Calibration of Condensation Particle Counters for Legislated Vehicle Number Emission Measurements

Abstract

Light duty vehicle emissions legislation requires calibration and validation of Condensation Particle Counters (CPCs). A workshop was organized at the European Commission's Joint Research Centre (JRC, Ispra, Italy) to study the effect of test aerosol materials (of their chemical composition) on CPC calibration results. The counting efficiencies of Combustion Aerosol STandard (CAST) particles at 23 nm (steep part of the counting efficiency curve) were found to be similar (0.53) to those of heavy duty diesel engine particles (0.57), the counting efficiencies of Emery oil were either similar (0.53) or higher (0.72), while those of tetracontane (C₄₀H₈₂) particles were much higher (0.86). However, tests performed at JRC after the workshop found much lower counting efficiencies for tetracontane particles (almost 0 at 23 nm) and variable results for NaCl (0.6 or lower for 23 nm) indicating the importance of the generation method and the thermal treatment of the generated aerosol. Measurement issues including calibration against an electrometer or a reference CPC, the effect of multiply charged particles on counting efficiencies, stability, repeatability, reproducibility and comparability of CPCs and electrometers of different manufacturers were also investigated.

1. INTRODUCTION

Particle number measurement has been introduced in Euro 5/6 light duty vehicles legislation based on the findings of the Particle Measurement Programme (PMP) (e.g., Giechaskiel et al. 2008a, b). The particle number measurement system consists of two main parts: the volatile particle remover (or sample preconditioning unit) and the particle number counter. The main requirements for the particle number counter according to the legislation are (Regulation 83):

•	To operate under full flow operating conditions (no bypass flow).
•	To have a linear response to particle concentrations over the full measurement range in single particle count mode.
•	To have a counting accuracy of ± 10% across the range from 1 cm− 3 to the upper threshold of the single particle count mode against a traceable standard (with up to 10% coincidence correction).
•	To have 0.50 ± 0.12 and > 0.90 counting efficiencies at particle electrical mobility diameters of 23 ± 1 nm and 41 ± 1 nm, respectively.

Condensation Particle Counters (CPC) compliant with these legislative requirements are known as PMP CPCs. These CPCs should be calibrated at least annually with one of the following two calibration methods:

•	Primary method: By comparison of the response of the CPC under calibration (test CPC) with that of a calibrated aerosol electrometer (AE) when simultaneously sampling electrostatically classified calibration particles, or
•	Secondary method: By comparison of the response of the CPC under calibration with that of a second one (reference CPC) which has been directly calibrated by the primary method.

These calibration procedures are not new. Both methods have been used by CPC manufacturers and research labs (e.g., Wiedensohler et al. 1997; Sem 2002). The advantage of the primary method is that the electrometer measurement can be traceable to international standards. Its disadvantages are that the particle charge state has to be known, and electrometer noise introduces measurement errors at low concentrations. On the other hand, the secondary method is less affected by the particle charge state, and it is able to measure very low concentrations. However, very few studies have compared the equivalence of these two methods (e.g., Fletcher et al. 2009).

Another open issue on the CPC calibration procedures, and maybe the most important, is the selection of the test aerosol material. CPC's counting efficiencies strongly depend on aerosol properties, thus the calibration curve is strictly valid for the test aerosol (e.g., Kesten et al. 1991; Sem 2002; Ankilov et al. 2002). As different CPC manufacturers may use different aerosol materials for calibration (e.g., TSI uses Emery oil and GRIMM uses NaCl), CPCs from different manufacturers may have different counting efficiencies. Since the PMP CPCs will be used to measure engine emissions, it would be desirable to use diesel engine exhaust aerosols for calibration. The properties of diesel aerosol, however, depend on many parameters (e.g., engine type, engine load, fuel) and diesel aerosol can contain a wide range of materials (e.g., soot, sulfuric acid, hydrocarbons). As it is very difficult to define a “diesel engine” aerosol, the main objective of this study is to identify a material with behavior similar to diesel exhaust aerosol. Moreover, to our knowledge no previous studies have investigated the calibration of n-butanol-based CPCs with diesel engine particles.

Another concern is how robust, complete and applicable are the proposed calibration/validation procedures. So far, CPC calibration was conducted by manufacturers (e.g., Sem 2002) or highly specialized labs (e.g., Stolzenburg and McMurry 1991; Wiedensohler et al. 1997; Kulmala et al. 2007). In the future, all labs that measure vehicle emissions will have to validate the performance of their CPCs on site. A clear CPC calibration procedure is needed. Furthermore, it is necessary to identify measurement uncertainties, such as CPC performance stability (one day), repeatability (different days), reproducibility (different labs). Specifically, the effect of multiply charged particles and mitigation approaches should be clearly addressed. These uncertainties, although well known (e.g., Rose et al. 2008), have not been clearly discussed or quantified so far for “every day” lab application. A workshop was organized by the European Commission's Joint Research Centre (JRC, Ispra, Italy) in December 2007 to address these issues. Various companies participated in this workshop with CPCs (GRIMM, TSI) and particle generators (TSI, GRIMM, AEA, Matter Eng.). JRC provided heavy duty diesel particles, tested the applicability of the required validation procedures, and coordinated the data analysis.

