Selective Detection of Target Volatile Organic

sensors Article

Selective Detection of Target Volatile Organic Compounds in Contaminated Humid Air Using a Sensor Array with Principal Component Analysis Toshio Itoh *, Takafumi Akamatsu, Akihiro Tsuruta and Woosuck Shin National Institute of Advanced Industrial Science and Technology (AIST), Shimo-shidami, Moriyama-ku, Nagoya 463-8560, Japan; [email protected] (T.A.); [email protected] (A.T.); [email protected] (W.S.) * Correspondence: [email protected]; Tel.: +81-52-736-7089 Received: 16 June 2017; Accepted: 18 July 2017; Published: 19 July 2017

Abstract: We investigated selective detection of the target volatile organic compounds (VOCs) nonanal, n-decane, and acetoin for lung cancer-related VOCs, and acetone and methyl i-butyl ketone for diabetes-related VOCs, in humid air with simulated VOC contamination (total concentration: 300 µg/m3 ). We used six “grain boundary-response type” sensors, including four commercially available sensors (TGS 2600, 2610, 2610, and 2620) and two Pt, Pd, and Au-loaded SnO2 sensors (Pt, Pd, Au/SnO2 ), and two “bulk-response type” sensors, including Zr-doped CeO2 (CeZr10), i.e., eight sensors in total. We then analyzed their sensor signals using principal component analysis (PCA). Although the six “grain boundary-response type” sensors were found to be insufficient for selective detection of the target gases in humid air, the addition of two “bulk-response type” sensors improved the selectivity, even with simulated VOC contamination. To further improve the discrimination, we selected appropriate sensors from the eight sensors based on the PCA results. The selectivity to each target gas was maintained and was not affected by contamination. Keywords: metal oxide; tin oxide; cerium oxide; principal component analysis; indoor air contamination

1. Introduction Human breath includes many volatile organic compounds (VOCs) that can be used as biomarkers for diseases. For example, breath exhaled from diabetes patients includes high concentrations of acetone and methyl i-butyl ketone (MiBK) [1], and breath exhaled from lung cancer patients includes higher concentrations of n-decane, nonanal, and acetoin than controls [2–4]. Therefore, breath-monitoring methods are desirable as diagnostic tools because they are fast and non-invasive diagnostic methods [5]. Consequently, many researchers have attempted to develop breath-monitoring systems [6–8]. One possible breath-monitoring system uses semiconductor metal oxide (MOx ) VOC sensors that were developed for a wide range of applications, including indoor air quality monitoring [9–15], mouth odor monitoring [16–19], and human health diagnosis [20–24]. In all cases, the selectivity of the gas sensor should be analyzed precisely due to the presence of potentially interfering gases. The selectivity of MOx sensors can be roughly controlled by adding noble metal catalysts [25]. Alternatively, a sensor array can be analyzed with statistical methods, such as principal component analysis (PCA) [20,21,26,27]. In one study, a screening apparatus was developed that included a sensor array with a solid phase microextraction (SPME) fiber as a VOC-condensing unit [20,21,28,29]. Some VOCs related to several diseases have been detected in exhaled breath at the ppb level [3], which is typically too low for detection by MOx sensors. The SPME fiber has the advantage of condensing VOCs from breath samples into gas bags to directly expose them to the sensor array. Byun et al. have reported a screening apparatus using a sensor array of four TGS series sensors (TGS 2600, 2610, 2610, and 2620; Figaro Engineering Sensors 2017, 17, 1662; doi:10.3390/s17071662

www.mdpi.com/journal/sensors

Sensors 2017, 17, 1662

2 of 14

Inc., Minoh, Japan) and a custom-built MOx sensor with an automated SPME desorption system for selectivity to lung cancer-related VOCs [20,21]. However, the sensor signals for target VOCs are also affected by the contamination of VOCs from ambient indoor air. To overcome this problem, the indoor air could be purified and maintained in the sensor chamber during the desorption of VOCs from the SPME fiber. In this case, however, the screening apparatus needs to be equipped with a complicated system, such as a delicate purification system for indoor air, a flow system for purified air into the sensor chamber, and an airtight injection joint for the SPME fiber. Therefore, we have investigated reducing the influence of indoor air contamination by selecting appropriate sensors from the sensor array based on statistical analysis. In this study, we investigated the effects of contamination on the monitoring of lung cancer- and diabetes-related VOCs under standard humid conditions (60% relative humidity (RH)). The European Commission Joint Research Centre Environmental Institute and Ministry of Health, Labour and Welfare, Japan, have published lists of VOCs that should be monitored for protecting human health [30,31]. We prepared mixtures of 31 VOCs to simulate possible indoor air contamination [15] according to the lists and definitions of VOCs from an international standard organization [32]. In our previous study, to simulate the diagnostic process, we used four TGS sensors defined by Byun et al. [20,21], and two Pt, Pd, and Au-loaded SnO2 (Pt, Pd, Au/SnO2 ) sensors in dry air [27]. The two Pt, Pd, Au/SnO2 sensors were developed for the detection of low concentrations of VOCs [25,33], including nonanal [24]. These six sensors are classified as “grain boundary-response type” MOx sensors, e.g., SnO2 . At low VOC concentrations, oxygen adsorbed from air on the surface removes electrons from the MOx conduction band and results in an electron-depleted layer, which acts as a potential barrier between neighboring grains [34,35]. At high VOC concentrations, the VOCs are oxidized by adsorbed oxygen, reducing the potential barrier, so that the concentration of VOCs can be analyzed by monitoring the electrical resistance. However, the described sensor array including the six sensors cannot discriminate lung cancer- and diabetes-related VOCs under humid conditions, even using PCA. Adsorbed water molecules block oxygen from adsorbing onto surface sites and decrease the intensity of the resistance change induced by oxidized VOCs. In this study, we used both “bulk-response type” sensors as well as “grain boundary-response type” sensors. The “bulk-response type” sensors, such as CeO2 and Ce1−x Zrx O2 [36,37], possess high oxygen diffusion coefficients to provide lattice oxygen for oxidizing VOCs. Therefore, because the “bulk-response type” sensors would likely be less affected by humidity, we added them into the sensor array to improve the performance. In addition, we investigated the use of eight sensors for monitoring lung cancer-related VOCs (n-decane, nonanal, and acetoin) and diabetes-related VOCs (acetone and MiBK) with simulated indoor air at a standard relative humidity. 2. Materials and Methods 2.1. Gas Sensors We selected four commercially available semiconductor metal oxide sensors (TGS 2600, 2602, 2610, and 2620; Figaro Engineering Inc., Minoh, Japan), two semiconductor Pt, Pd, Au/SnO2 sensors, and two semiconductor Zr-doped CeO2 sensors. The Pt, Pd, Au/SnO2 sensors and the Zr-doped CeO2 sensors were prepared according to previous reports [24,38]; photos of all sensors are shown in Figure 1, and the details of the sensors are summarized in Table 1. Pt, Pd, and Au aqueous colloid suspensions (particle size: 2–4 nm; Tanaka Kikinzoku Kogyo, Tokyo, Japan) were added to SnO2 powder (particle size: around 100 nm; Aldrich, St. Louis, MO, USA). The Pt, Pd, and Au content was 1 wt % of SnO2 . The mixture was stirred, dried, and subsequently heated at 400 ◦ C for 2 h in air. The resulting powder was added to an ethylcellulose-type organic dispersant to obtain a paste with powder/vehicle ratios of 1/16 and 1/8. The pastes were subsequently applied to surface-oxidized silicon substrates with platinum comb-type electrodes using a Musashi Engineering FAD-320s dispenser. The substrates were dried at 80 ◦ C for 2 h and subsequently annealed

