sensors Article

Fine Particle Sensor Based on Multi-Angle Light Scattering and Data Fusion Wenjia Shao, Hongjian Zhang and Hongliang Zhou * State Key Laboratory of Industrial Control Technology, College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China; [email protected] (W.S.); [email protected] (H.Z.) * Correspondence: [email protected]; Tel.: +86-571-8795-2253 Academic Editor: Paul Solomon Received: 20 February 2017; Accepted: 25 April 2017; Published: 4 May 2017

Abstract: Meteorological parameters such as relative humidity have a significant impact on the precision of PM2.5 measurement instruments based on light scattering. Instead of adding meteorological sensors or dehumidification devices used widely in commercial PM2.5 measurement instruments, a novel particle sensor based on multi-angle light scattering and data fusion is proposed to eliminate the effect of meteorological factors. Three photodiodes are employed to collect the scattered light flux at three distinct angles. Weather index is defined as the ratio of scattered light fluxes collected at the 40◦ and 55◦ angles, which can be used to distinguish the mass median diameter variation caused by different meteorological parameters. Simulations based on Lorenz-Mie theory and field experiments establish the feasibility of this scheme. Experimental results indicate that mass median diameter has less effect on the photodiode at the 55◦ angle in comparison with photodiodes at the 40◦ angle and 140◦ angle. After correction using the weather index, the photodiode at the 40◦ angle yielded the best results followed by photodiodes at the 55◦ angle and the 140◦ angle. Keywords: PM2.5; mass concentration; light flux; weather index; relative humidity

1. Introduction Fine particles with aerodynamic diameters of 2.5 µm or less (PM2.5) have received considerable attention because of their adverse impacts on human health, especially their ability to penetrate deep into our lungs and even blood streams. In 2013, the International Agency for Research on Cancer (IARC) classified Particulate Matter (PM) from outdoor air pollution as carcinogenic to humans [1]. PM2.5 has been listed in the revised Chinese National Ambient Air Quality Standards (NAAQS). The limit of daily average PM2.5 mass concentration is 35 µg/m3 while it is 10 µg/m3 inthe World Health Organization (WHO) air quality guideline. Monitoring networks have been established to record the levels of fine particulate matter (PM2.5) in China. There are several technologies that can be used to determine the PM2.5 mass concentrations, among which the Tapered Element Oscillating Microbalance (TEOM) and beta attenuation method are most popular. However, instruments based on these methods are quite expensive (>$20,000) and require regular maintenance and calibration. Many new particle sensors have emerged in recent years, which are based on particle charging [2], silicon carbide-field effect transistor(SiC-FET) technology [3], conductivity [4,5] and other techniques. Instruments based on light scattering, such as the DustTrak II Aerosol Monitor 8530, SidePak Aerosol Monitor AM510, and DC1700, can provide continuous PM2.5 mass concentrations at relatively low cost [6]. However, light scattering is affected by various factors, for example, the particle’s size, refractive index, density and shape. These instruments need proper calibration to achieve accurate measurement. Furthermore, these instruments do not perform well when the compositions of PM2.5 vary greatly. Therefore, most of these instruments are used indoors or in situations where particle composition remains constant. Numerous studies have been carried out to analyze the calibration Sensors 2017, 17, 1033; doi:10.3390/s17051033

www.mdpi.com/journal/sensors

Sensors 2017, 17, 1033

2 of 15

of different kinds of PM2.5 [7–9]. It is a great challenge to accurately estimate PM2.5 mass concentration for diverse kinds of PM2.5. Research has revealed that high Relative Humidity (RH) will have a strong effect on light scattering [10–15]. Ammonium sulfate and ammonium nitrate are the major semi-volatile components of PM2.5.The deliquescence point of ammonium sulfate and ammonium nitrate is 62% and 80%, respectively. Therefore, when RH is higher than 62%, particle-bound water increases. The particle’s mass will increase when RH exceeds 65% and will increase sharply when RH exceeds 80%, which will lead to the overestimation of PM2.5 [13]. Some researchers introduced the hygroscopic growth factor, which is the ratio of a particle’s diameter under high RH conditions to that under dry conditions, to assess the effect of RH on measurement results. The hygroscopic growth factor of particles in the range of 200 nm to 1 µm is about 1.5 when RH is 90% and reaches nearly 2.0 when RH is 95%. A little variation in RH can result in considerable changes in the hygroscopic growth factor, especially when RH exceeds 90% [16]. High RH will enlarge particle size and reduce the particle refractive index, resulting in a more smooth or spherical shape through the absorption of water by particles. Countless efforts have been made to eliminate the effects of RH. Sioutas et al. used a diffusion dryer to exclude aqueous components of PM2.5 in real-time PM2.5 mass concentration measurements [11]. However, adsorbed water needs to be driven off the desiccant in the diffusion dryer every 7 h. Others used heated inlets to remove aqueous components, which will cause excessive volatilization of labile PM2.5 constituents. Thus, additional instruments should be added to determine the mass of volatile component in PM2.5. For example, a Filter Dynamic Measurement System (FDMS) should be incorporated into a conventional TEOM monitor, which evaporates condensed water, to ensure the reliability of PM2.5 measurement results [17]. In order to diminish the influence of meteorological parameters such as RH on the PM2.5 sensors based on light scattering, a new particle sensor based on multi-angle light scattering and data fusion is proposed in this paper. The sensor presented here is a new version of the prototype developed by our group and shows better accuracy and stability with respect to thedevice reported earlier in reference [18]. The rest of the paper is arranged as follows: the design of the sensor is described in Section 2. Theory of data fusion and the signal processing method is described in Section 3. In Section 4, field experiments are carried out to verify the validity as well as evaluate the performance of the particle sensor. The final section, Section 5, summarizes the characteristics of the particle sensor. 2. Design of the Sensor and Particle Measurement System In order to overcome the drawbacks of the existing PM2.5 sensors based on light scattering, a new particle sensor with more than one optical detector was developed. Simulations were performed to obtain the appropriate parameters of the sensor. The distribution of the scattered light of a particle varies with the particle’s diameter, refractive index and shape. The Lorenz-Mie solution of Maxwell’s equations describes the scattering of an electromagnetic plane wave by a homogeneous sphere [19]. To optimize the mounting angle of the optical detector, scattered light flux collected by the optical detector is calculated for particles with different refractive indices and diameters [20]. The following simulation conclusions can be drawn: (1) forward scattered light flux is insensitive to the imaginary part of particle’s refractive index, which is consistent with conclusions in [21], (2) sideward scattered light flux is insensitive to the real part of a particle’s refractive index, and (3) backward scattered light flux is sensitive to a particle’s shape. A 3D view of the particle sensor is shown in Figure 1. The main body of the sensor, with the size of 120 mm × 80 mm × 80 mm, is produced by a 3D printer using photosensitive resin. Both the inside and outside walls of the main body are painted with black matt varnish to minimize the reflection of the scattered light and the effect of stray light. A light trap is constructed on the right side to suppress stray light from the light source. A stainless steel tube and a nozzle are introduced to reduce the settlement of particles in the gas tube and to focus the particle flow into the laser beam. The nozzle is also coated with black matt varnish to reduce stray light. A semiconductor laser is

Sensors 2017, 17, 1033

3 of 15

used to emit a 648 nm light beam in continuous wave mode. Air flow containing particles enters the optical chamber via the stainless steel tube, where it is illuminated by the laser beam. Scattered 2017, 17, 1033 3 of three 15 light in the Sensors scattering angle ranges of 40◦ ± 30◦ , 55◦ ± 30◦ and 140◦ ± 30◦ is captured by convex lenses and then focused onto three photo-detectors. Akey element of the photo-detectors is photodiode via the stainless steel tube, where it is illuminated by the laser beam. Scattered light in the scattering S2386-44K (Si-PD, Hamamatsu Hamamatsu, Japan) [22].convex A horn structure used to prevent angle ranges of 40 ± 30°, Photonics, 55 ± 30° and 140 ± 30° is captured by three lenses and then is focused onto three photo-detectors. Akey element of the photo-detectors is photodiode S2386-44K (Si-PD, the stray light of the laser beam from entering into the particle sensor and to stably maintain the power Photonics, Hamamatsu, Japan) [22]. A horn structure is used to prevent the stray light of the laserHamamatsu beam stable. A sheath gas tube is reserved to protect the optical systems from particle of the2017, laser Sensors 17,beam 1033 from entering into the particle sensor and to stably maintain the power of the laser 3 of 15 pollution. The of the optimized to ensure small particles can pollution. be detected beamdiameter stable. A sheath gastube tube isisreserved to protect the optical systems from particle The and stray diameter of thesteel tube issurfaces optimized toisthe ensure small particles canbeam. be detected andlight strayinlight to light owing tothereflections offtube, of tube can be via stainless where it illuminated by the suppressed. laser Scattered the owing scattering reflections tubeand can140 be suppressed. angle rangesoff ofsurfaces 40 ± 30°,of55the ± 30° ± 30° is captured by three convex lenses and then focused onto three photo-detectors. Akey element of the photo-detectors is photodiode S2386-44K (Si-PD, θ=100° Hamamatsu Photonics, Hamamatsu, Japan) [22]. A horn structure is used to prevent the stray light of the laser beam from entering into the particle sensor and to stably maintain the power of the laser beam stable. A sheath gas tube is reserved to protect the optical systems Convex from particle pollution. The lens θ=40° 140° photodiode 40° photodiode diameter of the tube is optimized to ensure small particles can be detected and stray light owing to reflections off surfaces of the tube can be suppressed. 55° photodiode

θ=100° Semiconductor laser

Horn structure

θ=55°

θ=40°

Convex lens

(a)

140° photodiode

(b)

40° photodiode

Figure 1. Diagram of the particle sensor: (a) 3D view; (b) Top-down view.