2. EXPERIMENTAL

2.1. Experimental Set Up

The schematic of the set up for CPC calibration is shown in Figure 1 (Marshall and Sandbach 2007). Each company sampled polydisperse aerosol from a generator. The aerosol first passed through a dilution bridge that controlled its concentration. Next, a differential mobility analyzer (DMA) was used to select particles of a given mobility diameter. The sheath-to-aerosol flow rate ratio of the DMA was set at > 5:1 to ensure a narrow “monodisperse” size distribution of different sizes (23, 41 nm and a larger diameter (> 50) for linearity tests). Filtered makeup flow air was added downstream of the DMA to maintain a flow balance. A mixing orifice was used to enhance turbulent mixing and to ensure uniform aerosol concentration. The aerosol flow was then split to the “test” CPC (with a nominal 50% cut-point (d ₅₀) around 23 nm), the “reference” CPC (d ₅₀ around 3 nm), and the aerosol electrometer. Short, conductive tubes that led to similar diffusion losses were used (Hinds 1999). When attainable, the counting efficiencies at 23 nm and 41 nm were measured at concentrations of ∼ 6000 cm^{− 3} to avoid uncertainties due to coincidence correction or electrometer noise. For the engine tests at 23 nm the concentration was only about 2500 cm^{− 3}. Electrometers have ± 1 fA noise at 1 s averaging time (± 375 cm^{− 3} at a flow rate of 1 L/min) (Sem 2002), which translates to an uncertainty of 15% for the engine tests at 23 nm and 6% for the other tests. Linearity was measured at a size > 50 nm at concentrations of 10000, 8000, 6000, 4000, 2000, and 0 cm^{− 3}. Each data point was recorded for 2 min at 1 Hz data acquisition rate.

FIG. 1 Schematic of the set up for the calibration of CPCs.

Before the beginning and after the end of the measurements a Scanning Mobility Particle Sizer (SMPS) (or the DMA was connected to the reference CPC) was used to measure the size distribution that entered the DMA. TSI and GRIMM used their own instrumentation and similar set ups and in most cases they measured in parallel. Each company performed separately the data acquisition, and the raw data were provided to JRC for analysis.

2.2. Instrumentation

Particle Generators

The generators used in this workshop included a heavy duty diesel engine as well as electrospray, evaporation-condensation, and CAST generators. The particle generators will be described briefly below.

Diesel Engine

An INECO Cursor 8 heavy duty engine without any aftertreatment was used as the diesel soot source. The TSI instruments sampled from the tailpipe through a 20 m line in parallel with an extra pump to reduce the residence time in this line (estimated 2.5 s). At medium load and speed (referred as “load”) the size distribution was monomodal (mean 60 nm, standard deviation σ = 1.8). At idle (referred as “idle”), the size distribution was bimodal, with mean 60 nm, σ = 1.8 and mean 18.5 nm, σ = 1.28, even though a thermodenuder was installed upstream of the DMA. Giechaskiel et al. (2009) showed that a thermodenuder without hot dilution is not adequate to remove completely semi-volatiles if their concentration is high, thus some semi-volatiles might have remained on the particles at the idle condition. For the engine tests GRIMM measured independently from the full dilution tunnel (CVS) downstream of a thermodenuder and an ejector dilutor at the same medium load and speed (referred as Engine “CVS” tests). Due to low particle concentrations (1000 cm^{− 3}) these measurements had high uncertainty (> 35%).

Evaporation-Condensation Technique

The aerosol generators (AEA prototypes) consisted of a ceramic crucible heated via an electric Bunsen. The bulk material (NaCl or tetracontane) was placed in the ceramic crucible and heated to near its boiling point. A small air flow was introduced into the crucible to displace vapor from the surface of the bulk material to a cooler region of the generator where condensation occurred. Filtered air was also added to dilute and cool the sample and to increase the flow rate. Particle diameters could be varied by controlling the rate of vapor transport from the crucible (via the crucible displacement air flow) and/or the subsequent dilution and cooling rate of the vapor (via the carrier air flow). Two different aerosol generators were used for NaCl (mean 75 nm, σ = 1.3) and for tetracontane particles (mean 21 nm and σ = 1.4, or mean 41 nm and σ = 1.6). A few additional tests were conducted by the JRC after the workshop with two different (but similar) tetracontane and NaCl evaporation-condensation generators. The main difference was that filtered N₂ was used as the displacement flow.