Sensors 2017, 17, 1662

3 of 14

at 500 ◦ C for 2 h in ambient air. The two Pt, Pd, Au/SnO2 sensors had a thickness of 2.8 and 4.6 µm for Sensors 2017, 17, 1662 3 of 14 the 1/16 and 1/8 powder/vehicle ratio pastes, and were referred to as “#33b” and “#31b,” respectively. Cerium nitrate pentahydrate (Kojundo Laboratory, Japan) zirconyl thickness of 2.8 and 4.6 μm for the 1/16 andChemical 1/8 powder/vehicle ratioSakado, pastes, and wereand referred to asnitrate dihydrate (Wako Pure Chemical Industries, Osaka, Japan) were dissolved in distilled water to obtain “#33b” and “#31b,” respectively. 3+ and 10 mmol/L Zr4+ , respectively. Aqueous ammonia (25%) was solutions containing 90 mmol/L Ce Cerium nitrate pentahydrate (Kojundo Chemical Laboratory, Sakado, Japan) and zirconyl nitrate dropwise dihydrate (Wako Chemical Industries, Osaka, Japan) were dissolved distilled water then added to the Pure aqueous solution to form a white precipitate. The in precipitate wasto filtered, obtainwith solutions containing 90 mmol/L Ce3+ and 10 mmol/LChemicals, Zr4+, respectively. ammonia and mixed commercialized carbon powder (Mitsubishi Tokyo,Aqueous Japan) using a Keyence (25%) was then added dropwise to the aqueous solution to form a white precipitate. The precipitate HM-500 hybrid mixer to give a precipitate/carbon powder weight ratio of 75:11. The mixture was then was filtered, and mixed with commercialized carbon powder (Mitsubishi Chemicals, Tokyo, Japan) dried at 70 ◦ C overnight and annealed at 900 ◦ C for 2 h to give the final product Ce0.9 Zr0.1 O2 (CeZr10) using a Keyence HM-500 hybrid mixer to give a precipitate/carbon powder weight ratio of 75:11. The as a fine powder. The powder was added an ethylcellulose-type dispersant to obtain mixture was then dried at 70 °C then overnight andtoannealed at 900 °C for 2 h organic to give the final product a paste aOpowder/vehicle of 1/4. Cewith 0.9Zr0.1 2 (CeZr10) as a fineratio powder. The powder was then added to an ethylcellulose-type organic The CeZr10topaste screen-printed onto theratio electrode dispersant obtainwas a paste with a powder/vehicle of 1/4. of alumina substrates with a platinum ◦ C for 15 min, CeZr10using paste awas screen-printed ontoKogyo the electrode alumina substrates a platinum comb-typeThe electrode New Long Seimitsu LS-150ofscreen printer, driedwith at 150 ◦ comb-type electrode using a New Long Seimitsu Kogyo LS-150 screen printer, dried at 150 °C and then subjected to four print/dry cycles. The resulting substrate was calcined at 500 Cfor for155 h and min, and then subjected to four print/dry cycles. The resulting substrate was calcined at 500 °C for 5 ◦ fired at 1100 C for 2 h to obtain the CeZr10 thick film sensor, referred to as “No. 9”. An additional h and fired at 1100 °C for 2 h to obtain the CeZr10 thick film sensor, referred to as “No. 9”. An Zr-doped CeO2 sensor was prepared in a similar manner. Powder of γ-alumina prepared as described additional Zr-doped CeO2 sensor was prepared in a similar manner. Powder of γ-alumina prepared in ouras previous report [39] was also added to an ethylcellulose-type organic dispersant to obtain a paste described in our previous report [39] was also added to an ethylcellulose-type organic dispersant with atopowder/vehicle of 1/2. Theratio γ-alumina paste was screen-printed onto the onto CeZr10 obtain a paste withratio a powder/vehicle of 1/2. The γ-alumina paste was screen-printed the thick film, and the thick print/dry process was repeated in athree similar manner to themanner CeZr10 CeZr10 film, and the print/dry processthree was times repeated times in a similar tothick the film. CeZr10 thick Subsequently, thescreen-printed CeZr10 paste was screen-printed onto the γ-alumina thick film, four Subsequently, thefilm. CeZr10 paste was onto the γ-alumina thick film, underwent ◦ C calcined underwent cycles,atand °C for 2suspension h. A Pt colloid suspension print/dry cycles,four andprint/dry was calcined 800was for 2 h. at A 800 Pt colloid (particle size:(particle 2 nm, Tanaka size: 2 nm, Tanaka Kikinzoku Kogyo, Tokyo, Japan) was added onto the surface of the resulting Kikinzoku Kogyo, Tokyo, Japan) was added onto the surface of the resulting CeZr10/Al 2 O3 /CeZr10 CeZr10/Al2O3/CeZr10 layered thick film and dried at 70 °C to obtain a Pt content of 3 wt % on the layered thick film and dried at 70 ◦ C to obtain a Pt content of 3 wt % on the upper CeZr10 thick film. upper CeZr10 thick film. This Zr-doped CeO2 sensor (CeZr10/Al2O3/Pt-CeZr10) was referred to as This Zr-doped CeO2 sensor (CeZr10/Al2 O3 /Pt-CeZr10) was referred to as “No. 71”. “No. 71”.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

Figure 1. Photos of all sensors: (a) TGS 2600; (b) TGS 2602; (c) TGS 2610; (d) TGS 2620; (e) #31b; (f)

Figure 1. Photos of all sensors: (a) TGS 2600; (b) TGS 2602; (c) TGS 2610; (d) TGS 2620; (e) #31b; (f) #33b; #33b; (g) No. 9; and (h) No. 71. (g) No. 9; and (h) No. 71. Table 1. Details of the four TGS, two Pt, Pd, Au/SnO2, and two Zr-doped CeO2 sensors used in this study.

Table 1. Details of the four TGS, two Pt, Pd, Au/SnO2 , and two Zr-doped CeO2 sensors used in this study. Name of Sensor Detected Compounds and Sensor Details TGS 2600 H2 and alcohol (manufactured by Figaro) Name of Sensor Compounds andbySensor TGS 2602 Alcohol, ammonia, Detected VOC, and H 2S (manufactured Figaro) Details TGS 2610 Liquefied petroleum gas (manufactured by Figaro) TGS 2600 H2 and alcohol (manufactured by Figaro) TGS 2620 Alcohol and solvent vapors (manufactured by Figaro) TGS 2602 #31b Alcohol, ammonia, VOC, and H2 S (manufactured by Figaro) 4.6 μm) VOCs (Pt, Pd, Au/SnO2; film thickness: TGS 2610 #33b Liquefied petroleum gas (manufactured by Figaro) 2; film thickness: 2.8 μm) VOCs (Pt, Pd, Au/SnO TGS 2620 No. 9 Alcohol andand solvent vapors (manufactured by Figaro) Oxygen VOCs (CeZr10; film thickness: 9 μm) #31b No. 71 VOCs (Pt, Pd, 4.6film µm)thickness: 9 μm/4 μm/9 μm) 2 ; film thickness: 2O3/Pt-CeZr10; Oxygen andAu/SnO VOCs (CeZr10/Al

#33b No. 9 No. 71

VOCs (Pt, Pd, Au/SnO2 ; film thickness: 2.8 µm) Oxygen and VOCs (CeZr10; film thickness: 9 µm) Oxygen and VOCs (CeZr10/Al2 O3 /Pt-CeZr10; film thickness: 9 µm/4 µm/9 µm)

Sensors 2017, 17, 1662

4 of 14

2.2. Preparation of Target VOCs The molecular structures of the target VOCs are shown in Figure 2. The target gases of nonanal (Tokyo chemical industry, Tokyo, Japan), n-decane (Tokyo chemical industry, Tokyo, Japan), and acetoin (Tokyo chemical industry, Tokyo, Japan) were generated from their solvents by a Gastec Permeator PD-1B gas generator. Target gases of acetone and MiBK were prepared using cylinders of standard gases (50 ppm in nitrogen; Sumitomo Seika Chemicals, Osaka, Japan). The simulated indoor air contaminants were also prepared using a cylinder of 31 different VOCs mixed in nitrogen (Sumitomo Seika Chemicals, Osaka, Japan), as shown in Table 2. Table 2. Concentrations and functional group classifications of each component in the mixture of 31 kinds of volatile organic compounds (VOCs) in nitrogen. Group

Aromatic Hydrocarbon

Aliphatic Hydrocarbon

Terpene

Component

Concentration in Cylinder [ppm]

Benzene

0.52

Toluene

0.50

o-Xylene

0.50

m-Xylene

0.50

Styrene

0.50

Ethylbenzene

0.50

n-Propylbenzene

0.49

1,2,3-Trimethylbenzene

0.50

1,2,4-Trimethylbenzene

0.50

1,3,5-Trimethylbenzene

0.50

o-Ethyltoluene

0.50

n-Hexane

0.50

2-Methylpentane

0.50

3-Methylpentane

0.50

n-Heptane

0.50

2,4-Dimethylpentane

0.51

n-Octane

0.50

2,2,4-Trimethylpentane

0.50

n-Nonane

0.50

n-Decane

0.50

n-Undecane

0.50

n-Dodecane

0.100

Methylcyclopentane

0.50

Cyclohexane

0.50

Methylcyclohexane

0.50

α-Pinene

0.50

β-Pinene

0.51

(+)-Limonene

0.50

Halide

p-Dichlorobenzene

0.50

Ester

Butyl acetate

0.99

Ketone

Methyl i-butyl ketone

2.99

Total

18.11 * * 18.11 [ppm] = 8.92 × 104 [µg/m3 ].

(Tokyo chemical industry, Tokyo, Japan), n-decane (Tokyo chemical industry, Tokyo, Japan), and acetoin (Tokyo chemical industry, Tokyo, Japan) were generated from their solvents by a Gastec Permeator PD-1B gas generator. Target gases of acetone and MiBK were prepared using cylinders of standard gases (50 ppm in nitrogen; Sumitomo Seika Chemicals, Osaka, Japan). The simulated indoor air2017, contaminants were also prepared using a cylinder of 31 different VOCs mixed in nitrogen Sensors 17, 1662 5 of 14 (Sumitomo Seika Chemicals, Osaka, Japan), as shown in Table 2. (a)

O H

(b)

(c)

(d) O

(e) O

O OH

Figure 2. Structural formulas of the target gases: (a) nonanal; (b) n-decane; (c) acetoin (3-hydroxy-2-

Figure 2. Structural formulas of the target gases: (a) nonanal; (b) n-decane; (c) acetoin (3-hydroxy-2butanone); (d) acetone; (e) methyl i-butyl ketone (MiBK; 2-butanone). Sensorsbutanone); 2017, 17, 1662 5 of 14 (d) acetone; (e) methyl i-butyl ketone (MiBK; 2-butanone). Table 2. Concentrations and functional group classifications of each component in the mixture of 31