55° photodiode Figure 1. Diagram of the particle sensor: (a) 3D view; (b) Top-down view.

The particle measurement system consists of theSemiconductor particlelasersampling unit, particle sensing unit and Horn structure θ=55° signal processing unit, as shown in Figure 2. The particle sampling unit comprises a PM2.5 impactor The particle measurement system consists of the particle sampling unit, particle sensing unit (TSI Inc., Shoreview, MN, USA), a miniature diaphragm gas pump D871-23 (Parker Inc., Cleveland, and signal OH, processing as shown in Figure 2. The particle sampling unit comprises USA) and unit, a gas flow impactor is used to pre- a PM2.5 (a) meter (Senlod Inc., Nanjing, China). The PM2.5(b) condition size range ofMN, the particles the particle sensor. By controlling theD871-23 pump speed impactor (TSI Inc., the Shoreview, USA),entering a miniature diaphragm gas pump (Parker Inc., Figure 1. Diagram of theparticles particle sensor: (a) 3D view; (b) Top-downdiameter view. at 3.0 L/min, the impactor can isolate with aerodynamic equivalent less than 2.5 Cleveland, OH, USA) and a gas flow meter (Senlod Inc., Nanjing, China). The PM2.5 impactor is used μm from coarse particles in the particle sensing unit. A multi-coil transformer and a LDO voltage Thethe particle system consistsentering of the particle particle unit andthe pump to pre-condition sizemeasurement range of the particles thesampling particleunit, sensor. Bysensing controlling regulator TPS 78633 are used to build up high-quality laser power supply. The signal processing unit signal processing unit, as shown in Figure 2. The particle sampling unit comprises a PM2.5 impactor speed at 3.0is L/min, the impactor can isolate particles withtheaerodynamic equivalent diameter less mainly a current-to-voltage converter. Weak current from photodiodes is amplified by LMP (TSI Inc., MN, USA), a miniature diaphragm gas precision. pump D871-23 (Parker Inc., an Shoreview, operational amplifier with ultra-low noise and high The stray light actsCleveland, as a kindand a LDO than 2.5 µm7721, from coarse particles in the particle sensing unit. A multi-coil transformer OH, USA) and a which gas flow meter slowly (Senlodover Inc., time Nanjing, The PM2.5 impactor is used to preof continuum, changes due China). to changes in temperature and pressure voltage regulator TPS 78633 are used to entering build up high-quality laser power supply. The signal condition the size range of the particles the particle sensor. By controlling the pump speed conditions or possible dust contamination in the optical chamber. The light scattered by the particles 3.0 L/min, the impactor can isolateThe particles with aerodynamic equivalent than 2.5 processingatisunit is mainly a current-to-voltage converter. Weak current from less the photodiodes is superimposed on this continuum. stray light signal can be assumed as adiameter continuous base-line from coarse particles inAs the particle sensing unit. Aultra-low multi-coil a LDO voltage The stray over a short time interval. shown in Figure 3, with a 50 Hz notch filtertransformer is employed to remove power amplified μm by LMP 7721, an operational amplifier noise andand high precision. regulatornoise, TPS 78633 used to build laser power supply. The signal processing unit and are a high-pass filterup ishigh-quality used to slowly remove stray signal in the signal light acts assupply a kind of continuum, which changes overlight timevoltage due to changes in temperature isconditioning mainly a current-to-voltage Weak current from the by photodiodes is amplified by data LMP circuit. Then, the converter. scattered light signal sare acquired an eight-path synchronous and pressure conditions oramplifier possible dust contamination inprecision. the optical chamber. The 7721, an operational noise high The stray light acts as alight kind scattered acquisition card with a samplewith rate ultra-low of 12.5 KHz andand processed using LabVIEW. of continuum, which changes on slowly time due to changes in temperature and can pressure by the particles is superimposed thisover continuum. The stray light signal be assumed conditions or possible dust contamination in the optical chamber. The light scattered by the particles as a continuous base-line over a short time interval. As shown in Figure 3, a 50 Hz notch filter is is superimposed on this continuum. The stray light signal can be assumed as a continuous base-line employed to remove power supply noise, and a high-pass filter is used to remove stray light voltage over a short time interval. As shown in Figure 3, a 50 Hz notch filter is employed to remove power signal in the signal conditioning circuit. thetoscattered lightlight signal saresignal acquired an eight-path supply noise, and a high-pass filterThen, is used remove stray voltage in theby signal conditioning circuit. Then, the scattered light signal sare by an eight-path synchronous synchronous data acquisition card with a sample rate ofacquired 12.5 KHz and processed using data LabVIEW. acquisition card with a sample rate of 12.5 KHz and processed using LabVIEW.

Figure 2.Composition of the particle measurement system.

Figure 2.Composition of the particle measurement system.

Figure 2. Composition of the particle measurement system.

Sensors 2017, 17, 1033

4 of 15

Sensors 2017, 17, 1033

50 Hz notch filter

4 of 15

High-pass filter

Voltage (V) Voltage (V)

50 Hz notch filter

Voltage (V) Voltage (V)

Voltage (V) Voltage (V)

Sensors 2017, 17, 1033

4 of 15

High-pass filter

0

0 Time (s)

0

Time (s)

Time (s)

2.5

0

0 Time (s)

0 Time (s)

Voltage (V)

2.0

Time (s)

1.5 2.5 1.0

Voltage (V)

2.0 0.5 1.5 0.0 1.0

0

20

40

60

80

100

120

100

120

Time (ms) 0.5 0.0

Figure 3.Signal processing steps.

0

40 60 80 Figure20 3. Signal processing steps.

Time (ms)

Similar to the RMS (Root Meam Square) moving average method in reference [18], a signal Figure 3.Signal processing steps. Similar toprogram the RMS (Root Meam Square) moving average method inscattered reference [18], a signal processing was written using LabVIEW. As illustrated in Figure 4, the light signals of the three photodiodes are conditioned by Median Filter, Moving Average Filter and so on. The processing program was written using LabVIEW. As illustrated in Figure 4, the scattered light Similar to the RMS (Root Meam Square) moving average method in reference [18], a signal rectangular window is applied to the record scattered light signals before RMS computation. In signals of the three photodiodes conditioned byillustrated Median Filter, Average Filter and so processing program was written are using LabVIEW. As in FigureMoving 4, the scattered light signals addition, the averaging time of the RMS value is one second. Mass concentrations as well as voltage on. The rectangular window is applied to the record scattered light signals before RMS computation. of the three photodiodes are conditioned by Median Filter, Moving Average Filter and so on. The signals are recorded once per second. In addition, the averaging of the valuescattered is one second. Mass concentrations as well as voltage rectangular window istime applied to RMS the record light signals before RMS computation. In Vout signals are recorded once per addition, the averaging timesecond. of the RMS value is one second. Mass concentrations as well as voltage signals are recorded once per second. Voltage (V) Voltage (V)

Vout Filter Median

MedianAverage Filter Moving Filter

Basic Averaged RMS Converter Average Filter

VRMS 

Average Filter Linear Transformation Linear CPM2.5 Transformation

VRMS

(a)

N

1 N

Vi 2

1 N

V

Time (s)

0 Time (s)

i 1

3

VRMS 

N

i

i 1

mass concentration (μg/m ) PM2.5 mass PM2.5 concentration (μg/m3)

Moving Average BasicFilter Averaged RMS Converter

0

2

LabVIEW signal processing LabVIEW signal processing

0 Time (s)

(b) 0

CPM2.5 VRMS Time (s) (b) waveforms. Figure 4.Schematic diagram of software signal processing: (a) flow diagram;

3. Theory of Data Fusion

(a)

(b)

Figure 4.Schematic diagram of software signal processing: (a) flow diagram; (b) waveforms.