Electrospray Technique

This method refers to the generation of liquid droplets by feeding a liquid solution or suspension through a capillary tube and applying an electric field to the liquid at the capillary tip (Chen et al. 1995). The electric field draws the liquid from the tip into a conical jet from which ultrafine charged droplets are emitted. Air (and optionally CO₂) is merged with the droplets and the liquid evaporates while the charge is neutralized by an ionizer. The result is a neutralized, monodisperse aerosol that is practically free of solvent residue. This method was used (TSI model 3480, S/N: 70515032) to electrospray Emery oil (PAO 4 cSt) for CPC calibration (mean diameters of 22, 40, and 54 nm with σ = 1.1). Emery oil is supposed to provide spherical particles of chemical composition representative of synthetic lube oil particles. Emery oil, a highly branched isoparaffinic polyalphaolefin (1-decene (tetramer) mixed with 1-decene (trimer), hydrogenated), usually consists of 82–85% C₃₀H₆₀ and 13–16% C₄₀H₈₀ polyalphaolefins by volume.

CAST (Combustion Aerosol STandard)

The CAST generators use a diffusion flame to form soot particles during pyrolysis. Within the soot-generating burner the flame is mixed with a quenching gas at a definite flame height. Consequently, combustion is quenched and a particle flow arises from the flame that leaves the combustion chamber. Sufficient quenching stabilizes soot particles and inhibits condensation in the particle stream when it escapes from the flame unit into ambient air. Subsequently, air is supplied to dilute the particle stream. The state of the flame and the features of generated soot particles are primarily a function of the flow settings of the gases. The mini-CAST generator from GRIMM (number distribution mean 20 or 30 nm with σ = 1.4, S/N: 001) and the CAST generator (mean 40 nm, σ = 1.4, S/N: 100907) from Matter Eng. were used.

Electrometers

Aerosol electrometer (AE) is a primary standard that measures the net charge of an aerosol. The charge is measured in a Faraday cup where the charge initiates a small current that is converted to a voltage using a high resistance resistor. The GRIMM model 5.705 (flow rate 1.5 L/min, S/N: 57050503) and the TSI 3068B (flow rate 1 L/min, S/N: 70601289) electrometers were used.

Particle Number Counters

GRIMM and TSI each tested at least 3 counters. Due to space limitations only the most representative results will be presented in this article. The results were similar for the other counters. GRIMM used one CPC (Model 5.403, S/N: 003) with d ₅₀ at 3 nm as a reference CPC for the secondary calibration method, and one test CPC (Model 5.404, S/N: 608) with d ₅₀ at 4.5 nm. All GRIMM CPCs operated at 1.5 L/min. The CPCs had been calibrated 6 months (test CPC) and 3 years (reference CPC) before the workshop using nebulised NaCl particles. TSI used a 3790 (S/N: 70721012) as test CPC (with d ₅₀ at 23 nm) with a flow rate of 1 L/min. It had been calibrated with Emery oil particles 6 months before the workshop.

Differential Mobility Analyzers

GRIMM used a Vienna-Type M-DMA (5 to 350 nm, S/N: 5UP60710) (Reischl et al. 1997). It was controlled and set to the specified sizes with a DMA controller. It had been last calibrated 7 months before the workshop. TSI used a 3080 electrostatic classifier with a nano-column DMA (calibrated 6 months before the workshop, S/N: 70424125). The flow rates of DMAs were controlled internally.

Flow, Temperature, and Pressure Measurements

Flow rate measurements were conducted with a soap bubble meter (mini-BUCK Calibrator M-5) (1–6000 cm³/min) with a ± 0.5% accuracy of the display reading. The ambient temperature and pressure remained constant during the measurements (21.5 ± 1°C and 98.5 ± 1.5 kPa, respectively). These ambient conditions were similar to those that the companies had during their calibrations, so no extra correction for the ambient conditions was needed.

2.3. GRIMM–TSI Comparison

For a limited number of tests some of the TSI and GRIMM instruments were used in parallel to compare directly the electrometers and the CPCs. Particles were generated by the TSI electrospray and classified by the GRIMM classifier. The GRIMM electrometer, the GRIMM test CPC model 5.404, the TSI electrometer and the TSI test CPC 3790 sampled the monodisperse aerosol in parallel (figure not shown). Only counting efficiencies at 23 nm and 41 nm were measured.

2.4. Effect of Multiply Charged Particles

Besides the monodisperse singly charged particles that exit a DMA, there is a smaller amount of multiply charged particles with the same electrical mobility but larger size (Baron and Willeke 2005). Multiply charged particles have a two fold effect:

•	The electrometer overestimates particle concentration due to the higher current generated by multiply charged particles. This can lead to lower test CPC linearity slopes and lower test CPC counting efficiencies.
•	The test CPCs seem to have higher counting efficiency because multiply charged particles are larger than singly charged particles with the same mobility diameter.

The contribution of these effects is difficult to calculate precisely, so the multiply charged fraction should be minimized. One rigorous way to correct the experimental error due to multiple charging is to carry out a Tandem Differential Mobility Analysis (TDMA) experiment to determine the fraction of multiply charged particles and correct the efficiency data. One simpler way to minimize multiple charging effects is to sample the test “monodisperse” aerosol from the right-hand side of the mode of the polydisperse aerosol size distribution. This procedure was followed for the measurements described in this article. The equations used to estimate the fraction of doubly to singly charged particles and its effect on the counting efficiencies can be found in the Supplemental information.

