applications involving fire detection and atmosphere environment monitoring.

[15] G.W. Hunter, P.S. Green Berg, J.C. Xu, B. Ward, D. Makel, P. Dutta, et al., Miniaturized sensor systems for early fire detection in spacecraft, SAE Tech. Pap. (2009), http://dx.doi.org/10.4271/2009-01-2469. [16] S.K. Friedlander, Dust, Smoke and Haze, 2nd ed., Oxford University Press, 2000, http://agris.fao.org/agris-search/search.do?recordID=US201300042742 (accessed 15.08.15). [17] S. Ev-p, Optical smoke detector, US Pat. . . .. (1995) 7107800. https://www. google.com/patents/US5451931 (accessed 18.11.15). [18] J. Wang, G. Xu, N. Wang, Mathematical model of calculating the flux of laser scattering by single particle in arbitraty directions, Appl. LASER (1995) 79–80.

Biographies References [1] TSI Incorporated, Particle Size, TSI Technical Notes (2013), http://www.tsi. com/uploadedFiles/ Site Root/Products/Literature/Technical Notes/Particle Size.pdf. [2] R.W. Shaw, Air pollution by particles, Sci. Am. 257 (1987) 96–103, http://dx. doi.org/10.1038/scientificamerican0887-96. [3] Schultze Krüll, Willms Tobera, Characterization of dust aerosols in false alarm scenarios, in: 15th Int. Conf. Autom. Fire Detect. AUBE’14, Duisburg, Germany, 2014. [4] Ruff Urban, Greenberg Meyer, Cleary Fisher, et al., The smoke aerosol modelling experiments (SAME) conducted on the international space station, in: 15th Int. Conf. Autom. Fire Detect. AUBE’14, Duisburg, Germany, 2014. [5] J. Dinaburg, D. Gottuk, Smoke alarm nuisance source characterization: review and recommendations, Fire Technol. (2015), http://dx.doi.org/10.1007/ s10694-015-0502-1, http://link.springer.com/article/10.1007/s10694-0150502-1 (accessed 11.01.16). [6] P. Baron, K. Willeke, Aerosol Measurement: Principles, Techniques, and Applications, 2nd ed., Wiley-Interscience, 2005, http://www.citeulike.org/ group/12297/article/2761053 (accessed 15.08.15). [7] X. Cai, Particle Size Measurement Techniques and Applications, Chemical Industry Press, 2010. [8] M. Loepfe, P. Ryser, C. Tompkin, D. Wieser, Optical properties of fire and non-fire aerosols, Fire Saf. J. 29 (1997) 185–194, http://dx.doi.org/10.1016/ S0379-7112(96)00048-3. [9] D.W. Weinert, T.G. Cleary, G.W. Mulholland, P.F. Beever, Light scattering characteristics and size distribution of smoke and nuisance aerosols, Fire Saf. Sci. (2003) 209–220, http://dx.doi.org/10.3801/IAFSS.FSS.7-209. [10] C.F. Bohren, D.R. Huffman, Absorption and Scattering of Light by Small Particles, Wiley-VCH Verlag GmbH & Co. KGaA, 1983, http://dx.doi.org/10. 1002/9783527618156. [11] P.S. Greenberg, D.G. Fischer, N. Glenn, Advanced Particulate Sensors for Spacecraft Early Warning Fire Detection, 40th Int. Conf. Environ. (2010) 1–10, http://arc.aiaa.org/doi/pdf/10.2514/6.2010-6243 (accessed 15.08.15). [12] C.H. Hulst, H.C. Van de Hulst, Light Scattering by Small Particles, Dover Publications, 1957, https://books.google.com/books?hl=zhCN&lr=&id=PlHfPMVAFRcC&oi=fnd&pg=PA1&dq =Light+scattering+by+small+particles+%5BM%5D. +Courier+Corporation,&ots=Gf1hPsqqHa&sig=aV2chKsFNCDisVZDyZvHVhY5s8 (accessed 14.08.15). [13] M. Cole, Aerosol characterisation for reliable ASD operation, in: 14th Int. Conf. Autom. Fire Detect. AUBE’09, Duisburg, Germany, 2009. [14] P. Greenberg, D. Fischer, Detection of spacecraft fire signatures and post-fire aerosols—part I: ground-based results, Fire Saf. J. (2015), http://www. sciencedirect.com/science/article/pii/S0379711215000569 (accessed 15.08.15).

Tian Deng is Ph.D. of School of Electronic Information and Communications at Huazhong University of Science and Technology, China. His current research area is the aerosol measurement and fire detection based on optical method.

Shu Wang is Professor and the dean of School of Electronic Information and Communications at Huazhong University of Science and Technology. He has over three decades of academic experience on the faculty at HUST. His research interests and expertise are aerosol measurement, fire smoke detection and acoustic gas detection. He has published over 100 journal papers.

Ming Zhu is Associate Professor of School of Electronic Information and Communications at Huazhong University of Science and Technology. He has the Ph.D. in Electronics and Information Engineering from HUST, awarded in 2008. He was a visiting scholar at Northwestern University of United States from 2014 to 2015. His current research areas range from aerosol measurement to acoustic gas detection.