Mass Measurement with the DMS500

Transcription

1 Mass Measurement with the DMS500 Jon Symonds & Kingsley Reavell

2 Contents Mass Calculation from Spectral Instruments why traditional problems Solutions lognormal mode identification particle density variation size calibration Lognormal parameterisation of DMS data in real-time Method of Mass Calculation with DMS Validation of Mass Measurement

3 Why Mass Measurement from Spectral Data? Why mass: Gravimetric measurements are the basis of current legislation and historical environmental data. With differences in particle sizing standards (DMA mobility classification / aerodynamic sizing / DMS electrical mobility classification), particle mass is unambiguously defined. Why from a spectral instrument: Improved sensitivity vs gravimetric methods Real-time data Measurement and differentiation of accumulation mode for unfiltered / partial filtered vehicles and DPF efficiency along with nucleation mode comprising the majority of mass emissions from DPF equipped vehicles.

4 Mass calculation from number:size data Measurement of concentration and size allows calculation of particulate mass. Mass Weighting Size:Number Spectra 2.00E E E E E E-03 Number concentration dn/dlogd p /cc 1.40E E E E E E E E E E-03 mass concentration dug/dlogd p /cc 4.00E E E E E E dp (nm)

5 Difficulties with Mass Calculation Sensitivity to calibration If mass ~ diameter 3, then 10% size error 30% mass error solved with traceable calibration via PSL spheres Noise in large particle sizes amplified by mass weighting Effect of spectral broadening Particle density effective density varies with particle size different for nucleation & accumulation modes diameter not well defined for agglomerates

6 Effect of mass weighting on noise Mass Weighted number:size spectra 2.50E+02 Volume concentration spectral density dv/dlogd p 2.00E E E E+01 Noise 0.00E Particle diameter, nm

7 Effect of Spectral Broadening Reweighting of broadened spectrum True total number = 1 CMD 60nm gsd = 1.5 true number spectrum resolved number spectrum true mass spectrum mass weighted resolved spectrum Measured mass median diameter = 116nm gsd = 1.6 total mass = 5.8e5 1.2E E E+05 dn/dlogdp Measured: number, CMD correct gsd = 1.6 True mass median diameter = 98nm gsd = 1.5 total mass = 4.5e5 6.0E+05 dm/dlogdp 4.0E E E Dp nm

8 Variable Particle Density Nucleation mode & accumulation obviously different. Accumulation variation demonstrated by various studies (DMA/APM, DMA/ELPI etc) for example, APM data (Park, Kittelson, McMurray): Accumulation mode density 1.4 density g/cc particle dia nm Environ Sci. Technology (2003)

9 Solution: Lognormal mode identification DMS normally produces up to 41 channels of size data Mode identification software can reduce this to just 6 for engine aerosol: Nucleation Accumulation GMD GSD N/cc GMD GSD conc = 1 ( logcmd log D ) df exp 2 2π logσ 2log g σ g p 2 dlog D p The nucleation and accumulation modes can then be considered separately. Factoring in air mass flow (and λ) gives total mass Software identifies if a mode is real relative to the noise base of the instrument, and if not discard it Reduces noise related error in mass estimation, especially in tail of accumulation mode Removes cross sensitivity introduced by crude size cut-off methods Improves spectral resolution

10 Lognormal Algorithm Bayes Theorem: Posterior Probability Prior Probability Likelihood P(w J) P(w) P(J w) Prior: Assumptions about size and width of aerosol modes, gives soft limits on fit Likelihood: Probability of the data (J, ring currents) given some lognormal parameters (w). Posterior: The fit criterion, to be maximised To fit: 1. Guess set of parameters and generate lognormal spectrum 2. Get ring currents for spectrum from transfer function 3. Compare these with measured ring currents, calculate likelihood (next slide) 4. Calculate Posterior from Likelihood and preset Prior 5. Iterate 1 4 to maximise Posterior