2.3. Sensor 2.3. Gas Gas kinds SensorofAnalysis Analysis volatile organic compounds (VOCs) in nitrogen.

The using in apparatus, as Group were Component Concentration Cylinder [ppm] The gas gas concentrations concentrations were measured measured using sensors sensors in aa flow flow in apparatus, as shown shown in in Figure Figure 3. 3. Benzene 0.52 The target gas was selected from nonanal, n-decane, acetoin, acetone, and MiBK. Depending the The target gas was selected from nonanal, n-decane, acetoin, acetone, and MiBK. Depending on on the Toluene 0.50 target ororacetoin the gas generator, or or 50 50 ppm of target gas, gas, 2.5 2.5 ppm ppm of ofnonanal, nonanal,n-decane, n-decane, acetoinininnitrogen nitrogenfrom from the gas generator, ppm o-Xylene 0.50 acetone or MiBK in nitrogen from the cylinder, was flowed at 200 or 10 mL/min, respectively. To of acetone or MiBK in nitrogen from the cylinder, was flowed at 0.50 200 or 10 mL/min, respectively. m-Xylene generate humid air, air, 200 200 mL/min of nitrogen and 100 mL/min of oxygen were introduced into a water Styrene 0.50 To generate humid mL/min of nitrogen and 100 mL/min of oxygen were introduced into Aromatic Ethylbenzene 0.50 bubbler to maintain a RH of 60%. To simulate contamination from indoor air, the mixture of 31 VOCs a water bubbler to maintain Hydrocarbona RH of 60%. To simulate contamination from indoor air, the mixture n-Propylbenzene 0.49 3 in nitrogen flowed was at 1.7flowed mL/min to maintain concentration of 300 μg/m3, which is the of 31 VOCs was in nitrogen at 1.7 mL/min atototal maintain a total 1,2,3-Trimethylbenzene 0.50 concentration of 300 µg/m , maximum concentration of total VOCs in indoor by the Federal Office for 1,2,4-Trimethylbenzene 0.50 air established which is theallowed maximum allowed concentration of total VOCsair in established indoor by the Federal 1,3,5-Trimethylbenzene 0.50 Environment in Germany. Additional dry nitrogen was flowed to adjust the total flow rate to 500 Office for Environment in Germany. Additional dry nitrogen was flowed to adjust the total flow rate o-Ethyltoluene 0.50 were 1 ppm, the N2/O2 ratio mL/min. Under Under these conditions, the concentrations of theoftarget gases to 500 mL/min. these conditions, the concentrations the target gases were 1 ppm, the N2 /O2 n-Hexane 0.50 was maintained at 4, and the RH was 0% or 60%. ratio was maintained at 4, and the RH2-Methylpentane was 0% or 60%. 0.50

Resistance

3-Methylpentane n-Heptane Target Gas 2,4-Dimethylpentane Selecting one target gas from n-Octane • NonanalAliphatic 2,2,4-Trimethylpentane Gas generator n-Nonane • n-Decane Hydrocarbon • Acetoin n-Decane • Acetone n-Undecane Cylinder • MiBK n-Dodecane Methylcyclopentane N2 Water Cyclohexane bubbler Methylcyclohexane O2 α-Pinene Terpene β-Pinene Contaminations (+)-Limonene 31 kinds of VOC Mixed gas Halide p-Dichlorobenzene • Flow rate:Ester 500 mL/min Butyl acetate • N2:O2 = Ketone 4:1 Methyl i-butyl ketone • Target gas: 1 ppm Total 

0.50 0.50 Sensor Chamber 0.51 0.50 0.50 0.50 0.50 0.50 Resistance [] 0.100 ࡾࢇ 0.50 ࡾࢇ 0.50 0.50 0.50 ࡾ ࡾࢍ ࢍ 0.51 0.50 Time 0.50 0.99 2.99 ࡿ ൌ Sensor response 18.11 *

Flow period Target gas Contaminations

ࡾࢇ ࡾࢍ

• Humidity: 60%RH * 18.11 [ppm] = 8.92 × 104 [μg/m3]. PCA • Contaminations: 0 or 300 g/m3

Figure of sensor sensor response. response. Figure 3. 3. Schematic Schematic diagram diagram of of the the flow flow apparatus apparatus for for the the measurement measurement of

All gas sensors were placed in a sensor chamber. The operating voltage of the heater of the four All gas sensors were placed in a sensor chamber. The operating voltage of the heater of the four commercially available TGS sensors was 5 V, as recommended by the manufacturer, and the commercially available TGS sensors was 5 V, as recommended by the manufacturer, and the operating operating temperature of the two Pt, Pd, Au/SnO2 sensors (#33b and #31b) and two Zr-doped CeO2 temperature of the two Pt, Pd, Au/SnO2 sensors (#33b and #31b) and two Zr-doped CeO2 sensors sensors (No. 9 and No. 71) were maintained at 250 °C and 400 °C, respectively. The electrical (No. 9 and No. 71) were maintained at 250 ◦ C and 400 ◦ C, respectively. The electrical resistance of the resistance of the eight sensors was measured by a Graphtec Midi Logger GL200 and a Keithley 2700 eight sensors was measured by a Graphtec Midi Logger GL200 and a Keithley 2700 digital multimeter digital multimeter for the TGS sensors and the Pt, Pd, Au/SnO2, and for the Zr-doped CeO2 sensors, respectively. The sensor response (r) is defined by Equation 1, r

Ra Rg

(1)

where Ra is the resistance in pure air or in simulated contaminated indoor air, and Rg is the resistance in pure air or in simulated contaminated indoor air with 1 ppm of the target gases.

Sensors 2017, 17, 1662

6 of 14

for the TGS sensors and the Pt, Pd, Au/SnO2 , and for the Zr-doped CeO2 sensors, respectively. The sensor response (r) is defined by Equation (1), r=

Ra Rg

(1)

where Ra is the resistance in pure air or in simulated contaminated indoor air, and Rg is the resistance in pure air or in simulated contaminated indoor air with 1 ppm of the target gases. To select appropriate sensors from the sensor array, the difference in sensor responses (D) between humid air with and without contamination is also defined in Equation (2), D=

r p − rs × 100 [%] rs

(2)

where rs and rp are the sensor responses in humid air with and without contamination, respectively. 2.4. Principal Component Analysis The PCA was carried out using the r values. First, normalized scores (xti ) were calculated according to Equation (3), (r − r t ) xti = ti (3) σt where t is the sensor index, i is the sensor response analysis index, rti is the sensor response analysis i of sensor t, rt is the average sensor response of sensor t, and σt is the standard deviation of sensor t. Correlation coefficients (cts ) were calculated according to Equation (4), cts =

Sts Stt Sss

(4)

where s is a sensor index and Sts is a sum of products of s and t (Equation (5)), imax

Sts =

∑ (rti − rt )(rsi − rs )

(5)

i =1

and Sss and Stt are the sum of the squares of s and t (Equation (6)). imax

Stt =

∑

Rsi − Rs


2

(6)

i =1

Eigenvalues (λ) were calculated according to Equation (7),

|A − λE| = 0

(7)

where A is the matrix shown in Equation (8), and E is the unit matrix.    A=  

c11 c21 .. . cm1

c12 c22 .. . cm2

··· ··· .. . ···

c1m c2m .. . cmm

     

(8)

c11 = c22 = . . . = c mm = 1, cts = cst , and m is the maximum number of sensors. Eigenvalues are then obtained for each sensor index (Equation (9)). λ = λ1 , λ2 , · · · , λ m ( λ1 > λ2 > · · · > λ m )

(9)

Sensors 2017, 17, 1662

7 of 14

Eigenvectors of number j (vj ) satisfy Equation (10).    A − λ j E vj = 0 (vj =   


a1j a2j .. . amj

   )  

(10)

Finally, PCA scores (Zji ) were evaluated according to Equation (11). Zji = a1j x1i + a2j x2i + . . . + amj xmi

(11)

when PCA scores in principal components 1 and 2 are plotted, the coordinates of the PCA scores are (Z1i , Z2i ). In this study, we carried out the PCA at six different combinations of sensor responses, relative humidity, and indoor air contamination, as summarized in Table 3. The method for selecting gas sensors for each type of analysis is described in Sections 3.1 and 3.2. Table 3. Selection of sensor responses in the six types of analysis. Sensors

Type I Type II Type III Type IV Type V Type VI

TGS 2600

TGS 2602

TGS 2610

TGS 2620

#31b

#33b

No. 9

No. 71

Humidity [%RH]

Contamination [µg/m3 ]

Results

v v v v v v

v v v v v -

v v v v v -

v v v v v v

v v v v v -

v v v v v v

v v v v

v v v v

0 60 0 60 60 60

0 0 0 0 0, 300 0, 300

Figure 6a Figure 6b Figure 6c Figure 6d Figure 7 Figure 9

* “v” means that the sensor responses (r) from the sensor were used.