Figure 4. Schematic diagram of software signal (a) flow An empirical Correction Factor (CF) was usedprocessing: to compensate thediagram; influence(b) ofwaveforms. RH [10,12,23]. However, empirical CF values vary greatly with particle composition and size distribution. 3. Theory of Data Fusion 3. Theory of Data Fusion formula will introduce substantial deviation when RH cannot be measured Moreover, an empirical An empirical Correction Factor (CF) was used to compensate the influence of RH [10,12,23]. precisely. Other than measuring RH or adding extra equipment to remove particle-bound water, the An empirical Correction Factor (CF) was used compensate the influence RH [10,12,23]. However, empirical CF values vary greatly with to particle composition and size of distribution. discrepancy of the scattering pattern under different RH conditions can be used to correct PM2.5 Moreover, an empirical formula vary will introduce deviation when RH cannot be measured However, empirical CF values greatly substantial with particle composition and size distribution. precisely. than formula measuringwill RHintroduce or adding extra equipment to remove particle-bound the Moreover, an Other empirical substantial deviation when RH cannotwater, be measured discrepancy the scattering pattern under different conditionstocan be usedparticle-bound to correct PM2.5water, precisely. Otherofthan measuring RH or adding extraRH equipment remove

the discrepancy of the scattering pattern under different RH conditions can be used to correct PM2.5

Sensors 2017, 17, 1033

5 of 15

measurements. According to Lorenz-Mie theory, when a particle’s diameter increases, the scattering light will turn to the forward direction. Multi-angle light scattering can be used to account for the effect of the hygroscopic growth factor and other weather conditions. A multi-angle light scattering technique has been used in the particle measurement field and can be used to distinguish the type of weather [24,25] as well as the typology of aerosols [26,27]. Based on the above analysis, the Weather Index (WI), defined as the ratio of scattered light flux collected by photodiodes at two forward angles, is proposed to distinguish different weather conditions, especially different RHs, as shown in Equation (1). WI =

F (40◦ ) F (55◦ )

(1)

where F (40◦ ) and F (55◦ ) are scattered light flux collected by the photodiodes at 40◦ and 55◦ angles, respectively, in the particle sensor. High RH will lead to a considerable hygroscopic growth factor and then a larger diameter for a single particle and a larger MMD (Mass Median Diameter) for a particle group. Simulations based on Lorenz-Mie theory are used to find the relationship between particle size distribution and WI. The angular scattered light flux of a single particle is calculated as follows: F (θ, D, m) =

Z ϕ2

dϕ

Z θ2

ϕ1

Is ( D, m, λ)r2 sin θdθ

(2)

θ1

where ϕ is the azimuth angle between the scattering plane and polarization direction, θ is the scattering angle, Is is scattering light intensity at one point in space, λ is the wavelength of the light source, D is the particle diameter, m = n + ki is a complex refractive index of the particle, n and k are the real and imaginary parts of the refractive index respectively. The log-normal distribution is often used to estimate the particle size distribution of aerosols. A particle’s diameter D is log-normally distributed if the logarithm of D and the particle’s volume V are normally distributed, " # 1 dV −(ln D − ln V MD )2 = √ exp d ln D V 2π ln σg 2(ln σg )2

(3)

V MD and σg are volume median diameter and geometric standard deviation, respectively. For a group of particles, angular scattering flux is calculated as follows: i= N

F (θ, m) =

∑

P( Di ) F (θ, Di , m)

(4)

i =1

P( Di ) is the ratio of the volume of a particle with diameter Di to the total volume of particles in the group: V ( Di ) P ( Di ) = (5) i= N V ( D ) ∑ i i =1

The typical size distribution of fine particle matter (PM2.5) is log-normal, the MMD is 0.3 µm, the geometric standard deviation is 2.0. VMD (Volume Median Diameter) and the MMD is the same if all the particles are assumed to have the same mass density. MMD of aerosol sulfate is typically 0.54 µm and 0.20 µm [28]. The aerosol size distribution of Los Angeles Smog is bimodal, and the size distribution in PM2.5 size range is a log normal distribution with an MMD of 0.3 µm and a geometric standard deviation of 2.25 [29]. In the simulations, MMD will vary from 0.05 µm to 2.5 µm while the geometric standard deviation will remain at 2.0. Single particles and particle groups are both considered. The real part of the refractive index will vary from 1.41 to 1.51, and the imaginary part

Sensors 2017, 17, 1033

6 of 15

Sensors 2017, 17, 1033 Sensors 2017, 17, 1033

6 of 15 6 of 15

realwill partvary of the refractive index vary from 1.41 to 1.51, and the imaginary part of with of theconsidered. refractive The index from 0.00 to 0.15.will Considering carbonaceous particles, particles considered. The realwill partvary of the refractive index will vary from 1.41 to 1.51, and the imaginary part of the refractive index from 0.00 to 0.15. Considering carbonaceous particles, particles with a a highthe refractive index are also taken into consideration. refractive index will vary from 0.00 to 0.15. Considering carbonaceous particles, particles with a high refractive index are also taken into consideration. Simulation results are shown in Figures 5 and 6. highSimulation refractive index are also taken consideration. results are shown ininto Figures 5 and 6. FromSimulation Figure 5, results it can are be concluded: (1) For a single particle, the weather index increases with shown in Figures 5 and 6. From Figure 5, it can be concluded: (1) For a single particle, the weather index increases with From Figure 5, it can be concluded: (1) For a single particle, the index increases with diameter and reaches the maximum value at about 1.0 µm; index of absorbing particles diameter and reaches the maximum value at about 1.0 (2) μm;the (2)weather the weather weather index of absorbing diameter and reaches the maximum value at about 1.0 μm; (2) the weather index of absorbing is larger thanisthat of non-absorbing particles; particles; and (3) the index decreases with with the real particles larger than that of non-absorbing andweather (3) the weather index decreases the part particles is the larger than that of non-absorbing particles; and (3) the index decreases of thereal refractive index when theparticle’s diameter is between 1.0 weather µm 1.0 andμm 1.5and µm. part of refractive index when theparticle’s diameter is between 1.5 μm. with the real part of the refractive index when theparticle’s diameter is between 1.0 μm and 1.5 μm. m=1.41+0.00i m=1.47+0.00i m=1.41+0.00i m=1.47+0.00i

5 5

m=1.43+0.00i m=1.49+0.00i m=1.43+0.00i m=1.49+0.00i

m=1.45+0.00i m=1.51+0.00i m=1.45+0.00i m=1.51+0.00i

12 10

4

10

Weather index Weather index

4

Weather index Weather index

m=1.43+0.05i m=1.43+0.10i m=1.43+0.05i m=1.43+0.15i m=1.43+0.10i m=1.80+0.50i m=1.43+0.15i m=1.80+0.50i

12

3 3

2 2

1

m=1.47+0.05i m=1.47+0.10i m=1.47+0.05i m=1.47+0.15i m=1.47+0.10i m=1.90+0.50i m=1.47+0.15i m=1.90+0.50i

m=1.51+0.05i m=1.51+0.10i m=1.51+0.05i m=1.51+0.15i m=1.51+0.10i m=2.00+0.50i m=1.51+0.15i m=2.00+0.50i

8 8 6 6 4 4 2

1

2

0

0

0.0 0 0.0

0.5

1.0

2.0

2.5

1.5 2.0 particle (um) Diameter of single particle (um)

1.5

2.5

0.5 1.0single Diameter of

0.0 0 0.0

(a) (a)

0.5

1.0

1.5

2.0

2.5

2.0 (um) Diameter of single particle (um)

2.5

0.5 1.0single particle 1.5 Diameter of

(b) (b)

Figure 5. Relationship between the diameter of single particle and the weather index: (a) non-

Figure 5. Relationship between the diameter of single particle and the index:index: (a) non-absorbing Figure 5. particles; Relationship between the diameter of single particle andweather the weather (a) nonabsorbing (b) absorbing particles. particles; (b) absorbing particles. absorbing particles; (b) absorbing particles. m=1.41+0.00i m=1.47+0.00i m=1.41+0.00i m=1.47+0.00i

3.0 3.0

m=1.43+0.00i m=1.49+0.00i m=1.43+0.00i m=1.49+0.00i

m=1.45+0.00i m=1.51+0.00i m=1.45+0.00i m=1.51+0.00i

8 8 7 7

m=1.43+0.05i m=1.43+0.10i m=1.43+0.05i m=1.43+0.15i m=1.43+0.10i m=1.80+0.50i m=1.43+0.15i m=1.80+0.50i

m=1.47+0.05i m=1.47+0.10i m=1.47+0.05i m=1.47+0.15i m=1.47+0.10i m=1.90+0.50i m=1.47+0.15i m=1.90+0.50i

m=1.51+0.05i m=1.51+0.10i m=1.51+0.05i m=1.51+0.15i m=1.51+0.10i m=2.00+0.50i m=1.51+0.15i m=2.00+0.50i

6

erd in W eW athe ath er in e xd e x

Weather index Weather index

2.5 2.5

6

5 5

4

2.0

4

3

2.0

3 2 2 1.5

1

0.0 1.5 0.0 MMD

0.5

1.0

1.5

2.0

2.5

0.5Median Diameter) 1.0 1.5 particle 2.0 2.5 (Mass of group (um) MMD (Mass Median Diameter) of particle group (um)

(a) (a)

0.0 0.5 1.0 1.5 2.0 2.5 1 0.0 0.5Median Diameter) 1.0 1.5particle 2.0 2.5 MMD (Mass of group (um)

MMD (Mass Median Diameter) of particle group (um)

(b) (b)