11 Likelihood Calculation Noise on electrometer rings measured during zero gives expected standard deviation (σ) of guessed currents (i) from measured currents (j) Likelihood per ring: P( j i, σ ) 1 ( j i) exp σ 2π 2σ = 2 2 Assume independent probabilities, total Likelihood for 22 rings: 22 1 ( j P( J w) = exp σ k 2π k = 1 k i k 2 k 2σ ( w)) 2

12 Example Diesel Engine Prior Probability GSD (width)

13 Let P(H m J) be the probability that there are truly m Lognormal modes, given the measured ring currents J. Then by Bayes theorem: (2) ) ( ), ( ), ( ) ( ), ( ), ( Product Rule : (1) )d, ( ) ( )d, ( ) ( Marginalisation rule : ) ( ) ( ) ( m m m m m m m m m H P H P H P I X P I X Y P I X Y P H P H P Y I Y X P I X P H P H P H P w w J w J w w J J J J = = = = L We assume that the prior probability of m modes, P(H m ), is uniform Equation (2) is now just the product of our original likelihood and prior, but now evaluated for a specific number of lognormals, i.e. the product of the likelihood of the overall spectrum and all the individual priors. (2) is substituted into (1) and evaluated. This takes considerable PC power in real-time! How many modes?

14 Automatic mode identification on NEDC Particle Sizes in Nucleation and Accumulation Modes Nucleation Mode Accumulation Mode Road Speed / kph Particle GMD / nm Road Speed / kph Time / s

15 Real-time Implementation Integrated into instrument interface Modes displayed in real-time along with conventional spectrum Summary parameters (size, concentration, σ g ) logged and available as analogue output signals Mass calculated and displayed, logged or available as analogue output.

16 Lognormal Data Processing: Reduced Noise Noise in DMS500 Particle Mass Concentration Response to 60 nm, σ g =1.9 Aerosol 1.E+04 Standard Lognormal 1.E+03 Abs error / μg / m 3 1.E+02 1.E+01 1.E+00 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Mass / μg / m 3

17 Lognormal Data Processing: Better Resolution dn/dlogdp /cc 299nm PSL Spheres - Standard DMS500 Spectral Output 1.20E E+06 Standard Lognormal 8.00E E E+05 SURFACTANT PEAK PSL PEAK 2.00E E Dp (nm) (Different Priors used to those for Diesel Engines!)

18 Density Function - SMPS Based upon work by Park, Cao, Kittleson and McMurry Accumulation mode mass Related DMA diameter to mass for heavy duty diesel via an effective density by cutting with Aerosol Particle Mass Analyser (APM) Recently shown to also apply to light duty diesel 4E E-18 load 0.1 y = 6.05E-24x 2.34E+00 mass kg 3E E-18 2E E y = 3.97E-24x 2.46E+00 y = 6.57E-24x 2.31E+00 y = 5.61E-24x 2.34E+00 1E-18 5E particle dia nm Environ Sci. Technology (2003) J.S. Olfert et al. in preparation

19 Accumulation Mode Size What size is this? DMS Calibrated with PSL Spheres Compare SMPS (mobility) sizing with DMS (electrical mobility) sizing for Diesel Agglomerates Test apparatus Vehicle (steady state) DPF Feedgas DMS 4:1 Dilution Air Sample Head HEPA Flow DMS 500 Filter Meter 7 slpm Heated 1 Atm 1 slpm 3080 DMA 8 slpm HR diluter OFF

20 DMA:DMS Diameter Relationship Diesel Agglomerates kph 4th Gear 70 kph 5th Gear 140 DMS Mean Diameter DMA Mean Diameter

21 New DMS Mass Weighting Rule DMA Diameter to Mass Relationship 4E E-18 3E E-18 Load y = 6.05E-24x 2.34E+00 Mass / kg 2E E-18 1E-18 5E DMA Dp / nm

22 New DMS Mass Weighting Rule DMS Diameter to Mass Relationship 4E E-18 Mass / Kg 3E E-18 2E E-18 Load y = 1.53E-25x 3.19E+00 1E-18 5E DMS Dp / nm