2.5. Effect of DMA Transfer Function on CPC Counting Efficiencies

The particle size selected by a DMA is not strictly monodisperse. Instead, it has a narrow size range defined by the DMA transfer function, even if contamination by multiply charged particles is not considered (Knutson and Whitby 1975). For a sheath-to-sample flow rate ratio of 5:1, the mobility range of particles exiting the DMA is Z*± 0.2Z*, where Z* is the DMA centroid mobility. This corresponds to a size range d _l −d _u (lower and upper diameters) of 21.0–25.7 nm for 23 nm, 37.4–45.9 nm for 41 nm, 54.7–67.2 nm for 60 nm. To estimate the effect of the transfer function on the counting efficiency at 23 nm (i.e., the effect of particles with Z different from Z*), we used a hypothetical PMP CPC with a steep counting efficiency curve (0.35 at 20 nm to 0.63 at 26 nm) to integrate concentrations of DMA classified particles. The different polydisperse aerosols prior to DMA classification were assumed to have lognormal distribution with different mean diameters and standard deviation. The DMA transfer function was assumed to be triangular (Knutson and Whitby 1975). We found that the transfer function had negligible effects on the measured counting efficiency (< 0.01 or < 2% for counting efficiency 0.5) for size distributions with σ > 1.2 or size distributions with σ < 1.2 and peak around 23 nm (± 2 nm). Similar conclusions were drawn by Alofs et al. (1995) and Rose et al. (2008). The effect would be even smaller for sheath-to-sample flow rate ratio of 10:1.

The broadening of the DMA transfer function due to particle diffusion was found to be negligible for particles larger than 20 nm and the flow rates selected (Stolzenburg 1988) and was not taken into account (Reischl et al. 1997; Mamakos et al. 2007).

2.6. Calculations

In the following results, the average of the 2 min electrometers readings were corrected for the zero levels (drift or offset) and their flow rates (although the correction was negligible). In general, the electrometer drift after the end of the tests with each material (30–60 min) was small (± 0.5 fA) with the exception of the engine tests where the drift was higher (± 2 fA) and the signal was unstable. The CPCs were not externally corrected for coincidence as both TSI 3790 and GRIMM 5.404 have inbuilt corrections. They were neither corrected for their flow rates, because users should have only one value to correct their results and not separately the flow and the slope. This effect would be around 1.2% for TSI 3790 (which had a flow rate of 0.988 L/min) and negligible for the GRIMM CPC 5.404. The effect of multiple charges was calculated through Equation (S9) (primary method) or Equation (S11) (secondary method), otherwise Equation (S15) or (S16) was used: equation numbers refer to the equations in the Supplemental information. We assumed that particle losses in sampling lines from the DMA outlet to test instruments were the same: no correction for sampling losses was considered.

The gradient (slope) and the square of the Pearson product moment correlation coefficient (R ²) during linearity test were calculated by forcing the fit to pass through the origin. As the slope and the R ² are not adequate indications of linearity and close agreement of two instruments (Fletcher et al. 2009 and Giechaskiel and Stilianakis 2009), the ratio of the CPC (under calibration) to the electrometer was calculated for each concentration. This ratio, according to legislation, should be between 0.9 and 1.1 for all concentrations tested. This ratio was also calculated for the 23 and 41 nm cases (it should be 0.38–0.62 and > 0.9, respectively). The coefficient of variance (CoV) of these ratios, the ratio of the standard deviation to the mean value, during the 2 min recordings was also calculated (stability of measurements).

For some tests, referred to as “repeated,” the method adopted by the Japanese National Institute of Advances Industrial Science and Technology (AIST) was followed: the DMA voltage was turned on/off alternatively for one minute and each measurement was repeated 3 times. The electrometer zero offset measured when the DMA voltage was off was subtracted from each measurement to reduce uncertainties due to electrometer drift.

3. RESULTS

3.1. Comparability of Electrometers

GRIMM and TSI electrometers showed similar concentrations when measuring in parallel (see section 2.3). The GRIMM electrometer measured 5% higher at 23 nm (concentration 8500 cm^{− 3}) and 1.5% higher at 41 nm (concentration 4500 cm^{− 3}). Hence, if the two companies calibrated the same CPC, the calibration factors would have < 5% difference. The 5% difference of the electrometers was considered acceptable as the uncertainty at these concentrations is 4–8% (noise 1 fA or 375 cm^{− 3}) (Sem 2002). In an earlier workshop conducted by TSI and AIST, the TSI electrometer was found to agree within 3% in the size range of 3.7–150 nm with the AIST electrometer (Wang et al. 2007).

3.2. Primary Method

The counting efficiencies and slopes of various materials for the three CPCs examined are presented in Table 1, Table 2, Table 3. These results were not corrected for multiply charged particles (i.e., Equation (S15) was used).