Figure 6. Relationship between MMD (Mass Median Diameter) of particle groups and the weather Figure 6. Relationship between MMD (Mass Median Diameter) of particle groups and the weather

index: (a) non-absorbing particles; (b) (Mass absorbing particles. Figure 6. Relationship between MMD Median Diameter) of particle groups and the weather index: (a) non-absorbing particles; (b) absorbing particles. index: (a) non-absorbing particles; (b) absorbing particles. From Figure 6, similar conclusions can be drawn: (1) For a group of non-absorbing particles, the From Figure 6, similar conclusions can beand drawn: (1) For group of non-absorbing particles, the weather index increases sharply with MMD reaches the amaximum value when MMD is about From Figure 6, similar conclusions can be drawn: (1) For a group of non-absorbing particles, weather index increases sharply with MMD and reaches the maximum value when MMD is about 0.6 μm and then decreases slowly. (2) For a group of absorbing particles, the weather index increases 0.6 μm and then decreases slowly. (2) For a group absorbing the value weather index increases the weather index increases sharply MMD andof reaches theparticles, maximum when is about with MMD and the weather indexwith decreases with the real part of the refractive index andMMD increases with MMD and the weather index decreases with the real part of the refractive index and increases 0.6 µm and then decreases slowly. (2) For a group of absorbing particles, the weather index increases with the imaginary part of the refractive index when MMD is between 0.5 μm and 1.0 μm. withWater theand imaginary part ofindex the when MMD isabsorption between μmbecome and 1.0 μm.and increases with MMD the weather with theand real part of the0.5 refractive index absorbs slightly in refractive thedecreases visibleindex spectrum, the will stronger with Water absorbs slightly in the visible spectrum, and the absorption will become stronger with wavelength [30].part The of operating wavelength the laser is 648 and the of water with the imaginary the refractive indexofwhen MMD is nm, between 0.5refractive µm and index 1.0 µm. wavelength [30]. The operating wavelength of the laser is 648 nm, and the refractive index of water −8 is m = 1.33 + 1.5 ×slightly 10 i, so in laser absorbed by water canthe be neglected. Water absorbs thelight visible spectrum, and absorption will become stronger with −8i, so laser light absorbed by water can be neglected. is m As = 1.33 + 1.5 × 10above, mentioned high RH will have three effects on the particle’s properties: (1)index particle wavelength [30]. The operating wavelength of the laser is 648 nm, and the refractive ofsize water is As mentioned above, highgrowth RH willfactor, have three effects onreal thepart particle’s (1) particle size increases with the hygroscopic (2) a particle’s of theproperties: refractive index decreases − 8 m = 1.33 + 1.5 × 10 i, so laser light absorbed by water can be neglected. increases the hygroscopic growth (2) a (3) particle’s real part of the refractive index decreases by addingwith the lower refractive index offactor, the water, a particle’s shape tends to becomesmoother by As above, high RH willofhave three effects on the shape particle’s (1) particle by mentioned adding the lower refractive index the water, (3) a particle’s tendsproperties: to becomesmoother by size

increases with the hygroscopic growth factor, (2) a particle’s real part of the refractive index decreases by adding the lower refractive index of the water, (3) a particle’s shape tends to becomesmoother by absorbing water. According to the simulation results above, as a whole, high RH will result

Sensors 2017, 17, 1033

7 of 15

Sensors 2017, 17, 1033

7 of 15

in a larger WI by making MMD larger, and the real part of the refractive index of the particle group 2017,smaller. 17, 1033 According 7 of 15of the absorbing water. to the simulation above,isasmuch a whole, highsignificant RH will result in athat larger will Sensors become Furthermore, the effectresults of MMD more than WI by making MMD larger, and the real part of the refractive index of the particle group will become refractive index. Our simulation results are in agreement with, e.g., Sioutas et al., who reported absorbing water. According to the simulation results above, as a whole, high RH will result in a larger smaller.[11] Furthermore, the effect of MMD is much most more important significant than that of theaffecting refractive index. in reference that thelarger, aerosol MMD is the parameter the response WI by making MMD and the real partsingle of the refractive index of the particleingroup will become Our simulation results are in agreement with, e.g., Sioutas et al., who reported in reference [11] that of PM2.5 measurement instruments based on light scattering, and the effect of particle chemical smaller. Furthermore, the effect of MMD is much more significant than that of the refractive index. the aerosol MMD is the single most important parameter in affecting the response of PM2.5 composition on PM2.5 measurement results much lessetimportant than particle size[11] distribution. Our simulation results are in agreement with,ise.g., Sioutas al., who reported in reference that measurement instruments based on light scattering, and the effect of particle chemical composition Therefore, the weather can bemost usedimportant to correct parameter the PM2.5 in measurements. the aerosol MMD isindex the single affecting the response of PM2.5 on PM2.5 measurement results is much less important than particle size distribution. Therefore, the measurement instruments based on light scattering, and the effect of particle chemical composition weather index can be used to correct the PM2.5 measurements. 4. Results andmeasurement Discussion results is much less important than particle size distribution. Therefore, the on PM2.5 weather index be used to correct the PM2.5 measurements. 4. Results andcan Discussion

Field experiments were conducted at the Wolongqiao monitoring station, anational monitoring station ofField theand National Ambient Air Quality Automaticmonitoring Monitoring Network, operated by the experiments were conducted at the Wolongqiao station, anational monitoring 4. Results Discussion ◦ 0 ◦ 0 37” E. Ministry of Environmental Protection China. Automatic The stationMonitoring is situatedNetwork, at 30 14 operated 44” N, 120 station of the National Ambient AirofQuality by 07 the Field experiments were conducted at the Wolongqiao monitoring station, anational monitoring Ministry of in Environmental of China. The station by is situated 30°14′44″ N, 120°07′37″ E. As illustrated Figure 7, Protection the station is surrounded forestsatand is suitable for studies station of the National Ambient Air Quality Automatic Monitoring Network, operated by the Asmeteorological illustrated in Figure 7, of thePM2.5 station monitoring. is surrounded There by forests and isexpressway suitable for studies on the side on the effect is a city to the north Ministry of Environmental Protection of China. The station is situated at 30°14′44″ N, 120°07′37″ E. meteorological of PM2.5 There is astation. city expressway to theexhaust north side theof stationmain of the andinaeffect tunnel themonitoring. southwest of the Automobile isof one Asstation illustrated Figure 7,tothe station is surrounded by forests and is suitable for studies onthe the and a tunnel to the southwest of the station. Automobile exhaust is one of the main sources of air sources of air pollution onPM2.5 weekends and holidays, during the meteorological effect of monitoring. There is awhile city expressway to working the north days, side ofair thequality station near pollution on weekends and holidays, while during the working days, air quality near the station is the station is better thansouthwest other areas of the city.Automobile The stationexhaust is equipped TEOM 1405-D and a tunnel to the of the station. is one with of thea main sources of monitor air better than other areas of the city. The station is equipped with a TEOM 1405-D monitor (Thermo 3 ) to pollution onScientific weekendsInc., and Waltham, holidays, while duringwith the working days, air quality near themeasure station isPM2.5 (Thermo Fisher MA, USA, a precision of ± 2.0 µg/m 3 Fisher Scientific Inc., Waltham, MA, USA, with a precision of ±2.0 μg/m ) to measure PM2.5 and PM10 than other areas of the city. The station is and equipped with a TEOM 1405-D monitor (Thermo and better PM10 mass concentrations simultaneously canasbe as aPM2.5 reference PM2.5 instrument mass concentrations simultaneously and can be used a used reference instrument in field 3) to measure PM2.5 and PM10 Fisher Scientific Inc., Waltham, MA, USA, with a precision of ±2.0 μg/m in field experiments. experiments. mass concentrations simultaneously and can be used as a reference PM2.5 instrument in field The experimental system is shown in Figure 8.The air isair introduced intointo a PM2.5 impactor through The experimental system is shown in Figure 8.The is introduced a PM2.5 impactor experiments. a gas path after filteringparticles the coarse particles using a Dorr-Oliver A commercial a gasthrough path after filtering the coarse using a Dorr-Oliver cyclone. Acyclone. commercial temperature The experimental system shown in Figure 8.The air is introduced into a PM2.5 impactor ◦ Cisand temperature sensor with precision ofan±0.5 and anwith RH sensor with aof precision of ±3% RH are sensor with a precision of ±a0.5 RH°Csensor a precision ±3% RH are used toused measure through a gas path after filtering the coarse particles using a Dorr-Oliver cyclone. A commercial to measure temperature and RH, respectively. The particle sensor is assembled into a metal chassis, temperature and RH, respectively. The particle sensor is assembled into a metal chassis, which temperature sensor with a precision of ±0.5 °C and an RH sensor with a precision of ±3% RH are used can which can effectively suppress electromagnetic interference and environmental noises. The output effectively suppress electromagnetic interference and environmental noises. The to measure temperature and RH, respectively. The particle sensor is assembled into output a metal signals chassis, from signals from the particle sensor were acquired using a DAQ (Data Acquisition) card (Daqvantech Inc., the particle sensor were suppress acquiredelectromagnetic using a DAQ interference (Data Acquisition) card (Daqvantech Suzhou, which can effectively and environmental noises. TheInc., output Suzhou, China) with a sample rate of 12.5 KHz. Then, the data were processed and displayed on a signals from the particle sensor were acquired using a DAQ (Data Acquisition) card (Daqvantech Inc., China) with a sample rate of 12.5 KHz. Then, the data were processed and displayed on a PC. PC. Suzhou, China) with a sample rate of 12.5 KHz. Then, the data were processed and displayed on a PC.

(a)

(b)

Figure (a) 7. Wolongqiao monitoring station: (a) location; (b) appearance. Figure 7. Wolongqiao monitoring station: (a) location; (b)(b) appearance. Figure 7. Wolongqiao Temp station: Hum(a) location; (b) appearance. Gas monitoring Dorr-Oliver Cyclone Dorr-Oliver Cyclone

Sensor Sensor Path Temp Hum Gas Path PM2.5 Sensor Sensor Impactor Particle PM2.5 Sensor Impactor Particle Sensor DAQ Card

PC

DAQ Card PC Figure 8. Schematic diagram of the experimental system.

Figure Schematic diagram diagram of system. Figure 8.8.Schematic ofthe theexperimental experimental system.