23 Mass Estimation from DMS500 DMS Calibration for Spheres Engine Aerosol Instrument Transfer Function DMS 500 Electrometer Currents Density as Function of DMA Measured Diameter (literature) DMS Response to DMA Cut Engine Aerosol Density as Function of DMS Measured Diameter Lognormal Fit of Accumulation Mode Mass Concentration Weighting Air Intake Mass Flow Exhaust Volume Flow Mass Integration λ MASS

24 Summary DMS500 Mass Calculation Lognormal mode identification reduces large particle noise by data redundancy discriminates between nucleation & accumulation modes to calculate soot mass, or apply different density laws. takes into account spectral blurring in instrument to correctly resolve mode widths. reduces amount of data to be handled by users allows nucleation mode to be measured and ignored at will, from one test. Mass weighting of particulate modes mass : diameter relationship derived from DMA literature, but modified due to different charge : drag ratio sizing basis.

25 Validating Mass Measurements Correlation is very easy to demonstrate: most instruments (not CPC!) are linear with increasing concentration, so just test at a variety of concentrations with similar aerosols even easier, test over different periods of emission. Neither of these give useful validation must ensure a variety of particle sizes in validation data set data presented here covers peaks between about 60nm and 140nm mobility diameter for different steady state conditions. confidence improved by agreement with a priori calculation, rather than just a posterior correlation

26 DPF Mass Correlation Validation Production vehicle (2.2l common rail) with DPF on chassis rolls and 2.0l common rail engine with prototype calibration on engine dyno. Measurements taken both by DPF weighing and filter paper measurement. Steady state test points selected for validation over maximum particulate size range. NEDC cycle mass for real application validation DMS500 sample drawn from pre-dpf flow via built-in dilution system and heated line. Mass flow calculated from intake MAF and exhaust lambda logged to DMS500 auxiliary analogue inputs.

27 Test Points NEDCs (sometimes inc EUDCs), warm and cold start 100 kph (5 th gear), 70 (4 th /5 th ), 32 (2 nd ) cruises , 650, 700 nm Fixed speed/load points on engine dyno DPF weighings (DMS primary + secondary dilution, heated line) Filter Paper in CVS Tunnel (DMS secondary dilution)

28 DMS500 Built-in 2 Stage Dilution Short unheated sample line Heated sample line Engine exhaust Cyclone separator (< 1um) Primary Dilution (Hot, dry, HEPA filtered) ~ 4 air:1 exhaust Heated restrictor Conductive heated sample line (~5m, <200C) Secondary (Rotating disc) diluter ~200:1 DMS500 Vacuum pump

29 Vehicle Tests: Mass Conc. over NEDC Real Time Accumulation Mode Mass Estimation Mass / (g/s) Road Speed / kph Mass / (g/s) Road Speed / kph Time / s

30 DMS Mass Test Results Filter Paper & DPF : DMS correlation : index = Dyno engine: prototype cal DPF weights DMS measurements within 10% of DPF mass measurement 0.1 Tunnel or Exhaust Mass Conc by DMS / μg/cc 0.01 CVS filter paper measurements Production vehicle DPF weights DMS underestimates filter paper mass: presumed due to adsorption of gas phase materials on filter Tunnel Mass Conc by Filter or Exhaust Mass Conc by DPF / μg / cc

31 Conclusions Mass calculation from spectral data allows correlation with gravimetric measurements and standards. Issues with mass calculation accuracy include size calibration, large particle noise floor and variable density. Mode identification reduces noise floor, improves size distribution accuracy and allows discrimination between nucleation and accumulation modes Variable particle density model modified for DMS size response Mass measurement validated on two drivetrains, against DPF and filter papers weights. Agreement with DPF soot mass within 10% Approximate 10% underestimate compared with filter paper weighings: assumed due to gas phase adsorption onto filter paper.