3. Results and Discussion 3.1. Sensor Responses and PCA in Pure Dry and Humid Air The sensor responses (S) of all eight sensors at 1 ppm of the target gases were collected under pure dry and pure humid air (i.e., dry and humid air without simulated contaminated indoor air). Figure 4 shows the dynamic resistance response of the six “grain boundary-response type” (TGS2600, 2602, 2610, 2620, #31b, and #33b) and two “bulk-response type” (Nos. 9 and 71) sensors at 1 ppm of acetone in pure dry and pure humid air. Semiconductor-type gas sensors decrease in resistance in response to VOCs. In the “grain boundary-response type” sensors, oxidation of VOCs consumes oxygen adsorbed on the surface of sensor materials, decreasing the resistance of the grain boundary. In contrast, the “bulk-response type” sensors provide lattice oxygen to oxidize VOCs, and the generated surface oxygen vacancies can diffuse rapidly into the bulk because of the high diffusion coefficient for oxygen [37], decreasing the resistance of the bulk. The base resistance of the “grain boundary-response type” sensors in pure humid air was lower than that in pure dry air except for TGS2602, whereas the “bulk-response type” sensors showed almost the same base resistance in pure dry and humid air. For TGS2602, the intensity of the resistance response to acetone decreased at 60% RH because the adsorption of oxygen is prevented by the adsorption of water. Therefore, the “bulk-response type” sensors were hardly affected by the humidity, because the resistances were controlled by a different mechanism.

ry-response type” sensors in pure humid air was lower than that in pure dry air except for 2, whereas the “bulk-response type” sensors showed almost the same base resistance in pure d humid air. For TGS2602, the intensity of the resistance response to acetone decreased at 60% ause the adsorption of oxygen is prevented by the adsorption of water. Therefore, the “bulke type” sensors were hardly affected Sensors 2017, 17, 1662by the humidity, because the resistances were controlled ferent mechanism. Sensors 2017, 17, 1662 8

10

107

TGS2600 TGS2602 TGS2610 TGS2620 #31b #33b No.9 No.71

7

10 Resistance []

Resistance []

8 of 14

8

10

6

10

105

104

103 0

8 of 14

6

10

105

104

(a) 10

20

30 40 Time [min]

50

60

70

103 0

(b) 10

20

30 40 Time [min]

50

60

70

Figure 4. Dynamic response the six “grain boundary-response type” and two “bulkFigure 4. Dynamic resistance responseresistance of the six “grainofboundary-response type” and two response to 1in ppm acetone in (a)(b) pure dry and (b) pure humid air. “bulk-response type” sensors to 1type” ppm sensors of acetone (a) of pure dry and pure humid air.

Figure 5 shows the average responses of all sensors to 1 ppm of each target gas. The average

Figure 5 shows the average responses of all sensors to 1 ppm of each target gas. The average sensor responses of the six “grain boundary-response type” sensors in pure dry air have been sensor responses of thereported six “grain boundary-response type” sensors in pure dry air have been reported previously [27], and the four commercial TGS sensors exhibit their characteristic sensor previously [27], and the four commercial TGS exhibit their characteristic sensor patterns in patterns in pure dry air. In sensors the Pt, Pd, Au/SnO 2 sensors, aldehydes were partially oxidized to acids at pure dry air. In the Pt, the Pd, working Au/SnOtemperature sensors, aldehydes were partially to acids atPt, thePd, working 2 (250 °C) [40]. Therefore,oxidized the two prepared Au/SnO2 sensors (#31b, ◦ temperature (250 C) [40]. the responses two prepared Pt, Pd, Au/SnO (#31b, #33b) show (acetoin) and 2 sensors groups, #33b) Therefore, show strong to easily oxidized functional i.e., alcohols strong responses to easily oxidized functional groups, i.e., alcohols (acetoin) and aldehydes (nonanal), aldehydes (nonanal), and low responses to difficult to oxidize functional groups, i.e., ketones (acetone and low responses to difficult to oxidize functional groups, i.e.,ofketones MiBK). Moreover, and MiBK). Moreover, the thicker film Pt, Pd, (acetone Au/SnO2and (#31b; 4.6 μm) shows stronger sensor to all2target specifically as sensor compared to a thinner film (#33b; 2.8 μm), which the thicker film of Pt,responses Pd, Au/SnO (#31b;gases, 4.6 µm) shows decane, stronger responses to all target we have previouslyto reported [27]. Although the µm), two Pt, Pd, Au/SnO 2 sensors (#31b and #33b) were gases, specifically decane, as compared a thinner film (#33b; 2.8 which we have previously both 250Au/SnO °C, analyzable gas(#31b flowed quickly 500 both mL/min in the reported [27]. Although theheated two Pt,atPd, and #33b)at were heated at sensor 250 ◦ C,chamber, cooling 2 sensors the sensors. Therefore, the thinner film (#33b) would be cooled below the appropriate temperature. analyzable gas flowed quickly at 500 mL/min in the sensor chamber, cooling the sensors. Therefore, thicker (#31b) alsothe shows a strong response to decane, but a weak the thinner film (#33b)The would be film cooled below appropriate temperature. The thicker filmresponse (#31b) to acetone and MiBK, even though all of these VOCs do not have easily oxidized functional Because the also shows a strong response to decane, but a weak response to acetone and MiBK, even though allgroups. of molecular weight of decane is larger than that of acetone and MiBK, the decomposition rate of decane these VOCs do not have easily oxidized functional groups. Because the molecular weight of decane would be expected to be greater than that of acetone and MiBK. is larger than that of acetone and MiBK, the decomposition rate of decane would be expected to be Compared to the “grain boundary-response type” sensors, the two “bulk-response type” sensors greater than that of acetone and MiBK. reveal different characteristic sensor patterns. The CeZr10 single layered film (No. 9) exhibits greater Compared to the “grain boundary-response type” sensors, the two “bulk-response type” sensors sensor responses to oxygen-containing hydrocarbons (acetoin, nonanal, MiBK, and acetone) than to reveal different characteristic sensor patterns. The CeZr10 single film (No.2O 9)3/Pt-CeZr10 exhibits greater unoxidized hydrocarbons (decane). In the caselayered of the CeZr10/Al multi-layered film (No. sensor responses to oxygen-containing hydrocarbons (acetoin, nonanal, MiBK, and acetone) than tobarely decreased 71), the response to all target gases was small. However, the response to decane unoxidized hydrocarbons In the caseupper of thelayer CeZr10/Al multi-layered film 2 O3 /Pt-CeZr10 with (decane). the addition of the (Pt-CeZr10), although the responses to oxygen-containing (No. 71), the response to all target gases was small. However, theCeZr10/Al response2O to3/Pt-CeZr10 decane barely decreased hydrocarbons decreased drastically. In the system, the lower CeZr10 layer with the addition of works the upper layer (Pt-CeZr10), although the responses to oxygen-containing as a gas-sensing layer because of contact with the platinum comb-type electrode. The upper hydrocarbons decreased drastically. theinsulated CeZr10/Al system, thethelower Pt-CeZr10 layer isIn then from lower CeZr10 layer by Al2O3CeZr10 layer in the middle. The 2 Othe 3 /Pt-CeZr10 upper Pt-CeZr10 layer acts as the catalytic layer. Therefore, the Pt-CeZr10 and the pure CeZr10 of the layer works as a gas-sensing layer because of contact with the platinum comb-type electrode. The type” materials canCeZr10 easily oxidize oxygen-containing upper Pt-CeZr10 layer“bulk-response is then insulated from the lower layer by the Al2 O3 layer hydrocarbons, in the middle.consuming them on the upper Pt-CeZr10 the CeZr10/Al 2O3/Pt-CeZr10 system. In contrast, The upper Pt-CeZr10 layer acts as the catalytic layer layer. of Therefore, the Pt-CeZr10 and the pure CeZr10 of decane would hardly be oxidized on the upper Pt-CeZr10 and, thus, would reach the lower CeZr10 the “bulk-response type” materials can easily oxidize oxygen-containing hydrocarbons, consuming layer. In layer pure of humid air, the characteristic sensor patterns of the “grain boundary-response type” them on the upper Pt-CeZr10 the CeZr10/Al 2 O3 /Pt-CeZr10 system. In contrast, decane would changed drastically. For the four commercial TGS sensors, the response to ketones for TGS hardly be oxidized on sensors the upper Pt-CeZr10 and, thus, would reach the lower CeZr10 layer. 2602 decreased drastically, and the other TGS sensors exhibited almost flat response patterns. For the In pure humid air, the characteristic sensor patterns of the “grain boundary-response type” two Pt, Pd, Au/SnO2 sensors, the characteristic strong responses to acetoin and decane from #31b sensors changed drastically. For the four commercial TGS sensors, the response to ketones for TGS decreased to almost the same response as #33b. However, the sensor response patterns of the “bulk2602 decreased drastically, and the other TGS sensors exhibited almost flat response patterns. For response type” sensors hardly changed in humid air. the two Pt, Pd, Au/SnO2 sensors, the characteristic strong responses to acetoin and decane from

Sensors 2017, 17, 1662

9 of 14

#31b decreased to almost the same response as #33b. However, the sensor response patterns of the “bulk-response type” sensors hardly changed in humid air. Sensors Sensors2017, 2017,17, 17,1662 1662

9 of 9 14of 14

(a) (b) (a) (b) Figure5. 5. Average sensor responses of the six “grain boundary-response and “bulk-response two “bulkFigure Average sensor responses of the six “grain boundary-response type”type” and two Figure 5. type” Average sensor of gases the six “grain type” air. and two “bulkresponse sensors to 1 responses ppm of target in (a) pure boundary-response dry air and (b) pure humid type” sensors to 1 ppm of target gases in (a) pure dry air and (b) pure humid air. response type” sensors to 1 ppm of target gases in (a) pure dry air and (b) pure humid air.