Sensors 2017, 17, 1033

8 of 15

The new particle Sensors 2017, 17, 1033 sensor was tested at the Wolongqiao monitoring station from 7 to 14 March 8 of 152016. Hourly PM2.5 and PM10 mass concentrations from the TEOM 1405-D monitor were recorded The newHalf-hourly particle sensormeteorological was tested at theparameters Wolongqiaosuch monitoring station from 7 to 14 March simultaneously. as visibility, barometric pressure, Hourly and PM10 mass concentrations the TEOM 1405-D monitor were recorded wind 2016. speed, windPM2.5 direction and weather types fromfrom a nearby weather station were also recorded. Half-hourly meteorological as to visibility, barometric pressure, wind Beforesimultaneously. experiments are conducted, zero-point parameters calibrationsuch needs be conducted using a HEPA (High speed, wind direction and weather types from a nearby weather station were also recorded. Before Efficiency Particulate Air) filter (TSI Inc., Shoreview, MN, USA). experiments are conducted, zero-point calibration needs to be conducted using a HEPA (High Figure 9 shows the relationship between the output voltages of the three photodiodes Efficiency Particulate Air) filter (TSI Inc., Shoreview, MN, USA). in the particle 1-h average PM2.5 given by TEOMin 1405-D. Figure sensor 9 showswith the relationship between the mass outputconcentrations voltages of the three photodiodes the Quadratic polynomial regression is adopted considering the nonlinearity. The degree particle sensor with 1-h average PM2.5 mass concentrations given by TEOM 1405-D.fitting Quadratic 2 ) and root mean square of curvilinear regression is examined by coefficient of determination (R polynomial regression is adopted considering the nonlinearity. The fitting degree of curvilinear 2) and error (RMSE). is a significant correlation between(Rthe output of thesquare photodiodes and the PM2.5 regressionThere is examined by coefficient of determination root mean error (RMSE). There is a significant correlation between the output of the photodiodes and the PM2.5 mass concentration. mass concentration. 400

0

25

50

75

100

125

150

175

200

2

Y=0.0082X +0.4506X+17.937 R2=0.9451 N=158

350 300 250 200 150 100

Original voltage signal (mV)

50

55° photodiode

0

Y=0.02417X2+0.8245X+21.9182 R2=0.9106 N=158

1000 800 600 400 200

40° photodiode 0 2

Y=0.0041X -0.0320X+70.9183 R2=0.9077 N=158

200 150 100 50

140° photodiode 0

0

25

50

75

100

125

150

175

200

Reference PM2.5 mass concentration (ug/m3)

Figure 9. Output voltage of the three photodiodes in the particle sensor vs. 1-h average reference Figure 9. Output voltage of the three photodiodes in the particle sensor vs. 1-h average reference PM2.5 mass concentration. PM2.5 mass concentration.

As illustrated in Figure10, 10, the the scatter of the 1-h average PM2.5 PM2.5 mass concentrations based As illustrated in Figure scatterplot plot of the 1-h average mass concentrations andand the the 40° photodiode in the in particle sensor follow two distinct trends. basedon onthe the55° 55◦photodiode photodiode 40◦ photodiode the particle sensor follow two distinct According to the variation of the weather index, the weather index is relatively stable, i.e., i.e., trends. According to the variation of the weather index, the weather index is relatively stable, approximately 3.0 for the first 20 h, and then drops sharply below 0.0 at 8 March 201614:00 with approximately 3.0 for the first 20 h, and then drops sharply below 0.0 at 8 March 2016 14:00 with decreasing PM2.5 mass concentrations; then, the weather index increases sharply to about 2.5 and decreasing PM2.5 mass concentrations; then, the weather index increases sharply to about 2.5 and becomes stable until the end of the measurement. One of the possible reasons that the weather index becomes the end of the the40° measurement. Onenot of receive the possible reasons thatsignal the weather index can stable be less until than zero is that photodiode does a sufficient voltage when the ◦ photodiode does not receive a sufficient voltage signal when the can bePM2.5 less than zero is that the 40 3 mass concentration is very low (e.g., less than 20 μg/m ) and the voltage of the 40° photodiode 3 ) and the voltage of the 40◦ photodiode PM2.5can mass concentration is very lownoise (e.g.,voltage less than 20 µg/mfrom even be negative when system is removed photodiode voltage. This abnormal can even be negative when system voltage is removed fromisphotodiode abnormal phenomenon only occurs whennoise the PM2.5 mass concentration less than 20 voltage. μg/m3andThis should be considered separately. One the of PM2.5 the possible solutions is toisuse mass be phenomenon only occurs when mass concentration lessthe thantraditional 20 µg/m3PM2.5 and should concentration of the 40° transformation of traditional the 40° photodiode voltage signal) considered separately. One ofphotodiode the possible(linear solutions is to use the PM2.5 mass concentration ◦ ◦ of the 40 photodiode (linear transformation of the 40 photodiode voltage signal) when the PM2.5

Sensors 2017, 17, 1033

9 of 15

Sensors 2017, 17, 1033

9 of 15

mass concentration is less than 20 µg/m3 . Because the particle size distribution can only change 3 PM2.5 concentration less than 20 μg/m the particle size of distribution can index. slowly, when data the points aremass broken into twois continuous parts. Because according the value the weather only change slowly, data points are broken into two continuous parts according the value of the The two parts are separated by 8 March 2016 at 14:00 h. Both parts of the data points were analyzed weather index. The two parts are separated by 8 March 2016 at 14:00 h. Both parts of the data points by linear regression, and the regression, results were indicating that during that different were analyzed by linear and excellent, the results were excellent, indicating duringperiods different of time, parameters of the PM2.5 particles are distinct, while parameters are stable during the same periods of time, parameters of the PM2.5 particles are distinct, while parameters are stable during the period of time.same period of time. 2016/03/07 12:00~2016/03/08 14:00 2016/03/08 15:00~2016/03/14 01:00

4.0

Y=2.936X-34.3 R2=0.9988 Y=2.399X-24.05 R2=0.998

1000

2016/3/8 14:00

3.5 3.0

800

W e ath er in d e x

40° photodiode voltage signal (mV)

1200

600

400

2.5 2.0 1.5 1.0

200

0.5

0 0

100

200

300

400

55° photodiode voltage signal (mV)

0.0 2016/3/7 2016/3/8 2016/3/9 2016/3/10 2016/3/11 2016/3/12 2016/3/13 2016/3/14 12:00 12:00 12:00 12:00 12:00 12:00 12:00 12:00

(a)

Date

(b)

Figure 10. Parameters of the weather index: (a) Scatter plot of the 55° photodiode voltage signals vs.

Figure 10. Parameters of the weather index: (a) Scatter plot of the 55◦ photodiode voltage signals the 40° photodiode voltage signals, (b) Variation of the weather index during the measurement ◦ photodiode voltage signals, (b) Variation of the weather index during the measurement period. vs. the 40 period. Variations of the meteorological parameters during the experiment are illustrated in Figure 11. Variations of the meteorological parameters during the experiment are illustrated in Figure 11. During the former period of time, temperature is higher than 12 °C;◦RH is mostly higher than 93% During RH; the visibility former period of time, temperature is higher than 12 C; RH is mostly higher than 93% is less than 3 km; and barometric pressure is lower than 1015 hPa, while during the RH; visibility is less than 3 km; and barometric pressurebutis remains lower than hPa, during the latter period of time, temperature changes significantly below1015 12 °C; RH while also varies ◦ latter period time, temperature changes significantly but remains below C; the RHsame also varies greatly,of but remains lower than 93% RH; visibility and barometric pressure follow12 almost visibility lower is greater than93% 3 km; andvisibility barometricand pressure is higher pressure than 1015 hPa during most of greatly,trend; but remains than RH; barometric follow almost the same the time. It is obvious that meteorological parameters during the former period of time are quite trend; visibility is greater than 3 km; and barometric pressure is higher than 1015 hPa during most different from those during the latter period of time. Therefore, it is reasonable that the weather index of the time. It is obvious that meteorological parameters during the former period of time are quite can help to distinguish meteorological parameters. MMD is the most important factor that affects the differentmeasurement from thoseof during the latter ofscattering time. Therefore, it is reasonable that the the weather instruments basedperiod on light [11]. It is obvious that MMD during two index can help to distinguish meteorological parameters. MMD is the most important periods of time is quite different. From the simulation results presented in chapter 3,factor a highthat WI affects the measurement of instruments based on the light scattering is obvious during the indicates a large MMD. Therefore, during former period of[11]. time,Ita larger MMD that can beMMD estimated, while of during of time, a smaller MMD is reasonable. According tointhe experimental two periods timethe is latter quiteperiod different. From the simulation results presented Chapter 3, a high WI results of other researchers, when MMD varies between 0.3 μm and 1.0 μm, the ratio of particle indicates a large MMD. Therefore, during the former period of time, a larger MMD can be estimated, measurement results based on light scattering method to results based on the dynamic weighing while during the latter period of time, a smaller MMD is reasonable. According to the experimental method will increase with increasing MMD [11]. results of other researchers, whenofMMD between 0.3 µm and using 1.0 µm, thephotodiode ratio of particle In Figure 12, the correction PM2.5 varies mass concentrations is displayed the 40° measurement resultsAccording based ontolight scattering toofresults based dynamic weighing as an example. the weather index,method the points the scatter ploton arethe divided into two the blackwith onesincreasing and red ones. For different methodgroups: will increase MMD [11]. groups, a unique quadratic polynomial regression ◦ photodiode model is12, used describe the between the outputisvoltage of theusing photodiode the 40° In Figure thetocorrection ofrelationship PM2.5 mass concentrations displayed the 40at as angle and the reference PM2.5 concentrations. The results are very good, with R2 ranging between an example. According to the weather index, the points of the scatter plot are divided into two groups: 0.9817 and 0.9575. During the former period of time, the data points follow a linear trend while during the black ones and red ones. For different groups, a unique quadratic polynomial regression model the latter period of time, the data points follow a non-linear trend. It is not unusual. During the whole is used experiment, to describewe the relationship the output voltage of the photodiode at the 40◦ angle used an index of between 1.5 to describe the nonlinear relationship between scattered light 2 and theflux reference PM2.5 concentrations. results are very good, expression. with R ranging between 0.9817 and corresponding particle volume,The as illustrated in the following and 0.9575. During the former period of time, the data points follow a linear trend while during the (6) V  F 1.5 latter period of time, the data points follow a non-linear trend. It is not unusual. During the whole where V is the particle volume and F is scattered light flux. experiment, we used an index of 1.5 to describe the nonlinear relationship between scattered light flux and corresponding particle volume, as illustrated in the following expression.