TABLE 1 Test TSI CPC 3790 calibration counting efficiencies according to the primary method without multiple charge correction (Supplemental information, Equation (S15)). Number in parenthesis after the material is the size (in nm) at which the linearity checks were conducted (for the last 3 columns). The fourth column gives the minimum and maximum ratios of the test CPC to electrometer concentrations. Numbers after ± indicate the CoV (ratio of the standard deviation to the mean value) of the ratio of the test CPC to the electrometer during the 2 min measurement

Material (nm)	23 nm	41 nm	> 50 nm	Slope	R ^2
NaCl (80)	—	—	0.856–0.919	0.894	0.9995
Tetracontane (70)	0.862(± 2.9%)	1.033(± 2.6%)	0.938–0.970	0.943	0.9999
CAST (60)	0.543(± 1.1%)	0.931(± 2.6%)	0.921–0.945	0.926	0.9999
Mini-CAST (50)	0.522(± 1.3%)	0.916(± 4.8%)	0.911–0.976	0.924	0.9981
Emery oil (55)	0.724(± 0.9%)	0.984(± 1.9%)	0.948–0.975	0.973	1.0000
Emery oil (repeated)	0.532(± 0.8%)	0.947(± 1.8%)	0.945–0.987	0.951	0.9998
Engine idle (120)	0.835(± 4.4%)	0.941(± 1.9%)	0.772–0.844	0.784	0.9984
Engine load (120)	0.573(± 2.3%)	0.919(± 3.2%)	0.874–1.058	0.885	0.9966

TABLE 2 Test GRIMM CPC 5.404 calibration counting efficiencies according to the primary method

Material (nm)	23 nm	41 nm	> 50 nm	Slope	R ^2
NaCl (50)	—	—	0.766–0.790	0.785	0.9997
Tetracontane (70)	0.810(± 5.7%)	0.935(± 5.6%)	0.908–0.996	0.913	0.9999
CAST (60)	0.599(± 6.9%)	0.911(± 2.8%)	0.905–0.945	0.919	0.9997
Mini-CAST (50)	0.566(± 5.1%)	0.865(± 3.4%)	0.887–0.944	0.936	0.9998
Emery oil (55)	0.726(± 5.9%)	0.954(± 3.1%)	0.949–0.963	0.954	0.9999
Emery oil (combined)	0.624(± 5.2%)	0.913(± 4.4%)	—	—	—
Engine CVS (70)	—	0.806(± 8.2%)	0.721–0.762	0.731	0.9996

TABLE 3 Reference GRIMM CPC 5.403 calibration counting efficiencies according to the primary method

Material (nm)	23 nm	41 nm	> 50 nm	Slope	R ^2
NaCl (50)	—	—	0.822–0.878	0.854	0.9994
Tetracontane (70)	0.946(± 5.6%)	0.965(± 5.4%)	0.914–0.985	0.960	0.9992
CAST (60)	0.968(± 6.3%)	0.964(± 2.8%)	0.945–0.969	0.951	0.9999
Mini-CAST (50)	0.905(± 4.2%)	0.946(± 3.3%)	0.963–1.000	0.986	0.9992
Emery oil (55)	0.952(± 5.6%)	0.976(± 3.1%)	0.960–1.026	1.007	0.9986
Engine CVS (70)	—	0.853(± 8.5%)	0.718–0.777	0.730	0.9980

TSI 3790

As shown in Table 1 the concentration ratios and the slopes were between 0.91 and 0.99 (requirement 0.9–1.1) except for the NaCl and engine cases. The R ² was greater than 0.996 (0.97 required) for all materials in the concentration range tested (up to 10000 cm^{− 3}). The slope of the TSI CPC 3790 according to the original calibration was 0.932 (with Emery oil). In this workshop, it was 0.95–0.97 (with Emery oil). This 0.03 difference of the slope (3%) indicates that a < 5% (reproducibility) uncertainty should be expected.

The 41 nm counting efficiency requirement (> 0.9) was met for all materials. The CoV of the tests was < 5% indicating the expected stability of the measurements. Similar uncertainty was reported by Liu et al. (2005) who calibrated 3010 and 3010D CPCs and by Hermann et al. (2007). The 23 nm counting efficiency of the TSI test CPC 3790 met the requirements (0.38–0.62) for diesel (load), CAST, and Emery oil (repeated) (Table 1 and Figure 2). The engine exhaust at idle mode had very high counting efficiency at 23 nm (0.83), but lower efficiency at 120 nm (0.78). This unexpected behaviour may be attributed to: (1) The size distribution at idle was bimodal, therefore a large fraction of particles were multiply charged when the DMA selected 23 nm or 41 nm particles; or (2) The number concentration during these measurements was 2500 cm^{− 3}, thus the uncertainty of the electrometer was 15%. Tetracontane particles had significantly higher counting efficiencies at 23 nm (0.86) than the other materials. These results will be further discussed in Section 3.4.

FIG. 2 Primary method counting efficiency of TSI CPC 3790 (according to Equation (S15), Supplemental information). Error bars give the counting efficiency corrected for doubly charged particles (Equation (S9)).