Figure 6 shows the PCA scores and eigenvectors using six (“grain boundary-response type”) or Figure 66 shows the and eigenvectors using six type”) Figure shows the PCA PCA scores scores and eigenvectors using six (“grain (“grain boundary-response type”) or eight (“grain boundary-response type” and “bulk-response type”) sensorsboundary-response in pure dry and humid air. or eight (“grain boundary-response type” and “bulk-response type”) sensors in pure eight (“grain boundary-response type” and “bulk-response type”) sensors dry and humid air. All cumulative variances of principal components (PCs) 1 and 2 in Figure 6 are over 80%. To retain All cumulative principal (PCs) 1 and 2 in Figure are 80%. To retain the originality variances of the data of [41], the first components two PCs are normally sufficient in this 6study. In Figure All cumulative variances of principal components are over over 80%. 6a, Tothe retain the originality the data[41], [41], the first twoPCs PCs are normally sufficient in In Figure 6a, PCA uses sensor responses from the sixtwo “grain boundary-response type”in sensors instudy. pure air; 6a, the the the of the data the first are normally sufficient thisthis study. Indry Figure the PCA uses sensor from the six “grain boundary-response type” sensors in sensors pure sensor response dataresponses are from the same PCA that uses the sensor responses from the in air; PCA uses sensor responses from the six “grain boundary-response type” sensors in six pure drydry air; the pure and contaminated dry airthe [27]. In dry conditions, discriminating between all target gases is in the sensor response from the same PCA that uses the sensor responses from sensors sensor response datadata are are from same PCA that uses the sensor responses from thethe sixsix sensors possible. However, in humid conditions, the scores of decane overlapped those MiBKgases and is in pure and contaminated dryair air[27]. [27].InIndry dryPCA conditions, discriminating between alloftarget pure and contaminated dry conditions, discriminating between all is were close to those of acetone. Therefore, discrimination between all gases is not possible, as shown possible. However, in humid conditions, the PCA scores of decane overlapped those of MiBK and possible. in humid conditions, PCA scores of decane overlapped those of MiBK and in Figure 6b, because the response patterns from the “grain boundary-response type” sensors to in were close those ofofacetone. Therefore, discrimination between allall gases is not possible, as shown were closeto to those acetone. Therefore, discrimination between gases is not possible, as shown decane became similar to MiBK and acetone, as shown in Figure 5b. Figures 6c and 5d show the Figure 6b, because the response patterns from the “grain boundary-response type” sensors to decane in Figure 6b, because the response patterns from the “grain boundary-response type” sensors to resultssimilar of the PCA for the “grain boundary-response type” and “bulk-response type”the sensors. The became to MiBK and acetone, as shown in Figure 5b. Figures 6c and 5d show results of the decane became similar to MiBK and acetone, as shown in Figure 5b. Figures 6c and 5d show the addition sensorboundary-response responses from the “bulk-response type” sensors can sensors. improve The the selectivity all PCA forof theof “grain type” and “bulk-response addition oftosensor results the PCA for the “grain boundary-response type” andtype” “bulk-response type” sensors. The target gases. In humid conditions, Figure 6d shows that the PCA scores of decane separated from responses theresponses “bulk-response type” sensors cantype” improve the selectivity to the all target gases. addition offrom sensor from the “bulk-response sensors can improve selectivity to In all those of MiBK and acetone, since the “bulk-response type” sensors keep their characteristic patterns humid conditions, Figure 6d shows that the PCA scores of decane separated from those of MiBK and target gases. In humid conditions, Figure 6d shows that the PCA scores of decane separated from in humid conditions, as shown in Figure 5b. Therefore, the “bulk-response type” sensors should play acetone, “bulk-response sensors keep their characteristic patterns in humid conditions, those of since MiBKthe and acetone, sincetype” the “bulk-response type” sensors keep their characteristic patterns an important role in gas discrimination under humid conditions. as shown in Figure 5b. Therefore, the “bulk-response type” sensors should play an important role in in humid conditions, as shown in Figure 5b. Therefore, the “bulk-response type” sensors should play Eigenvector PC1 Eigenvector PC1 gas discrimination under humid conditions. an important role in gas discrimination under humid conditions.

Nonanal

Eigenvector PC1 #31b 0.4 #33b #31b

Acetoin 0.8 1.2

Decane

0.0

0.4

-0.4

0-1

Acetone

0

0.0

2602 2610 2600 2620 MiBK

Decane 1

2

-0.8

-0.4 3

4

-0.8

MiBK -1

0

1

(a)

2

PC1 (62.8%)

3

4

5

-0.5

3

0.0

0.5

1.0

Eigenvector PC1 0.0 Nonanal 0.5

-0.5

1.0 1.0

2610 1

0

-1

2

2620

0

0.5

2602 Nonanal

1.0

2600

2610

Decane 2620 1 MiBK 2600

-2 -2

5

AcetonePC1 (62.8%) -2 -2

2

0.8

Acetoin #33b 2602 2610 2600 2620

Nonanal

-1

3

0.4

10

-2 -1 -2

0.8

1.2

Eigenvector PC2

0.0

0.8

#31b #33b 2602

-1

0

1

2

3

0.5

-0.5

#31b #33b Acetoin

Decane Acetone MiBK

-1

0.0

4

-0.5

PC1 (61.0%)

-2 -2

Acetone -1

Acetoin

(b) 0

1 PC1 (61.0%)

(a)

(b) Figure 6. Cont.

2

3

0.0

4

Eigenvector PC2

-0.4

0.4

PC2 (33.6%) PC2 (33.6%)

PC2 (24.2%) PC2 (24.2%)

21

0.0

Eigenvector PC2 Eigenvector PC2

-0.4

2

Sensors 2017, 17, 1662 Sensors 2017, 17, 1662

10 of 14 10 of 14

Sensors 2017, 17, 1662

PC2 (27.6%) PC2 (27.6%)

2

1

Nonanal Nonanal

1

0.8

0.4

0

-1 -1

-2

Decane

Decane -3 -3 -2 -1 -3 -3 -2 -1

0

1

2 0 1 2 PC1 (54.5%) PC1 (54.5%)

3 3

4 4

0.0 PC1 Eigenvector

0.5

2610 Nonanal No.9 2610 Nonanal2602 No.9 2602

2620

2

2600

1

0.5

0.0

2

0.4

#33b 2602 0.0 Acetoin #33b 2610 0.0 2602 Acetone Acetoin 2600 2610 Acetone 2600 2620 -0.4 No.71 MiBK 2620 -0.4 No.71 MiBK

0

-2

0.8

0.8

No.9 #31b No.9 #31b

2

-0.5 3 -0.5 3

0.8

0.4

2620

2600

1

0 0 -1 -1

0.0 #31b 0.0 #31b #33b #33b

MiBK

No.71MiBK No.71

Decane Acetone Decane Acetone

-2 -2

-0.8 5 -0.8 5

0.5 0.5 Eigenvector PC2 Eigenvector PC2

3

0.0

PC2 (30.2%) PC2 (30.2%)

-0.4

Eigenvector PC2PC2 Eigenvector

-0.4

3

10 of 14

Eigenvector PC1

Eigenvector PC1 0.0 0.4 Eigenvector PC1

-0.5

Acetoin -0.5

-3 -3

-2 -2

-1 -1

(c) (c)

0

2 Acetoin 3 2 3

1

0 1 PC1 (52.9%) PC1 (52.9%)

(d) (d)

Figure 6. 6. Principal component component analysis (PCA) scores and eigenvectors from (a) six “grain boundaryFigure scores and andeigenvectors eigenvectorsfrom from(a) (a)six six“grain “grain boundaryFigure 6.Principal Principal componentanalysis analysis (PCA) scores boundaryresponse type” sensors in pure dry air; (b) six “grain boundary-response type” sensors in pure humid response type” boundary-responsetype” type”sensors sensorsininpure pure humid response type”sensors sensorsininpure puredry dryair; air; (b) (b) six “grain boundary-response humid air; (c)(c) six “grain boundary-response type” and two “bulk-response “bulk-responsetype” type”sensors sensorsinin inpure pure dry air; and air; (c) six “bulk-response type” sensors pure dry air; and air; six“grain “grainboundary-response boundary-responsetype” type” and and two dry air; and (d) six “grain boundary-response type” and two “bulk-response type” sensors in pure humid air. “grainboundary-response boundary-responsetype” type”and andtwo two “bulk-response “bulk-response type” (d)(d) sixsix “grain type” sensors sensorsin inpure purehumid humidair. air.