V ∝ F1.5 where V is the particle volume and F is scattered light flux.

(6)

Sensors 2017, 17, 1033 Sensors 2017, 17, 1033

10 of 15 10 of 15

Sensors 2017, 17, 1033

10 of 15 2016/3/8 14:00 Barometric pressure Visibility

2016/3/8 14:00

1035 1030

10 Barometric pressure Visibility

1030 1025 1025 1020

5 5

1020 1015 1015 1010

0

1010 1005

20 25

100% 80%

15 20

60%

5 10 5

-5

0

2016/3/7 12:00 -5

RH

80% 60%

10 15

RH

Temperature (°C) Temperature (°C)

0 100%

Temperature Relative humidity Temperature Relative humidity

25 1005

0

10

Visibility (km) Visibility (km)

Barometric pressure （hPa） Barometric pressure （hPa）

1035

40% 40% 20% 2016/3/8 12:00

2016/3/7 12:00

2016/3/9 2016/3/10 2016/3/11 2016/3/12 2016/3/13 2016/3/14 12:00 12:00 12:00 12:00 12:00 12:00 20%

2016/3/8 12:00

Date 2016/3/11 2016/3/12 2016/3/13 2016/3/14 2016/3/9 2016/3/10 12:00 12:00 12:00 12:00 12:00 12:00

Date Figure Figure11. 11.Meteorological Meteorologicalparameters parametersduring duringthe themeasurement measurementperiod: period:Temperature, Temperature,RH, RH,visibility visibility and barometric pressure. and barometric pressure. parameters during the measurement period: Temperature, RH, visibility Figure 11. Meteorological

and barometric pressure. 2016/03/07 12:00~2016/03/08 14:00 2016/03/08 15:00~2016/03/14 01:00 2 12:00~2016/03/08 14:00 Y=-0.01X2016/03/07 +5.8046X-6.279 2 2016/03/08 15:00~2016/03/14 01:00

R 1200 1000

=0.9817 2 2 Y=-0.01X +5.8046X-6.279 +1.058X+8.895 Y=0.019X 2 R2=0.9817 R =0.9575 1000 Y=0.019X2+1.058X+8.895 800 R2=0.9575

40° photodiode voltage signal (mV)

40° photodiode voltage signal (mV)

1200

600

400

200

0

800

600

400

200 0

25 0

50

75

100

125

150

175

200

Reference PM2.5 mass concentration (ug/m3)

0

25

50

75

100

125

150

175

200

Reference PM2.5 mass concentration (ug/m3)

Figure 12. Relationship between PM2.5 mass concentrations measured using the 40° photodiode and ◦ Figure Relationship between mass measured reference PM2.5 mass concentrations at Wolongqiao monitoring station.using Figure12. 12. Relationship betweenPM2.5 PM2.5 massconcentrations concentrations measured usingthe the4040°photodiode photodiodeand and reference PM2.5 mass concentrations at Wolongqiao monitoring station. reference PM2.5 mass concentrations at Wolongqiao monitoring station.

This index is based on the theory of Fraunhofer diffraction, which is valid for particles larger than 2This μm index and isiswidely in theory PM10 instruments [31,32]. this indexlarger can This index isbased basedused onthe the theorymeasurement Fraunhofer diffraction,which whichisTherefore, isvalid validfor forparticles particles larger on ofofFraunhofer diffraction, be accurately applied when a larger MMD is estimated.instruments However, the resultsTherefore, will showthis nonlinearity than μmand and widely used PM10 measurement instruments [31,32]. Therefore, thisindex indexcan can than 22µm isiswidely used ininPM10 measurement [31,32]. when a smaller MMD is estimated. beaccurately accuratelyapplied appliedwhen whenaalarger largerMMD MMDisisestimated. estimated.However, However,the theresults resultswill willshow shownonlinearity nonlinearity be The following example demonstrates how the WI is used to correct the 40° photodiode voltage when smallerMMD MMD estimated. when a asmaller isisestimated. signal.The Thefollowing following example demonstrates WIWI is used to correct the 40° voltage demonstrateshow howthethe is used to correct thephotodiode 40◦ photodiode Y  0.1807X  0.9104 When WI is higher than 2.7, the linear equation is used to correct the signal. signal. voltage ◦ 40° photodiode signal obtain adjusted PM2.5 concentrations of the 40° photodiode 0.1807X 0.1807X  0.9104 WhenWI WIvoltage higher thanand 2.7,the thelinear linear equation is used to correct When isishigher than 2.7, equation YY =mass + 0.9104 is used to correct the 40the ◦ photodiode for thephotodiode formervoltage period of time; Xand is the 40° photodiode voltage signal, and Y is adjusted PM2.5 40° voltage signal obtain adjusted PM2.5 mass concentrations of the photodiode photodiode signal and obtain adjusted PM2.5 mass concentrations of the the 4040° ◦ mass concentration. forthe theformer formerperiod periodofoftime; time;XXisisthe the4040°photodiode photodiodevoltage voltagesignal, signal,and andYYisisthe theadjusted adjustedPM2.5 PM2.5 for 2

mass concentration. When WI is equal to or less than 2.7, the equation Y  0.0003148X  0.4109X  6.4123 is mass concentration. used toWhen correct 40° photodiode voltage and obtain PM2.5 2mass concentrations WIthe is equal to or less than 2.7, signal the equation Y adjusted 0.0003148X  0.4109X  6.4123of is theused 40° photodiode for the latter period of time. to correct the 40° photodiode voltage signal and obtain adjusted PM2.5 mass concentrations of PM2.5 mass fortime. all experimental processes can be calculated based on theAdjusted 40° photodiode for theconcentrations latter period of these two equations. Adjusted PM2.5 mass concentrations for all experimental processes can be calculated based on these two equations.

Sensors 2017, 17, 1033

11 of 15

When WI is equal to or less than 2.7, the equation Y = −0.0003148X2 + 0.4109X + 6.4123 is used to correct the 40◦ photodiode voltage signal and obtain adjusted PM2.5 mass concentrations of the 40◦ photodiode for the latter period of time. Adjusted PM2.5 mass concentrations for all experimental processes can be calculated based on these two Sensors 2017,equations. 17, 1033 11 of 15 Figure 13 displays the relationship between PM2.5 mass concentrations of three photodiodes FigureTEOM 13 displays the relationship PM2.5 mass concentrations of three photodiodes and reference 1405-D monitored between 1-h average PM2.5 mass concentrations before and after and reference TEOM 1405-D monitored 1-h average PM2.5 mass concentrations before after after correction. It can be seen that the measurement accuracy and linearity are improvedand greatly correction. It can be seen that the measurement accuracy and linearity are improved greatly after ◦ ◦ correction. Photodiodes at the angles of 40 and 55 perform very well. These results are consistent correction. Photodiodes at the angles of 40° and 55° perform very well. These 2results are consistent with results using an expensive optical particle counter (Grimm 1.108) (R = 0.97) [33] and better with results using an expensive optical particle counter (Grimm 1.108) (R2 = 0.97) [33] and better than 2 than field field experiments experimentsusing usingananaerosol aerosol monitor DustTrak = 0.859) In summary, PM2.5 monitor DustTrak (R2 (R = 0.859) [34].[34]. In summary, PM2.5 mass mass concentrations measured by the three photodiodes are similar if MMD is stable. When MMD varies, concentrations measured by the three photodiodes are similar if MMD is stable. When MMD varies, ◦ and 55◦ can the measurements will be affected. After correction by WI, photodiodes at the angle of 40 the measurements will be affected. After correction by WI, photodiodes at the angle of 40° and 55° resultcan in result a favorable performance. in a favorable performance. 0

20

40

60

80

100

120

140

160

180

200

0

Y=0.9092X+5.0714 R2=0.9310 RMSE=12.5690 ug/m3 N=158

200

150

40

60

80

100

120

150

160

180

200

100

Adjusted PM2.5 mass concentration (ug/m3)

50

55° photodiode 0

Y=0.9020X-0.3330 R2=0.9021 RMSE=17.9239 ug/m3 N=158

200

150

100

50

40° photodiode 0

Y=0.9453X+2.5288 R2=0.9251 RMSE=12.8692 ug/m3 N=158

200

150

50

55° photodiode 0

Y=0.9623X+2.5796 R2=0.9663 RMSE=8.3751 ug/m3 N=158

200

150

100

50

40° photodiode 0

Y=0.9581X+2.2724 R2=0.9485 RMSE=10.4077 ug/m3 N=158

200

150

100

100

50

50

140° photodiode

140° photodiode 0

140

Y=0.9618X+2.5288 R2=0.9640 RMSE=8.6649 ug/m3 N=158

200

100

Unadjusted PM2.5 mass concentration (ug/m3)

20

0

20

40

60

80

100

120

140

160

180

Reference PM2.5 mass concentration (ug/m3)

(a)

200

0

0

20

40

60

80

100

120

140

160

180

200

Reference PM2.5 mass concentration (ug/m3)

(b)

Figure 13. Scatter plot of the 1-h average PM2.5 mass concentrations using the reference Tapered

Figure 13. Scatter plot of the 1-h average PM2.5 mass concentrations using the reference Tapered Element Oscillating Microbalance (TEOM) 1405-D monitor at Wolongqiao monitoring station vs. Element Oscillating Microbalance (TEOM) 1405-D monitor at Wolongqiao monitoring station vs. PM2.5 PM2.5 mass concentrations based on three angular photodiodes in the particle sensor: (a) unadjusted; mass(b) concentrations based on three angular photodiodes in the particle sensor: (a) unadjusted; adjusted. (b) adjusted.

