Particle size
The spread often matters more than the mean.
D50 tells you the middle of a powder; D10 and D90 tell you the truth. Processing behavior, packing, coating, calendering, is set by the spread and the tails, which is why two powders with identical medians can run completely differently.
What it measures
A particle-size distribution is the full population, summarized by its cumulative percentiles:
- D10, D50, D90: the sizes below which 10%, 50%, and 90% of the population falls. D50 is the median; the pair D10/D90 bounds the working spread.
- Span: (D90 − D10)/D50, the standard width metric. Narrow span packs predictably; wide span segregates, settles, and clogs.
- The model: powder populations from grinding and precipitation are typically log-normal, so fits and comparisons should happen in log space; bimodal distributions (deliberate blends, or accidental fines) need two components, not one forced average.
The same statistics come from different measurements, laser diffraction, or counted populations from SEM/TEM image analysis, and they do not agree by default: one is volume-weighted, the other number-weighted, and the weighting changes the percentiles substantially.
How to read the output
Read the shape before the numbers. A clean single mode that fits log-normal is one population; a shoulder on the fine side is a second population, attrition fines, broken secondary particles, that a D50 will never reveal. Then read the tails against the process: the D90 end interacts with coating-gap and calendering clearances (one oversize particle scratches a kilometer of electrode), while the D10 end sets surface area, binder demand, and slurry rheology. Track the ratio D90/D10 across lots; drift there with stable D50 is the classic silent change.
A real use case
A coating line starts logging intermittent streak defects. The cathode powder lot in use passes incoming QC, D50 is dead on spec, because the spec only lists D50. The full distributions tell a different story: the defect-era lot carries a high-side tail, D90 up roughly a third versus reference lots, putting its largest particles at the coating-gap clearance. The streaks correlate with that tail, the spec gains D90 and span limits, and the supplier’s classifier, not the coater, turns out to be the process that changed.
Common mistakes
- Specifying and trending only D50. Most particle-driven process failures live in the tails the median cannot see.
- Comparing number-weighted (microscopy) and volume-weighted (laser diffraction) distributions as if they were the same quantity.
- Fitting one log-normal to a bimodal population, the fit converges and the parameters mean nothing.
- Counting too few particles in image-based PSD; the D90 estimate is exactly where small samples are least reliable.
- Ignoring dispersion quality: agglomerates measured as single particles inflate the coarse tail and then vanish when the sonication changes.
Distributions, percentiles, and lot-to-lot drift
Niobia builds particle-size distributions from both routes: segmentation and counting on SEM images, histograms annotated with mean and standard deviation, and Gaussian or log-normal fitting of nanoparticle populations from TEM, including aspect-ratio characterization for anisotropic particles. From any distribution it reports the percentile set (D10/D50/D90, span) and overlays lots for comparison, so a tail shift like the D90 drift above shows up as a flagged difference between curves, not a surprise on the coater. When the data source is number-weighted it says so, keeping the comparison against volume-weighted lab data honest.
Frequently asked
Why do my laser diffraction and SEM-based sizes disagree?
Different weighting. Laser diffraction reports volume-weighted percentiles, where one 50 µm particle outweighs thousands of 1 µm fines; image counting is number-weighted, where the fines dominate. Both are correct; convert or compare like-for-like.
What does span actually predict?
Packing and flow behavior. Narrow-span powders pack reproducibly and coat smoothly; wide-span powders segregate in handling, pack densely but variably, and drag oversize particles into clearance-limited steps like coating and calendering.
Is a bimodal distribution always a problem?
No, deliberate coarse/fine blends are bimodal by design to raise packing density. The problem is unplanned bimodality: attrition fines or broken secondary particles appearing as a new mode that the slurry and binder system was never formulated for.