3.2. Sensor Responses and PCA in Humid Air with Contaminations 3.2. SensorResponses Responsesand andPCA PCAin inHumid Humid Air Air with with Contaminations Contaminations 3.2. Sensor Figure 7a shows the PCA scores and eigenvectors using six six“grain “grain boundary-response type” and Figure showsthe thePCA PCAscores scoresand and eigenvectors eigenvectors using and Figure 7a7ashows usingsix “grainboundary-response boundary-responsetype” type” and 3 of VOC contamination. PC1 3 two “bulk-response type” sensors in humid air with 0 and 300 μg/m two “bulk-response type” sensors humid with 0 and μg/m VOC contamination. PC1 3 of of two “bulk-response type” sensors in in humid air with 0 and 300300 µg/m VOC contamination. PC1 and and PC2 80.3%.The ThePCA PCAscores scorestend tendtotomove move a negative and PC2showed showedaasufficient sufficientcumulative cumulative variance of 80.3%. inin a negative PC2 showed a sufficient cumulative variance of 80.3%. The PCA scores tend to move in a negative directionparallel paralleltotothe thePC2 PC2axis axis with with contaminated contaminated indoor direction direction indoorair, air,and andscatter scatterininthe thepositive positive direction direction parallel to the PC2 axis with contaminated indoor air, and scatter in the positive direction paralleltotothe thePC1 PC1axis axis with with target target gases gases containing containing easily parallel easily oxidized oxidizedfunctional functionalgroups groupssuch suchas as parallel to the PC1 axis with target gases containing easily oxidized functional groups such as alcohols alcohols (acetoin) and aldehydes (nonanal). As shown in Figure 7a, PC1 tends to be related to thethe alcohols (acetoin) and aldehydes (nonanal). shown in Figure 7a, PC1 tends to be related to (acetoin) and aldehydes (nonanal). As shown in Figure 7a, PC1 tends to be related to the functional functionalgroup groupofofthe thetarget targetgases. gases. PC2 PC2 tends tends to because functional to be be related relatedto tototal totalorganic organiccarbon carbon(TOC), (TOC), because group of the target gases. PC2 tends to be related to total organic carbon (TOC), because the PC2 the PC2 score decreases with increasing concentration of indoor air contaminants. Thus, the PC2 the PC2 score decreases with increasing concentration of indoor air contaminants. Thus, the PC2 score decreases with increasing concentration of indoor air contaminants. Thus, the PC2 includes includes information about the contaminants, but PC1 does not. Therefore, the PCA scores with PC1 includes information about the contaminants, but PC1 does not. Therefore, the PCA scores with PC1 information about theprepared contaminants, but PC1 does not. from Therefore, the PCA scores withinPC1 and7b. PC3 andPC3 PC3were were also to eliminate eliminate interference and also prepared to interference from contamination, contamination,asasshown shown inFigure Figure 7b. were also prepared toFigure eliminate interference from contamination, as shown in Figure 7b. PC1 and PC3 PC1 and PC3 in 7b show that the PCA scores from each target gas with or without PC1 and PC3 in Figure 7b show that the PCA scores from each target gas with or without incontamination Figure 7b show the PCA scores from each target do gas with or without contamination are almost arethat almost identical, so the the PC1 PC1 and contamination. contamination are almost identical, so and PC3 PC3 donot notinclude includethe theinfluence influenceofof contamination. identical, so the PC1 and PC3 do not include the influence of contamination. As shown in Figure As shown in Figure 7b, PC3 also tends to be related to the functional groups of the target gases.7b, As shown in Figure 7b, PC3 also tends to be related to the functional groups of the target gases. PC3 also tends be related the functional of the target gases. However, theare PCA scores However, the to PCA scores oftoacetone coincidegroups with those of MiBK, and those of acetoin also closeof However, the PCA scores of acetone coincide with those of MiBK, and those of acetoin are also close acetone coincide with those of MiBK, and those of acetoin are also close to those of nonanal. to those of nonanal. to those of nonanal.

-1

-1 -2

-2 -3 -3

-3 -3

-2

-2

-1

-1

0.0 #31b 0.0 #33b #31b Acetoin #33b 0 0

Acetoin

0

1

PC1 0(48.5%) 1

(a) PC1 (48.5%)

2

2

0 300 0 00 300 300 300

300 300

-0.5 3

3

-0.5

1 No.71 1 No.71

2620 0

2600

0

0.5 0.51.0 1.0

0.0

Decane 0 300 Decane 300 0 300 0300 0300 0 300 0

(12.1%) PC3PC3 (12.1%)

0

Acetone

0 0 0 300 0 0 300 No.71 0 0 300300 300 300 300 300 No.71 300 300 300 300 300 300 300 300 300 300

2 2

Eigenvector Eigenvector PC2PC2

PC2 (31.8%) PC2 (31.8%)

0

Eigenvector PC1 Eigenvector PC1 0.0

-0.5 -0.5

2620

Acetoin 0.5 2610 0 0.5 00 300 2610 #33b Acetoin 0 #31b 00 300 #33b 300 300 #31b 00 300 3000 300 0.0 2602 300 300 00 300 Nonanal 300 3000 2602

2600

-1

Acetone

-1

Acetone -3

-3

-2

-2

00

-1

0

-1PC1 (48.5%) 0

0.0

Nonanal

300 MiBK 300 300 0 0300 300 0300300 MiBK 300 300 00 0 0300 300 0300

-0.5

-0.5

No.9

No.9 2

1

1

Eigenvector PC3 Eigenvector PC3

Eigenvector PC1 Eigenvector 0.0 PC1 0.5 0.5 0 26200.0 0 0 3 2600 2620 2610 0.5 0 Nonanal 0 0 2600 2610 0.5 2 Nonanal 0MiBK 300 Decane 2 No.9 300 300 0 0 0MiBK 300 2602 0 No.9 300 1Decane 300 0 0 2602 Acetone 0 1 0 0 0 -0.5 3-0.5

3

2

3

(b) PC1 (48.5%)

(a) and PC1 and PC2 eigenvectors and (b) PCA scores (b) and PC1 and PC3 Figure 7. (a) PCA scores Figure 7. (a) PCA scores and PC1 and PC2 eigenvectors and (b) PCA scores and PC1 and PC3 eigenvectors from scores six “grain boundary-response type” andand two(b) “bulk-response type”PC1 sensors in Figure 7. (a)from PCA PC1 and PC2 eigenvectors PCA scores and and PC3 eigenvectors six “grainand boundary-response type” and two “bulk-response type” sensors in humid 3 humid air with 0 and 300 μg/m of indoor air VOC contaminants. 3 eigenvectors from six “grain boundary-response type” and two “bulk-response type” sensors in air with 0 and 300 µg/m of indoor air VOC contaminants. humid air with 0 and 300 μg/m3 of indoor air VOC contaminants.

Sensors 2017, 17, 1662

11 of 14

Sensors 2017, 17, 1662

11 of 14

As shown in Figure 7a, on the PC2 axis, the PCA scores of MiBK are larger than those of acetone, As shown in Figure 7a, on the PC2 axis, the PCA scores of MiBK are larger than those of acetone, and those of nonanal are larger than those of acetoin. Therefore, the PC2 includes the molecular weight and those of nonanal are larger than those of acetoin. Therefore, the PC2 includes the molecular of the target gases and information about contamination. In Figure 7a, the eigenvectors of TGS 2600, weight of the target gases and information about contamination. In Figure 7a, the eigenvectors of 2610, and 2620 strongly contribute to PC2. As described above, PC1 and 3 in Figure 7b include barely TGS 2600, 2610, and 2620 strongly contribute to PC2. As described above, PC1 and 3 in Figure 7b any information about contamination. Moreover, the eigenvectors of TGS2600 and 2620 in Figure 7b include barely any information about contamination. Moreover, the eigenvectors of TGS2600 and are so small that7b they contribute the discrimination of discrimination each target gas.ofFigure 8 shows 2620 in Figure arehardly so small that they to hardly contribute to the each target gas. the average sensor responses of all sensors to 1 ppm of target gases in humid air with contamination, Figure 8 shows the average sensor responses of all sensors to 1 ppm of target gases in humid air withand the differences inand sensor responses (D) with and without Indeed, sensors such as TGS contamination, the differences in sensor responses (D)contamination. with and without contamination. Indeed, 2600 and 2620 with large D are strongly affected by contamination, suggesting that their elimination sensors such as TGS 2600 and 2620 with large D are strongly affected by contamination, suggesting from sensor array would reduce influence contamination. thatthe their elimination from the sensorthe array would of reduce the influence of contamination.

(a)

(b)

Figure 8. (a) Average sensor responses of all sensors to 1 ppm of target gases in humid air with 300 Figure 8. (a) Average sensor responses of all sensors to 1 ppm of target gases in humid air with μg/m3 of VOC contamination; and (b) difference in sensor responses (D) between 1 ppm of target 300 µg/m3 of VOC contamination; and (b) difference in sensor responses (D) between 1 ppm of target gases in pure humid air and those in humid air with 300 μg/m3 of VOC contamination. gases in pure humid air and those in humid air with 300 µg/m3 of VOC contamination.

Moreover, the removal of one sensor from a group of sensors with high correlation coefficients the removal of onethe sensor from a group ofscores sensors[41]; with high correlation coefficients hasMoreover, been reported to not affect distribution of PCA the correlation coefficients arehas been reported to not affect the distribution of PCA scores [41]; the correlation coefficients are shown shown in Table 4. In addition, #31b and #33b have a strong dependence on PC1, as shown in Figure in Table 4. In addition, #31b andalmost #33b have a strong dependence on PC1, as shown in Figure 7a. Their 7a. Their eigenvectors have the same magnitude and direction, and their high correlation eigenvectors have the that samethey magnitude andinterrelated. direction, and their high correlation coefficient coefficient of 0.99almost indicates are highly Consequently, we eliminated #31b of 0.99 indicates that they are highly interrelated. Consequently, we eliminated #31b because of itsto high because of its high correlation with #33b and TGS2600 and 2620 because of their strong responses correlation with as #33b and TGS2600 contamination, discussed above. and 2620 because of their strong responses to contamination, as discussed above. Sensor TGS2600 Sensor TGS2602 TGS2610 TGS2600 TGS2620 TGS2602 #31b TGS2610 #33b TGS2620 #31b No. 9 #33b No. 71 No. 9 No. 71

Table 4. Correlation coefficients of the eight sensors. Table 4. Correlation coefficients of the eight sensors.