As illustrated in Table 1, after correction by the WI, the three photodiodes can achieve accurate measurement PM2.5 However, there arethe stillthree large photodiodes biases in the three As illustratedofin Tableconcentrations. 1, after correction by the WI, can photodiodes, achieve accurate especially when PM2.5 concentrations are low. The minimum PM2.5 concentrations of the three measurement of PM2.5 concentrations. However, there are still large biases in the three photodiodes, photodiodes are about 8 μg/m3 while the minimum TEOM 1405-D is 2 μg/m3.The possible reason especially when PM2.5 concentrations are low. The minimum PM2.5 concentrations of the three may be that the measurement model, which describes the relationship between the PM2.5 photodiodes are about 8 µg/m3 while the minimum TEOM 1405-D is 2 µg/m3 .The possible reason concentration and output voltage of the photodiode, does not fit the data well when the PM2.5 may concentration be that the measurement model, which describes the relationship between the PM2.5 concentration is low.

and output voltage of the photodiode, does not fit the data well when the PM2.5 concentration is low.

Sensors 2017, 17, 1033 Sensors 2017, 17, 1033

12 of 15 12 of 15

Table1.1.Summary Summaryof of1-h 1-haverage averagePM2.5 PM2.5measurements measurements(µg/m (μg/m33) by different instruments. instruments. Table

Instrument TEOM 1405-D TEOM 1405-D 55° photodiode 55◦ photodiode 40° photodiode 40◦ photodiode photodiode 140◦ 140° photodiode Instrument

Mean 63.72 63.72 63.81 63.81 63.89 63.89 63.32 63.32

Mean

Median Minimum Maximum Median Minimum Maximum 52.00 2.00 179.00 52.00 2.00 179.00 45.86 8.57 188.13 45.86 8.57 188.13 47.00 7.74 7.74 184.56 47.00 184.56 47.20 8.20 8.20 190.68 47.20 190.68

CorrectedPM2.5 PM2.5mass massconcentrations concentrationsmeasured measured photodiode according to different Corrected byby thethe 40◦40° photodiode according to different WI WI and reference PM2.5 mass concentrations are presented in Figure 14. It can be seen that corrected and reference PM2.5 mass concentrations are presented in Figure 14. It can be seen that corrected PM2.5 PM2.5concentrations mass concentrations the trend of reference PM2.5 concentrations Figure mass followfollow the trend of reference PM2.5 massmass concentrations veryvery well.well. Figure 14 14 clearly reveals several temporal PM2.5 concentration patterns during the experiment. The PM2.5 clearly reveals several temporal PM2.5 concentration patterns during the experiment. The PM2.5 concentrations dropped dropped sharply sharply with with fresh, fresh, clean clean air air brought brought by by aa cold cold wave wave from from the the northwest northwest concentrations during the the evening evening of of 77 March March 2016 2016 until until noon noon on on88March March2016. 2016. Then, Then, the the PM2.5 PM2.5 concentrations concentrations during increased slowly with temperature, and the effect of the cold wave vanished. increased slowly with temperature, and the effect of the cold wave vanished.

PM2.5 mass concentration (ug/m3)

225 200

Reference by TEOM-1405D 40° photodiode adjusted

175 150 125 100 75 50 25 0 2016/3/7 2016/3/8 2016/3/9 2016/3/10 2016/3/11 2016/3/12 2016/3/13 2016/3/14 12:00 12:00 12:00 12:00 12:00 12:00 12:00 12:00

Date

Figure Comparison of of 1-h 1-h average average PM2.5 PM2.5 mass massconcentrations concentrationsobtained obtainedsimultaneously simultaneouslyusing usinga Figure 14. 14. Comparison aTEOM TEOM1405-D 1405-Dmonitor monitoratatWolongqiao Wolongqiao monitoring monitoring station station and adjusted PM2.5 mass concentrations mass concentrations based 40◦40° photodiode in the sensorsensor according to different weather indices (the regression basedon onthe the photodiode in particle the particle according to different weather indices (the results are presented Figure 13b, center plot). regression results areinpresented in Figure 13b, center plot).

5. Conclusions Conclusions 5. In this this paper, paper, aa novel novel particle particle sensor sensor based based on on multi-angle multi-angle light light scattering scattering and and data data fusion fusion is is In presented. In weather index index is is introduced introduced to to describe describe MMD MMD presented. In the design of the particle sensor, a weather variationof of PM2.5 PM2.5 particles particles caused caused by by different different meteorological meteorologicalparameters parameterssuch suchas as RH. RH. High High RH RH will will variation enlarge the particle carried out to to demonstrate thatthat the the WI can enlarge particle diameter diametersignificantly. significantly.Simulations Simulationsare are carried out demonstrate WI distinguish the the variation of of MMD caused simulation can distinguish variation MMD causedbybydiverse diversemeteorological meteorological parameters. parameters. The simulation resultsindicate indicatethat thataa larger larger MMD MMD introduced introducedby byhigh highRH RH will will produce produceaa higher higher WI WI while while the the particle particle results refractive index index has has less lessimpact impacton onthe theWI. WI. refractive Field experiments experimentswere were carried outfurther to further the ofability of theindex weather index to Field carried out to verify verify the ability the weather to distinguish distinguish the variation of MMD caused by different meteorological parameters. The field the variation of MMD caused by different meteorological parameters. The field experiment results experiment results demonstrated that for a larger MMD, a linear between the demonstrated that for a larger MMD, a linear relationship occursrelationship between theoccurs values measured values measured by the in three thethe particle sensor andresults. the reference TEOM results. by the three photodiodes the photodiodes particle sensorinand reference TEOM However, for a smaller However, forisaasmaller MMD, there is abetween nonlinear between measured byand the MMD, there nonlinear relationship therelationship values measured bythe thevalues three photodiodes three photodiodes and the reference TEOM results. One possible reason is that thesignal index processing used in the the reference TEOM results. One possible reason is that the index used in the signal processing of the particleused sensor is widely used in PM10 measurement instruments. Therefore, of the particle sensor is widely in PM10 measurement instruments. Therefore, it is possible that it isweather possibleindex that can the weather indexthe canindex help to toreduce adjust the the impact index to the impact of MMD in the help to adjust of reduce MMD in further research. further research. Experimental results showed that the photodiode at the 55° angle can achieve good performance. After correction by the WI, photodiodes at 40° and 55° angles can both achieve good performance.

Sensors 2017, 17, 1033

13 of 15

Experimental results showed that the photodiode at the 55◦ angle can achieve good performance. After correction by the WI, photodiodes at 40◦ and 55◦ angles can both achieve good performance. This was presumably due to two reasons: (1) variations of the nonlinear relationship between scattered light flux and corresponding particle volume introduced by MMD variation are smallest for the 55◦ photodiode, while after correction by the weather index, the effect of MMD variation vanishes, and the effect of variation of a particle’s refractive index is smallest for the 40◦ photodiode; (2) the noise of the voltage signal is highest for the 140◦ photodiode and lowest for the 55◦ photodiode. It should be noted that more work needs to be done to further improve the performance of the particle sensor. For example, more field experiments should be carried out to investigate the relationship between WI and MMD of particle groups. In addition, the measurement precision, as well as the detection limit, has yet to be improved. Furthermore, a virtual impactor is a potential alternative, which can relieve the maintenance work. Acknowledgments: This work was supported by the State Key Program of National Natural Science Foundation of China (Grant No. 11132008). Author Contributions: Wenjia Shao, Hongjian Zhang and Hongliang Zhou conceived and designed the experiments; Wenjia Shao performed the experiments; Wenjia Shao and Hongliang Zhou analyzed the data; Wenjia Shao and Hongliang Zhou wrote the paper. Conflicts of Interest: The authors declare no conflict of interest.

References 1.

2. 3.

4. 5.

6.

7.

8.

9.

10. 11.