The Emery oil data taken with the standard experimental procedure had a higher efficiency (0.72) than earlier calibration data for the same unit (see calibration data in Figure 2). The counting efficiency at 41 nm was also higher compared to the Emery oil (repeated) and the factory calibration value. It was suspected that the isopropanol alcohol buffer solution used for electrospray had not evaporated completely before entering the CPC, resulting in higher counting efficiencies. A counting efficiency test was conducted at TSI after the workshop by passing the electrosprayed Emery oil particles through a diffusion dryer filled with activated charcoal (or bypassing the diffusion dryer) before the DMA. However, no difference in the CPC counting efficiency was observed with or without passing particles through the diffusion dryer. The high Emery oil efficiency in this test could be partly due to the uneven concentration split between the electrometer and CPC or due to electrometer drift during the duration of this particular test. However, the most probable explanation is that the Emery oil has higher efficiencies than the CAST particles at 23 nm. Comparison of the original Emery oil calibration data with the workshop results (0.13 difference with normal data and 0.05 with repeated data) gives an indication of the (reproducibility) uncertainty at the steep part of the curve. These uncertainties are similar to the difference Heim et al. (2004) observed at the counting efficiency measurements of a GRIMM CPC over a two years period (differences in counting efficiencies of ∼ 0.1 at the slope and much less at bigger diameters). High uncertainty (20%) of the slope of the counting efficiency was also reported by Hermann et al. (2007).

Figure 2 shows the counting efficiencies reported in Table 1 for the TSI 3790 (for diameters > 50 nm the slope is given). For the sake of clarity, error bars with measurement uncertainties, which have been reported in Table 1, are not displayed. However, error bars are given that show the efficiency calculated after correcting for doubly charged particles (Equation (S9) with measured f _{+ +}). This effect was important for the engine (load) (correction +0.075 for 41 and 120 nm) and NaCl cases (correction +0.055 for the 80 nm) and negligible for the rest (correction < 0.02). When this correction is applied to the NaCl and engine data, their slopes become similar to the slopes of the other materials, indicating the importance of this correction.

GRIMM 5.404

Similar results with the TSI test CPC were also found for the counting efficiencies of the test GRIMM model 5.404 CPC (Table 2). It is important to note the low slope with NaCl particles (size 50 nm). This has to do only partly with the multiply charged particles. Even if this correction is taken into account, the slope is less than with the other materials. The lower slope can be explained by the fact that the 50 nm size is still at the increasing part of the counting efficiency curve (Wang et al. 2007). This effect was not observed at the TSI 3790 in this workshop because the measurements were conducted with 80 nm particles. Thus, for linearity tests it is very important to choose a size that the counting efficiency has reached its maximum value. In Table 2, a test is named “Emery oil combined,” where the two companies measured in parallel. The “combined” counting efficiency at 23 nm was 0.62, while during the measurements with the standard procedure it was 0.72. These tests show that at the steep part of the counting efficiency curve the (repeatability) uncertainty is high (0.1 or 16%). The tests with this CPC also showed that the counting efficiency of Emery oil particles at 23 and 41 nm is the same or slightly higher (+0.1) than the counting efficiency of CAST particles. The diesel engine (CVS) tests had very high uncertainty (> 35%) due to the low concentration measured (1000 cm^{− 3}): thus, no sound conclusions can be drawn.

GRIMM 5.403

The GRIMM reference CPC had 23 nm efficiencies of 0.91–0.95 for all materials and 0.94–0.98 for 41 nm (except for the engine case, Table 3). The ratios and slopes of the linearity checks were 0.95–1.0 (lower for NaCl and engine) and the R ² values were > 0.998.

3.3. Secondary Method

According to the secondary method, the CPCs under evaluation are compared with a CPC calibrated with the primary method. The reference CPC used by GRIMM was CPC model 5.403 (d ₅₀ at 3 nm). No (counting efficiency) correction was applied to the reference CPC. This correction should have been ∼ 0.93, ∼ 0.96, and ∼ 0.99 (see Table 3) for the 23, 41, and > 50 nm cases (depending also on the material of the primary calibration of the reference CPC). The secondary calibration results are shown in Table 4. The 23 nm efficiencies were 0.62 for CAST particles but higher for Emery oil (0.76) and even higher for tetracontane (0.87). The 41 nm efficiencies were > 0.91 for all materials. Note that 41 and 70 nm diesel particles had high efficiencies (0.94) regardless of their low concentration (1000 cm^{− 3}). With the primary method the results were 0.72–0.80 (Table 2). The secondary method is much less susceptible to low concentrations than the primary method. The linearity slope for the CPC model 5.404 was ∼ 0.95. The gradient seemed to be material independent for diesel, CAST, tetracontane and Emery oil particles. The gradient for NaCl was ∼ 0.03 less mainly due to the effect of doubly charged particles. However, the effect was much lower compared to the primary method (where the doubly charged particles effect on the slope was > 0.1). This can be explained by the different effects of the doubly charged particles on the CPCs and the electrometers (Equation (S13) and (S14), see discussion of Figure S1b). The R ² was greater than 0.997 (0.97 required) for all materials in the concentration range tested (up to 10000 cm^{− 3}).