TGS 2600 TGS 2602 TGS 2610 TGS 2620 #31b #33b 1 TGS TGS TGS TGS #31b #33b 2600−0.19 2602 1 2610 2620 0.41 0.70 1 1 0.84 0.18 0.58 1 −0.19 1 0.24 −0.19 1 0.41−0.56 0.70 0.75 1 −0.26 0.99 1 0.84−0.60 0.18 0.64 0.58 0.13 1 −0.56 0.75 0.62 0.24 0.33 −0.19 0.22 1 0.053 0.26 0.15 −0.600.38 0.64 −0.68 0.13 −0.056−0.26 0.065 0.99 −0.72 1−0.68 0.053 0.62 0.33 0.22 0.26 0.15 0.38 −0.68 −0.056 0.065 −0.72 −0.68

No. 9

No. 71

No. 9

1 −0.74 1 −0.74

No. 71

1 1

Figure 9 shows the PCA scores and eigenvectors from three “grain boundary-response type” (TGS2602, TGS2610, and #33b) and two “bulk-response type” sensors (No. 9 and No. 71) in humid Figure 9 shows the PCA scores and eigenvectors from three “grain boundary-response type” air with and without contamination. From PC1 and 2 in Figure 9a, the PCA scores can discriminate (TGS2602, TGS2610, and #33b) and two “bulk-response type” sensors (No. 9 and No. 71) in humid air all target gases and are hardly affected by the contamination. Thus, the discrimination between target with andcan without contamination. From PC1 and PC2 sensors in Figurefrom 9a, the scores can to discriminate gases be improved by selecting appropriate thePCA sensor array eliminate all target gases and are hardly affected by the contamination. Thus, the discrimination between target unnecessary information from the total data set. gases can be improved by selecting appropriate sensors from the sensor array to eliminate unnecessary information from the total data set.

Sensors 2017, 17, 1662

12 of 14

Sensors 2017, 17, 1662

12 of 14 Eigenvector PC1

Eigenvector PC1

PC2 (22.3%)

No.71

MiBK

300 300 300

0

Acetone

-2

0 0

#33b 0 0

-2

-1

0 0 0 300 300

2602 No.9

0

300 0 300 300

-1

Nonanal

2610

3000 300 0 300 0

1

2

0.5

0.0

0 00 300 300 300

300 3000 300 0 0

#33b 0.5

2610

No.71 0

2602 0.0

300 300 00 300

Nonanal

0

Acetone

300 MiBK 300 300 300 300 00 300 0 000

-1

-1.0 3

0.5

Acetoin

Decane 1

-0.5

300 00 0 300 300 Acetoin

0

1.0

0.0

-3

-2

-1

0

Eigenvector PC3

Decane 1

-0.5

0.5

PC3 (16.6%)

0.0

Eigenvector PC2

2

-0.5

-0.5 No.9 1

PC1 (59.5%)

PC1 (59.5%)

(a)

(b)

2

3

Figure 9. (a) PCA scores and PC1 and PC2 eigenvectors and (b) PCA scores and PC1 and PC3 Figure 9. (a) PCA scores and PC1 and PC2 eigenvectors and (b) PCA scores and PC1 and PC3 eigenvectors from three “grain boundary-response type” (TGS2602, TGS2610, and #33b) and two eigenvectors from three “grain boundary-response type” (TGS2602, TGS2610, and #33b) and two “bulk-response type” sensors (No. 9 and No. 71) in pure humid air and humid air with 300 μg/m3 of “bulk-response type” sensors (No. 9 and No. 71) in pure humid air and humid air with 300 µg/m3 of VOC contamination. VOC contamination.

4. Conclusions 4. Conclusions All eight sensors exhibit decreases in resistance in response to target VOCs. The base resistances All eight exhibit decreases in resistance in response target The base of almost all sensors the “grain boundary-response type” sensors in pure to humid airVOCs. were lower thanresistances those in of pure almostdry all the “grain boundary-response type” sensors in pure humid air were lower than those air, and the intensities of their resistance responses decreased. Furthermore, the in pure dry air, andsensor the intensities their resistance responses decreased. theflat characteristic characteristic patterns ofofall target gases decreased drastically andFurthermore, became nearly patterns. sensor patterns of all target gases decreased drastically and nearly flat patterns. However, However, the “bulk-response type” sensors showed almost thebecame same base resistance in pure dry and thehumid “bulk-response almost the same base resistance pure dry and humid air, and the type” sensorsensors patternsshowed of the “bulk-response type” sensors hardly in changed in humid air. resultspatterns showedofthat the “bulk-response type” sensors play an important air,The andPCA the sensor the “bulk-response type” sensors hardly changed in humidrole air. in The discriminating eachthat target in humid air.type” The PC2 scoreplay tends be relatedrole to the total organic PCA results showed theVOC “bulk-response sensors antoimportant in discriminating carbon (TOC) thus, included relatedtotobe VOC contamination, the PC1 and PC3 each target VOCand, in humid air. Theinformation PC2 score tends related to the totaland organic carbon (TOC) scores tend to be related to the functional groups of the target gases. From PC1 and PC2, we removed and, thus, included information related to VOC contamination, and the PC1 and PC3 scores tend to contributions contaminations, i.e., large differences in sensor sensors responses besensors related with to thelarge functional groupstoofVOC the target gases. From PC1 and PC2, we removed with 3 (D) between pure humid air and pure humid air with 300 μg/m of VOC contamination, to reduce large contributions to VOC contaminations, i.e., large differences in sensor responses (D) between thehumid effects air of contamination on the sensor array. Consequently, selectivity for to thereduce target the gases was of 3 of VOC contamination, pure and pure humid air with 300 µg/m effects maintained, and the VOC contamination did not greatly affect the measurements. contamination on the sensor array. Consequently, selectivity for the target gases was maintained, and theAcknowledgments: VOC contamination did not greatly affect the measurements. We express our sincere thanks to Hyung-Gi Byun (Kangwon National University, South Korea) for his kind advice on principal component analysis. Acknowledgments: We express our sincere thanks to Hyung-Gi Byun (Kangwon National University, South Korea) for his kind Toshio advice Itoh on principal component analysis.and designed the present study; Toshio Itoh Author Contributions: and Woosuck Shin conceived carried out the gas sensor analyses and discriminated from the sensor responsesthe using principal component Author Contributions: Toshio Itoh and Woosuck Shin conceived and designed present study; Toshio Itoh analyses; analyzed and discussed the experimental results; Toshio Itoh wrote theprincipal paper. component carried out all theauthors gas sensor analyses and discriminated from the sensor responses using analyses; all authors analyzed and discussed the experimental results; Toshio Itoh wrote the paper. Conflicts of Interest: The authors declare no conflicts of interest. Conflicts of Interest: The authors declare no conflicts of interest.

References References 1.

Wang, C.; Mbi, A.; Shepherd, M. A Study on Breath Acetone in Diabetic Patients Using a Cavity Ringdown Breath with Blood Glucose andUsing Glycohemoglobin A1C. Wang, C.;Analyzer: Mbi, A.; Exploring Shepherd,Correlations M. A Studyof onBreath BreathAcetone Acetone in Diabetic Patients a Cavity Ringdown IEEE Analyzer: Sens. J. 2010, 10, 54–63. Breath Exploring Correlations of Breath Acetone with Blood Glucose and Glycohemoglobin A1C. 2. IEEE Poli,Sens. D.; Carbognani, P.; Corradi, M.; Goldoni, M.; Acampa, O.; Balbi, B.; Bianchi, L.; Rusca, M.; Mutti, A. J. 2010, 10, 54–63. [CrossRef] Exhaled volatile organic compounds patients M.; withAcampa, non-small cell lung B.; cancer: CrossL.; sectional 2. Poli, D.; Carbognani, P.; Corradi, M.; in Goldoni, O.; Balbi, Bianchi, Rusca, and M.; nested Mutti, A. short-term follow-up study. Respir. Res. 2005, 6, 71–80. Exhaled volatile organic compounds in patients with non-small cell lung cancer: Cross sectional and nested 3. short-term Fuchs, P.;follow-up Loeseken, C.; Schubert, W. Breath gas aldehydes as biomarkers of lung cancer. Int. study. Respir.J.K.; Res.Miekisch, 2005, 6, 71–80. [CrossRef] [PubMed] J. Cancer 2010, 126, 2663–2670. 3. Fuchs, P.; Loeseken, C.; Schubert, J.K.; Miekisch, W. Breath gas aldehydes as biomarkers of lung cancer. Int. J. Cancer 2010, 126, 2663–2670. [CrossRef] [PubMed]

1.

Sensors 2017, 17, 1662

4.

5. 6.

7.

8.

9. 10. 11.

12.

13. 14.

15. 16. 17.

18.

19. 20. 21. 22. 23.

24.

