TL;DR
This paper introduces a novel Monte Carlo-based framework to quantify the uncertainty in airborne particle size distributions measured by Scanning Mobility Particle Sizers (SMPS). By meticulously modeling various sources of uncertainty, including experimental dispersion and physical approximations, the authors provide a more robust estimation of measurement uncertainty than traditional methods.
What Problem Does This Paper Address?
Accurately measuring the size distribution of airborne particles is crucial in many fields, from air quality monitoring to nanotechnology. Scanning Mobility Particle Sizers (SMPS) are widely used for this purpose, but their measurements inherently carry uncertainty. Traditional methods often simplify or neglect certain physical processes and experimental variations, leading to an underestimation of this uncertainty. This paper tackles the challenge of comprehensively accounting for these uncertainties, particularly those arising from the complex physics governing particle behavior within the SMPS and the inherent limitations of the measurement model itself.
Methods
The core of this work is the application of a Monte Carlo (MC) method for uncertainty propagation. This approach is implemented in a two-stage process:
- Modeling Experimental Dispersion: This stage accounts for the random variations in measured particle counts, either by modeling them as a Poisson distribution (for single scans) or a multivariate Gaussian distribution (for multiple scans).
- Handling Modeling Errors: This is the more novel aspect. Instead of using fixed models for physical processes, the authors randomly draw parameters and functions from their estimated distributions. This includes:
- Bipolar Charging Law: Modeled using a multivariate Gaussian distribution to capture variations from different theoretical approaches.
- Slip Correction Factors: Modeled using truncated normal distributions.
- DMA Transfer Function: A combination of convective and convective-diffusive models is used, with a random threshold to account for transition uncertainties.
- Other Sources: Various other parameters like viscosity, temperature, pressure, and flow rates are also modeled with appropriate statistical distributions.
The MC simulations then propagate these uncertainties through the entire measurement model to estimate the uncertainty in the particle size distribution (PSD).
Datasets
The methodology was tested using Standard Reference Materials (SRMs), specifically monosize polystyrene latex spheres (PSL) of 100 nm, 200 nm, and 450 nm. For the 200 nm PSL, experimental data was collected. For the 100 nm and 450 nm PSL, "TSI free data sets" were used, which are obtained by downloading data from the AIM (Aerosol Instrument Manager) software. This approach allowed for comparison with existing methods and evaluation across different particle sizes and data sources.
Key Insights and Ideas
- Modeling Errors Matter: The study strongly emphasizes that neglecting the uncertainty associated with the underlying physical models (e.g., bipolar charging, slip correction) can lead to a significant underestimation of the overall measurement uncertainty.
- Conservative but Realistic: The statistical models used are described as potentially "conservative" because they account for a wide range of existing theories and experimental variations. However, this conservatism is presented as a strength, aiming for a more realistic representation of uncertainty.
- Median and Mode as Robust Indicators: The median and mode diameters were found to be less sensitive to fluctuations compared to the mean and geometric mean diameters, providing a more stable measure of central tendency.
- Repeatability vs. Full Uncertainty: The study highlights that the uncertainty derived from the full uncertainty analysis (including modeling errors) is consistently larger than that obtained from repeatability analysis alone, underscoring the importance of accounting for all sources.
Key Results
The paper demonstrates that the SMPS is well-suited for accurate sizing, achieving very low uncertainties for particle mobility diameter. For example, the median particle sizes had relative standard uncertainties of:
- ±0.17% for 100 nm PSL
- ±0.83% for 200 nm PSL
- ±0.34% for 450 nm PSL
However, the uncertainty in the peak particle number concentration was found to be significantly larger, reaching ±40% for the 200 nm PSL. This is attributed to the repeatability of particle count measurements and the conservative modeling of the bipolar charging law. The authors also showed that their method's estimates of PSD were consistent with those from AIM software, though their uncertainty estimates were generally broader.
Limitations and Future Work
- Unmeasured Quantities: The study acknowledges that some quantities (e.g., aerosol generation, particle drying) are not measured and thus not included in the uncertainty propagation.
- Model Updates: The authors recommend updating the statistical models with new experimental data or publications to improve accuracy and reduce conservatism.
- Further Sensitivity Analysis: The paper suggests that a global sensitivity analysis is needed to clearly identify the main contributors to uncertainty and to achieve a more refined numerical evaluation of their respective contributions.
- Ionic Population Characterization: A more accurate characterization of the ionic population could significantly reduce particle number concentration uncertainty.
Summary updated