TABLE 4 Test GRIMM CPC model 5.404 calibration counting efficiencies according to the secondary method. No linearity (slope) correction for the reference CPC

Material (nm)	23 nm	41 nm	> 50 nm	Slope	R ^2
NaCl (50)	—	—	0.889–0.933	0.919	0.9999
Tetracontane (70)	0.867(± 5.5%)	0.970(± 5.2%)	0.940–1.011	0.951	0.9994
CAST (60)	0.618(± 4.4%)	0.945(± 2.2%)	0.948–0.975	0.960	0.9996
Mini-CAST (50)	0.619(± 3.9%)	0.915(± 2.5%)	0.891–0.968	0.950	0.9998
Emery oil (55)	0.762(± 2.9%)	0.977(± 2.7%)	0.927–1.003	0.947	0.9985
Engine CVS (70)	—	0.944(± 4.7%)	0.973–1.015	1.000	0.9992

Comparison of the results of the primary and secondary methods for the GRIMM test CPC model 5.404 shows that the slopes with the secondary method were slightly higher (∼ 0.02). However, if the slope of the reference CPC model 5.403 was taken into account then there would be no difference. The counting efficiencies at 23 nm and 41 nm with the secondary method were around 0.05 and 0.03 higher, respectively. This had to do with the 0.93 and 0.96 efficiency of the reference CPC at these diameters. These corrections should be taken into account for accurate results.

Legislation doesn't specify that the reference CPC should have efficiencies ∼ 1 at diameter > 23 nm. Thus, in principle, one CPC with a d ₅₀ around 23 nm could be used. Such a test was conducted by comparing the test CPC model 5.404 with a “reference” CPC with d ₅₀ at 23 nm. The ratio of the concentrations of the two CPCs was 1 ± 0.02 for all the sizes and materials tested (with one exception at 23 nm that had 0.08 difference). When using a reference CPC with d ₅₀ at 23 nm, its counting efficiencies must be known to calculate the counting efficiencies of the test CPC. However, the efficiency uncertainty near 23 nm is high. Small deviation can lead to large errors. Therefore, it is recommended that the reference CPC has counting efficiency ∼ 1 near 23 nm.

3.4. Validation at Labs

Figure 3 compares the results of the JRC workshop with earlier measurements at JRC for the same 3790 CPC with diesel engine particles from the same heavy duty engine at the same medium load and speed. Error bars show the counting efficiencies corrected for double charges. Note that multiple charging is important for 41 and 120 nm with the primary method. However, for the secondary method, although there is a considerable amount of doubly charged particles (10%), the effect is negligible due to the high counting efficiency (> 0.9) of the test CPC at 41 and 120 nm (see Supplemental information, Figure S1b). At 10 and 23 nm the doubly charged fraction is low (2.5%) so its effect is small. The two calibration results are in very good agreement, indicating that labs can do the validation tests (with the secondary method).

FIG. 3 Comparison of JRC workshop results with earlier measurements at JRC for the same CPC for heavy duty diesel engine particles (engine at medium load).

Figure 4 shows the tetracontane and NaCl efficiency data from the JRC workshop. Counting efficiencies of tetracontane and NaCl for another TSI CPC 3790 measured by the JRC lab with the secondary method are also plotted in Figure 4. The tetracontane and NaCl particles were generated by two different (but similar) evaporation-condensation generators. The NaCl counting efficiency was measured with and without passing the aerosol through a thermodenuder (at 300°C).

FIG. 4 Results from the JRC workshop TSI 3790 CPC (with the primary method, solid symbols) and from JRC lab with another TSI 3790 CPC (secondary method, open symbols) for NaCl (squares and triangles) and tetracontane (circles) particles.

Significant differences in counting efficiencies for thermally or non-thermally treated NaCl particles were observed with the later having higher counting efficiencies. Higher counting efficiencies of NaCl particles with relative humidity of 50% compared to relative humidities of 25% and 0% were also reported in Sem (2002). One possible explanation is that non-thermally treated NaCl particles contain water (NaCl is very hygroscopic) that could have condensed from the carrier air of the generator or the sheath air in the DMA. Since n-butanol is soluble in water, wet NaCl will have higher efficiencies than dry NaCl particles. Another possible explanation is the differences in particle shape and structure due to the thermal treatment of the aerosol. The particles generated by evaporation-condensation technique (and possibly the dry particles downstream of the thermodenuder) are highly agglomerated. After exposure to water vapour at relative humidity below the deliquescence threshold (75%) they become more compact with reduced shape factor (Krämer et al. 2000).

Larger differences were observed with the tetracontane particles, probably due to differences in their generation. In the JRC tests the displacement flow was filtered N₂, to avoid contamination from particles, and condensation took place at ambient temperature. In the JRC workshop the tetracontane particles were generated using air as the displacement flow that was not dried and condensation took place at < 5°C. It is possible that some water had condensed on the tetracontane particles, thereby increasing their counting efficiencies. In addition, impurities of the displacement flow might have also contributed to these higher efficiencies. Hering et al. (2005) found differences in the counting efficiencies of DOS particles in a water CPC due to contamination of NaCl. As extra solubility tests didn't show any significant solubility of tetracontane in n-butanol, the JRC lab results should be more representative of the tetracontane behavior.