13 of 14

Song, G.; Qin, T.; Liu, H.; Xu, G.B.; Pan, Y.Y.; Xiong, F.X.; Gu, K.S.; Sun, G.P.; Chen, Z.D. Quantitative breath analysis of volatile organic compounds of lung cancer patients. Lung Cancer 2010, 67, 227–231. [CrossRef] [PubMed] Amann, A.; Spanel, P.; Smith, D. Breath analysis: The approach towards clinical applications. Mini Rev. Med. Chem. 2007, 7, 115–129. [CrossRef] [PubMed] Peng, G.; Tisch, U.; Adams, O.; Hakim, M.; Shehada, N.; Broza, Y.Y.; Billan, S.; Bortnyak, R.A.; Kuten, A.; Haick, H. Diagnosing lung cancer in exhaled breath using gold nanoparticles. Nat. Nanotech. 2009, 4, 669–673. [CrossRef] [PubMed] Westhoff, M.; Litterst, P.; Freitag, L.; Urfer, W.; Bader, S.; Baumbach, J.-I. Ion mobility spectrometry for the detection of volatile organic compounds in exhaled breath of patients with lung cancer: Results of a pilot study. Thorax 2009, 64, 744–748. [CrossRef] [PubMed] Mazzone, P.J.; Hammel, J.; Dweik, R.; Na, J.; Czich, C.; Laskowski, D.; Mekhail, T. Diagnosis of lung cancer by the analysis of exhaled breath with a colorimetric sensor array. Thorax 2007, 62, 565–568. [CrossRef] [PubMed] Herberger, S.; Herold, M.; Ulmer, H.; Burdack-Freitag, A.; Mayer, F. Detection of human effluents by a MOS gas sensor in correlation to VOC quantification by GC/MS. Build. Environ. 2010, 45, 2430–2439. [CrossRef] Yang, J.-W.; Cho, H.-J.; Lee, S.-H.; Lee, J.-Y. Characterization of SnO2 ceramic gas sensor for exhaust gas monitoring of SVE Process. Environ. Monit. Assess. 2004, 92, 153–161. [CrossRef] [PubMed] Wolfrum, E.J.; Peterson, R.M.M.D.; Sluiter, J. Metal oxide sensor arrays for the detection, differentiation, and quantification of volatile organic compounds at sub-parts-per-million concentration levels. Sens. Actuators B Chem. 2006, 115, 322–329. [CrossRef] Szczurek, A.; Maciejewska, M.; Flisowska-Wiercik, B.; Bodzoj, L. Application of a sensor system for determining the kind and quantity of two component VOC mixtures in air after the use of solvents. J. Environ. Monit. 2009, 11, 1942–1951. [CrossRef] [PubMed] Wolfrum, E.J.; Meglen, R.M.; Peterson, D.; Sluiter, J. Calibration transfer among sensor arrays designed for monitoring volatile organic compounds in indoor air quality. IEEE Sens. J. 2006, 6, 1638–1643. [CrossRef] Itoh, T.; Matsubara, I.; Kadosaki, M.; Sakai, Y.; Shin, W.; Izu, N.; Nishibori, M. Effects of High-Humidity Aging on Platinum, Palladium, and Gold Loaded Tin Oxide—Volatile Organic Compound Sensors. Sensors 2010, 10, 6513–6521. [CrossRef] [PubMed] Itoh, T.; Matsubara, I.; Nishibori, M.; Izu, N.; Shin, W. Calibration Gas Preparation for Non-Disposable Portable MOx, PID, and IER VOC Detectors. Sens. Lett. 2012, 10, 985–992. [CrossRef] Manorama, S.; Devi, G.S.; Rao, V.J. Hydrogen sulfide sensor based on tin oxide deposited by spray pyrolysis and microwave plasma chemical vapor deposition. Appl. Phys. Lett. 1994, 64, 3163–3165. [CrossRef] Shimura, M.; Yasuno, Y.; Iwakura, M.; Shimada, Y.; Sakai, S.; Suzuki, K.; Sakamoto, S. A new monitor with a zinc-oxide thin film semiconductor sensor for the measurement of volatile sulfur compounds in mouth air. J. Periodontol. 1996, 67, 396–402. [CrossRef] [PubMed] Tanda, N.; Washio, J.; Ikawa, K.; Suzuki, K.; Koseki, T.; Iwakura, M. A new portable sulfide monitor with a zinc-oxide semiconductor sensor for daily use and field study. J. Dent. 2007, 35, 552–557. [CrossRef] [PubMed] Itoh, T.; Taguchi, Y.; Izu, N.; Matsubara, I.; Nakamura, S.; Suzuki, K.; Kanda, K.; Shin, W.; Nishibori, M. Alternating Current Impedance Analysis of CeO2 Thick films as Odor Sensors. Sensors 2011, 9, 703–705. Byun, H.-G.; Persaud, K.C.; Pisanelli, A.M. Wound-State Monitoring for Burn Patients Using E-Nose/SPME System. ETRI J. 2010, 32, 440–446. [CrossRef] Yu, J.-B.; Byun, H.-G.; Zhang, S.; Do, S.-H.; Lim, J.-O.; Huh, J.-S. Exhaled Breath Analysis of Lung Cancer Patients Using a Metal Oxide Sensor. J. Sens. Sci. Technol. 2011, 20, 300–304. [CrossRef] Hossein-Babaei, F.; Amini, A. A breakthrough in gas diagnosis with a temperature-modulated generic metal oxide gas sensor. Sens. Actuators B Chem. 2012, 166–167, 419–425. [CrossRef] ´ Dymerski, T.; G˛ebicki, J.; Wi´sniewska, P.; Sliwi nska, ´ M.; Wardencki, W.; Namie´snik, J. Application of the Electronic Nose Technique to Differentiation between Model Mixtures with COPD Markers. Sensors 2013, 13, 5008–5027. [CrossRef] [PubMed] Itoh, T.; Nakashima, T.; Akamatsu, T.; Izu, N.; Shin, W. Nonanal gas sensing properties of platinum, palladium, and gold-loaded tin oxide VOCs sensors. Sens. Actuators B Chem. 2013, 187, 135–141. [CrossRef]

Sensors 2017, 17, 1662

25. 26. 27.

28. 29.

30.

31. 32.

33.

34. 35. 36. 37.

38.

39.

40.

41.

14 of 14

Sakai, Y.; Kadosaki, M.; Matsubara, I.; Itoh, T. Preparation of total VOC sensor with sensor-response stability for humidity by noble metal addition to SnO2 . J. Ceram. Soc. Jpn. 2009, 117, 1297–1301. [CrossRef] Zaromb, S.; Stellar, J.R. Theoretical basis for identification and measurement of air contaminants using an array of sensors having partially overlapping sensitivities. Sens. Actuators 1984, 6, 225–243. [CrossRef] Itoh, T.; Akamatsu, T.; Izu, N.; Shin, W.; Byun, H.-G. Monitoring of disease-related volatile organic compounds in simulated room air. In Proceedings of the 2014 IEEE Sensors, Valencia, Spain, 2–5 November 2014; pp. 1427–1430. Cesare, F.D.; Pantalei, S.; Zampetti, E.; Macagnano, A. Electronic nose and SPME techniques to monitor phenanthrene biodegradation in soil. Sens. Actuators B Chem. 2008, 131, 63–70. [CrossRef] Natale, C.D.; Macagnano, A.; Martinelli, E.; Paolesse, R.; D’Arcangelo, G.; Roscioni, C.; Finazzi, A.; D’Amico, A. Lung cancer identification by the analysis of breath by means of an array of non-selective gas sensors. Biosens. Bioelectron. 2003, 18, 1209–1218. [CrossRef] European Commission Joint Research Centre Environment Institute. Indoor Air Quality and Its Impact on Man—Report No. 19: Total Organic Compounds (TVOC) in Indoor Air Quality Investigations; European Commission Joint Research Centre Environment Institute: Ispra, Italy, 1997. Ministry of Health, Labour and Welfare Japan. Midterm Report of Sick Building Syndrome Committee—Parts 6 and 7; Ministry of Health, Labour and Welfare Japan: Tokyo, Japan, 1999. ISO 16000-6. Indoor Air—Part 6: Determination of Volatile Organic Compounds in Indoor and Test Chamber Air by Active Sampling on Tenax TA Sorbent, Thermal Desorption and Gas Chromatography Using MS/FID; ISO: Geneva, Switzerland, 2004. Kadosaki, M.; Sakai, Y.; Tamura, I.; Matsubara, I.; Itoh, T. Development of oxide semiconductor thick film gas sensor for the detection of total volatile organic compounds. Electron. Commun. Jpn. 2008, 93, 34–41. [CrossRef] Yamazoe, N.; Shimanoe, K. Theory of power laws for semiconductor gas sensors. Sens. Actuators B Chem. 2008, 128, 566–573. [CrossRef] Yamazoe, N.; Shimanoe, K. New perspectives of gas sensor technology. Sens. Actuators B Chem. 2009, 138, 100–107. [CrossRef] Izu, N.; Shin, W.; Matsubara, I.; Murayama, N. Development of Resistive Oxygen Sensors Based on Cerium Oxide Thick Film. J. Electroceram. 2004, 13, 703–706. [CrossRef] Izu, N.; Shin, W.; Matsubara, I.; Murayama, N. Evaluation of response characteristics of resistive oxygen sensors based on porous cerium oxide thick film using pressure modulation method. Sens. Actuators B Chem. 2006, 113, 207–213. [CrossRef] Itoh, T.; Izu, N.; Akamatsu, T.; Shin, W.; Miki, Y.; Hirose, Y. Elimination of Flammable Gas Effects in Cerium Oxide Semiconductor-Type Resistive Oxygen Sensors for Monitoring Low Oxygen Concentrations. Sensors 2015, 15, 9427–9437. [CrossRef] [PubMed] Itoh, T.; Uchida, T.; Matsubara, I.; Izu, N.; Shin, W.; Miyazaki, H.; Tanjo, H.; Kanda, K. Preparation of γ-alumina large grain particles with large specific surface area via polyol synthesis. Ceram. Int. 2007, 41, 3631–3638. [CrossRef] Itoh, T.; Lee, D.; Goto, T.; Akamatsu, T.; Izu, N.; Shin, W.; Kasuga, T. Analysis of Recovery Time of Pt- Pd- and Au-Loaded SnO2 Sensor Material with Nonanal as Large-Molecular-Weight Volatile Organic Compounds. Sens. Mater. 2016, in press. [CrossRef] ASrivastava, K.; Dravid, V.P. On the performance evaluation of hybrid and mono-class sensor arrays in selective detection of VOCs: A comparative study. Sens. Actuators B Chem. 2006, 117, 244–252. [CrossRef] © 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/).