Does a two-height BDST line predict the third height?

A frozen check across six media, three concentrations, and two usable flow conditions. The line is fitted on 7.5 cm and 10 cm service times, then used to predict 12.5 cm before that observation is looked at.

This is a public 2019 dataset, analysed as an external check. The data is transcribed from one licensed article artifact, and the analysis is frozen against its SHA-256. It checks service time at one held-out bed height. It does not validate a breakthrough curve, a mass-transfer zone, or scale-up capacity.

Verdict

HELD-OUT ERROR STAYS BELOW 20 PERCENT

The independent two-height line predicts all 36 observed 12.5 cm service times within 20 percent. The median error is 7.9 percent, and 26 of 36 checks are within 10 percent.

Read that against what is missing. The result covers 36 of 54 possible strata, and nothing is imputed for the rest.

Training fit is not validation

Two points define a line exactly. This study fits 7.5 cm and 10 cm service times.

It predicts 12.5 cm before comparing with the held-out observation. Each medium, concentration, and flow condition stays separate.

Median error 7.9% absolute relative, held out

Largest error 19.3% worst held-out stratum

Valid strata 36 / 54 18 excluded, none imputed

Within 10% 26 of 36 held-out checks

All 36 held-out errors remain below 20 percent

The independent two-height line predicts all 36 observed 12.5 cm service times within 20 percent. The median error is 7.9 percent. 26 of 36 checks are within 10 percent.

Observed and independently predicted held-out service times on logarithmic axes. — Each dot is one held-out 12.5 cm observation.

The result is not driven by one material

Absolute relative held-out errors grouped by filter medium. — Six valid checks remain for each medium. The upper dashed line is 20 percent.

Medium	n	Median	Maximum
RS	6	8.8%	19.3%
RSMN	6	10.2%	14.3%
GS1	6	6.8%	9.8%
GS2	6	5.4%	16.0%
GS1MN	6	8.5%	14.7%
GS2MN	6	8.4%	10.9%

Errors are absolute relative errors at the held-out height. The median absolute error is 9.45 min, but service times span orders of magnitude.

18 strata are not validation cases

The source prints slowest-flow predictions at 12.5 cm. It does not print the matching observations.

No value is imputed. The source omits slowest-flow held-out observations. The result covers 36 of 54 possible strata and only the 5 percent threshold.

Velocity and diameter disagree

Printed velocities imply a column diameter near 2.86 cm, not the stated 2.3 cm.

Do not report recovered N0 as a physical capacity. The page reports service-time error only.

A frozen, narrow protocol

Parse Tables 1 through 3 from the frozen XML artifact.
Keep each medium, concentration, and printed velocity separate.
Apply T_12.5 = 2T₁₀ − T_7.5 at C_t/C₀ = 0.05.
Compare with the 12.5 cm observation. Exclude missing observations.

What this does not show

This checks service time at one held-out bed height. It does not validate a full breakthrough curve, a mass-transfer zone, or scale-up capacity.

The article does not state that its predictions use the same two-height calibration. They remain a separate comparison. The source has 27 of 36 predictions within 10 percent.

No adjustable control is shipped. Changing the threshold or training heights would change the frozen validation question.

Reproducible from one licensed article artifact

Article DOI	10.1016/j.dib.2019.104751
Archive record	PMC6880139
Licence	CC BY 4.0
Access date	2026-04-22
XML SHA-256	`794d28a8dbe8c7d86b24e9fc83bf327d8d44823aec1aa3fb0416b66159b3bc34`