Some extra tests were also conducted to investigate the effect of mixing and splitting of the flow (no figure shown). It was found that there must be a mixing orifice after the DMA, or at least the tube must be long enough (> 20 diameters) to ensure adequate mixing. The splitting is also important especially when the flows of the instruments are different. A Y-type splitter should be used. All the above tests show that although some results might be repeatable, they are not necessary accurate. Labs that use their own generators and calibration systems should make sure that they control humidity and impurities in their experimental set ups, and that the aerosol concentration is uniformly split to instruments.

4. SUMMARY AND CONCLUSIONS

Recently the particle number method was introduced in the Euro 5/6 light duty regulation. Calibration of Condensation Particle Counters (CPCs) includes linearity measurements (slope 0.9–1.1, R ² > 0.97) and counting efficiency measurements (at 23 nm it should be 0.38–0.62, at 41 nm > 0.9). Labs will have to demonstrate compliance of their counters with a traceable standard within a 12-month period prior to the vehicle emissions test. Compliance can be demonstrated by either the primary (comparison of the test CPC with an electrometer) or secondary method (comparison with a CPC already calibrated with the primary method). Proper calibration of CPCs is necessary as the concentration linearity slope will have to be taken into account for the calculations of particle number emissions.

A workshop was organised at the European Commission's Joint Research Centre (JRC) to investigate the effect of the aerosol material (chemical composition) and the uncertainties of the calibration procedures. GRIMM and TSI provided CPCs, while AEA, Matter Eng., GRIMM and TSI provided particle generators (evaporation-condensation, CAST, electrospray). Heavy duty diesel engine (w/o aftertreatment) particles were also produced at idle and at a medium load modes. The data were evaluated by JRC. The main conclusions are discussed in the following sections.

Multiply Charged Particles Effect

Multi-charge effect is important when f _{+ +} (ratio of doubly to singly charged particles) is > 5%. This translates to size distributions with σ > 1.3 and measurements of large particles (> 50 nm), assuming that the selected size is on the right side of the mean of the size distribution.

Primary Method

For the linearity tests the material or size (> 50 nm) was small (with f _{+ +} < 3%) (i.e., slopes were within ± 0.02). A high multi-charge effect (f _{+ +}∼ 6%, polydisperse size distribution at DMA inlet with mean 60 nm and σ = 1.82) led to lower slope estimation for NaCl and engine particles (0.05 less than the true one). An underestimation of the slope was also observed with 50 nm NaCl particles because the counting efficiency at this size was not maximum (i.e., this size was still at the increasing part of the counting efficiency curve). The R ² was > 0.997 for all cases.

The counting efficiency at the steep part of the curve (23 nm) was found to depend significantly on the chemical composition of the aerosol particles. Tetracontane particles showed the highest efficiencies (0.86); CAST particles had similar efficiencies (0.53) with non-volatile diesel engine soot (0.57); Emery oil particles had also similar efficiencies (0.53) or slightly higher (0.72). The uncertainty of the measurements (expressed as difference between two measurements on different days or from different labs) was generally 0.05 for counting efficiencies > 0.9 and 0.1 for counting efficiencies at the steep part of the counting efficiency curve. Thus, manufacturers should aim for a slope around unity and a counting efficiency at 23 nm of 0.5 to have best performance safety margin. Labs that check their CPCs should use the same material with the one that the CPC had been calibrated. If a different material is used its effect on the counting efficiencies should be taken into account.

Secondary Method

Generally, similar counting efficiencies, slopes and R ² were found with the secondary method as with the primary method when using a reference CPC with a low cut-off size (3 nm). If the counting efficiencies of the reference CPC were taken into account, the primary and secondary methods were equivalent. With the secondary method the multi-charge effect was much smaller, as it was partially cancelled out from the two CPCs. The secondary method is highly recommended for labs that want to verify the proper operation of their CPCs, provided that they have a reliable reference CPC.

Validation at Labs

Experiments at the JRC after the workshop measured counting efficiencies of thermally treated (through a thermodenuder at 300°C) NaCl particles that were almost 0.2 lower than the efficiencies of non-thermally treated NaCl particles. Tetracontane particles were found to have lower counting efficiencies than the workshop results (almost 0 at 23 nm). However, experiments with diesel aerosol measured efficiencies similar to the workshop results. Thus, labs should be extremely careful in their aerosol generation method. Special attention should be paid to humidity and impurities, as they were found to influence considerably the measured counting efficiencies.

Acknowledgments

[Supplementary materials are available for this article. Go to the publisher's online edition of Aerosol Science and Technology to view the free supplementary files.]

The authors thank L. Keck for his helpful comments. The assistance of M. Carriero, S. Alessandrini, F. Forni, H. Jörgl, and G. Winkler during the tests is also highly appreciated.

Notes

*Current address: Desert Research Institute, 2215 Raggio Pkwy., Reno, NV 89512, USA.