Loomis, D.; Huang, W.; Chen, G. The International Agency for Research on Cancer (IARC) evaluation of the carcinogenicity of outdoor air pollution: Focus on China. Chin. J. Cancer 2014, 33, 189–196. [CrossRef] [PubMed] Zhang, C.; Zhu, R.; Yang, W.M. A Micro Aerosol Sensor for the Measurement of Airborne Ultrafine Particles. Sensors 2016, 16, 9. [CrossRef] [PubMed] Lloyd Spetz, A.; Huotari, J.; Bur, C.; Bjorklund, R.; Lappalainen, J.; Jantunen, H.; Schütze, A.; Andersson, M. Chemical sensor systems for emission control from combustions. Sens. Actuators B Chem. 2013, 187, 184–190. [CrossRef] Hagen, G.; Feistkorn, C.; Wiegartner, S.; Heinrich, A.; Bruggemann, D.; Moos, R. Conductometric soot sensor for automotive exhausts: Initial studies. Sensors 2010, 10, 1589–1598. [CrossRef] [PubMed] Carminati, M.; Pedalà, L.; Bianchi, E.; Nason, F.; Dubini, G.; Cortelezzi, L.; Ferrari, G.; Sampietro, M. Capacitive detection of micrometric airborne particulate matter for solid-state personal air quality monitors. Sens. Actuators A Phys. 2014, 219, 80–87. [CrossRef] Sousan, S.; Koehler, K.; Thomas, G.; Park, J.H.; Hillman, M.; Halterman, A.; Peters, T.M. Inter-comparison of low-cost sensors for measuring the mass concentration of occupational aerosols. Aerosol Sci. Technol. 2016, 50, 462–473. [CrossRef] Jiang, R.T.; Acevedo-Bolton, V.; Cheng, K.C.; Klepeis, N.E.; Ott, W.R.; Hildemann, L.M. Determination of response of real-time SidePak AM510 monitor to secondhand smoke, other common indoor aerosols, and outdoor aerosol. J. Environ. Monit. 2011, 13, 1695–1702. [CrossRef] [PubMed] Dacunto, P.J.; Cheng, K.C.; Acevedo-Bolton, V.; Jiang, R.T.; Klepeis, N.E.; Repace, J.L.; Ott, W.R.; Hildemann, L.M. Real-time particle monitor calibration factors and PM2.5 emission factors for multiple indoor sources. Environ. Sci. Process. Impacts 2013, 15, 1511–1519. [CrossRef] [PubMed] Dacunto, P.J.; Klepeis, N.E.; Cheng, K.C.; Acevedo-Bolton, V.; Jiang, R.T.; Repace, J.L.; Ott, W.R.; Hildemann, L.M. Determining PM2.5 calibration curves for a low-cost particle monitor: Common indoor residential aerosols. Environ. Sci. Process. Impacts 2015, 17, 1959–1966. [CrossRef] [PubMed] Donateo, A.; Contini, D.; Belosi, F. Real time measurements of PM2.5 concentrations and vertical turbulent fluxes using an optical detector. Atmos. Environ. 2006, 40, 1346–1360. [CrossRef] Sioutas, C.; Kim, S.; Chang, M.C.; Terrell, L.L.; Gong, H. Field evaluation of a modified DataRAM MIE scattering monitor for real-time PM2.5 mass concentration measurements. Atmos. Environ. 2000, 34, 4829–4838. [CrossRef]

Sensors 2017, 17, 1033

12.

13.

14. 15.

16.

17.

18.

19. 20. 21. 22. 23.

24. 25. 26.

27.

28. 29. 30. 31. 32.

14 of 15

Chakrabarti, B.; Fine, P.M.; Delfino, R.; Sioutas, C. Performance evaluation of the active-flow personal DataRAM PM2.5 mass monitor (Thermo Anderson pDR-1200) designed for continuous personal exposure measurements. Atmos. Environ. 2004, 38, 3329–3340. [CrossRef] Cropper, P.M.; Hansen, J.C.; Eatough, D.J. Measurement of light scattering in an urban area with a nephelometer and PM2.5 FDMS TEOM monitor: Accounting for the effect of water. J. Air Waste Manag. Assoc. 2013, 63, 1004–1011. [CrossRef] [PubMed] Baynard, T.; Garland, R.M.; Ravishankara, A.R.; Tolbert, M.A.; Lovejoy, E.R. Key factors influencing the relative humidity dependence of aerosol light scattering. Geophys. Res. Lett. 2006, 33, 4. [CrossRef] Karagulian, F.; Belis, C.A.; Lagler, F.; Barbiere, M.; Gerboles, M. Evaluation of a portable nephelometer against the Tapered Element Oscillating Microbalance method for monitoring PM2.5. J. Environ. Monit. 2012, 14, 2145–2153. [CrossRef] [PubMed] Liu, P.F.; Zhao, C.S.; Goebel, T.; Hallbauer, E.; Nowak, A.; Ran, L.; Xu, W.Y.; Deng, Z.Z.; Ma, N.; Mildenberger, K.; et al. Hygroscopic properties of aerosol particles at high relative humidity and their diurnal variations in the North China Plain. Atmos. Chem. Phys. 2011, 11, 3479–3494. [CrossRef] Grover, B.D.; Kleinman, M.; Eatough, N.L.; Eatough, D.J.; Hopke, P.K.; Long, R.W.; Wilson, W.E.; Meyer, M.B.; Ambs, J.L. Measurement of total PM2.5 mass (nonvolatile plus semivolatile) with the Filter Dynamic Measurement System tapered element oscillating microbalance monitor. J. Geophys. Res. 2005, 110, 9. [CrossRef] Shao, W.; Zhang, H.; Zhou, H. A new particle sensor based on true RMS value measurement. In Proceedings of the IEEEInternational Instrumentation and Measurement Technology Conference (I2MTC), Taipei, Taiwan, 23–25 May 2016; pp. 1–6. Mie, G. Pioneering mathematical description of scattering by spheres. Ann. Phys. 1908, 25, 337. Mätzler, C. MATLAB functions for Mie scattering and absorption, version 2. IAP Res. Rep. 2002, 8, 1–24. Renard, J.B.; Thaury, C.; Mineau, J.-L.; Gaubicher, B. Small-Angle Light Scattering by Airborne Particulates: Environnement S.A. Continuous Particulate Monitor. Meas. Sci. Technol. 2010, 21, 085901. [CrossRef] Datesheet of S2386-44K. Available online: http://www.hamamatsu.com/jp/en/product/category/3100/ 4001/4103/S2386-44K/index.html (accessed on 5 April 2017). Laulainen, N. Summary of Conclusions and Recommendations from a Visibility Science Workshop; Technical Basis and Issues for a National Assessment for Visibility Impairment; US DOE, Pacific Northwest Laboratory: Richland, WA, USA, 1993. Li, C.; Peng, P. Visibility measurement using multi-angle forward scattering by liquid droplets. Meas. Sci. Technol. 2012, 23, 105802. [CrossRef] Peng, P.; Li, C.W. Visibility measurements using two-angle forward scattering by liquid droplets. Appl. Opt. 2016, 55, 3903–3908. [CrossRef] [PubMed] Renard, J.B.; Dulac, F.; Berthet, G.; Lurton, T.; Vignelles, D.; Jegou, F.; Tonnelier, T.; Jeannot, M.; Coute, B.; Akiki, R.; et al. LOAC: A small aerosol optical counter/sizer for ground-based and balloon measurements of the size distribution and nature of atmospheric particles—Part 1: Principle of measurements and instrument evaluation. Atmos. Meas. Tech. 2016, 9, 1721–1742. [CrossRef] Renard, J.B.; Dulac, F.; Berthet, G.; Lurton, T.; Vignelles, D.; Jegou, F.; Tonnelier, T.; Jeannot, M.; Coute, B.; Akiki, R.; et al. LOAC: A small aerosol optical counter/sizer for ground-based and balloon measurements of the size distribution and nature of atmospheric particles—Part 2: First results from balloon and unmanned aerial vehicle flights. Atmos. Meas. Tech. 2016, 9, 3673–3686. [CrossRef] Hering, S.V.; Friedlander, S. Origins of aerosol sulfur size distributions in the Los Angeles basin. Atmos. Environ. 1982, 16, 2647–2656. [CrossRef] Whitby, K.; Husar, R.; Liu, B. The aerosol size distribution of Los Angeles smog. J. Colloid Interface Sci. 1972, 39, 177–204. [CrossRef] Pope, R.M.; Fry, E.S. Absorption spectrum (380–700 nm) of pure water. 2. Integrating cavity measurements. Appl. Opt. 1997, 36, 8710–8723. [CrossRef] [PubMed] Gu, F.; Yang, J.; Wang, C.; Bian, B.; He, A. Mass concentration calculation with the pulse height distribution of aerosols and system calibration. Optik 2010, 121, 1–10. [CrossRef] Wang, X.; Chancellor, G.; Evenstad, J.; Farnsworth, J.E.; Hase, A.; Olson, G.M.; Sreenath, A.; Agarwal, J.K. A Novel Optical Instrument for Estimating Size Segregated Aerosol Mass Concentration in Real Time. Aerosol Sci. Technol. 2009, 43, 939–950. [CrossRef]

Sensors 2017, 17, 1033

33.

34.

15 of 15

Giorio, C.; Tapparo, A.; Scapellato, M.L.; Carrieri, M.; Apostoli, P.; Bartolucci, G.B. Field comparison of a personal cascade impactor sampler, an optical particle counter and CEN-EU standard methods for PM10, PM2.5 and PM1 measurement in urban environment. J. Aerosol Sci. 2013, 65, 111–120. [CrossRef] Kim, J.Y.; Magari, S.R.; Herrick, R.F.; Smith, T.J.; Christiani, D.C. Comparison of fine particle measurements from a direct-reading instrument and a gravimetric sampling method. J. Occup. Environ. Hyg. 2004, 1, 707–715. [CrossRef] [PubMed] © 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).