How far can routine process records predict final turbidity in industrial beer crossflow filtration? A capability-boundary study on 5000 production batches

Abstract Crossflow filtration is the last separation step before bright beer, and final turbidity is its key quality attribute; yet the most recent systematic review of 188 brewing-AI studies shows that filtration remains largely unstudied. This work presents a systematic capability assessment of turbidity prediction in industrial beer crossflow filtration, based on 5000 production batches with 37 recorded variables (35 features: 17 tank-1, 17 tank-2 and a dual-tank indicator, plus batch number and turbidity). Ten regression model families were evaluated under two feature sets with 5-fold cross-validation, supported by learning-curve analysis, dose–response testing and permutation importance. The best models (PLS and Ridge, 35 features) reached an RMSE of 1.404 NTU and R² = +0.084, only marginally above the mean baseline, and no non-linear model outperformed the linear ones. The learning curve plateaus at about 2000 batches with a train–test gap below 0.02, showing that the ceiling is a property of the recorded features rather than of sample size. The strongest single predictor was total recovered yeast from tank 1 (Spearman ρ = −0.128), with a strictly monotonic dose–response across quintiles (2.016 → 1.505 NTU; Jonckheere–Terpstra p < 10 − 15 ), in the direction expected from a particle mass balance. Exceedance classification reached AUC 0.64–0.69 but F1 collapsed at high thresholds, supporting weak ranking only. The recorded process variables therefore carry at most 8–10% of the variance in final turbidity; usable soft sensing at this station requires additional measurement points, in particular pre-filtration online turbidity, transmembrane pressure and flux, and filtration duration and volume.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-17
DOI
https://doi.org/10.21203/rs.3.rs-11487956/v1
Primary Topic
Fault Detection and Control Systems
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

How far can routine process records predict final turbidity in industrial beer crossflow filtration? A capability-boundary study on 5000 production batches

Yuyang Xia, Han Dong, Gongquan Weng, Peide Yang et al.
Research Square
Fault Detection and Control Systems
preprint

How far can routine process records predict final turbidity in industrial beer crossflow filtration? A capability-boundary study on 5000 production batches

Yuyang Xia, Han Dong, Gongquan Weng, Peide Yang, Chongyi Zhu, Songdi Lin
preprint en

Abstract

Abstract Crossflow filtration is the last separation step before bright beer, and final turbidity is its key quality attribute; yet the most recent systematic review of 188 brewing-AI studies shows that filtration remains largely unstudied. This work presents a systematic capability assessment of turbidity prediction in industrial beer crossflow filtration, based on 5000 production batches with 37 recorded variables (35 features: 17 tank-1, 17 tank-2 and a dual-tank indicator, plus batch number and turbidity). Ten regression model families were evaluated under two feature sets with 5-fold cross-validation, supported by learning-curve analysis, dose–response testing and permutation importance. The best models (PLS and Ridge, 35 features) reached an RMSE of 1.404 NTU and R² = +0.084, only marginally above the mean baseline, and no non-linear model outperformed the linear ones. The learning curve plateaus at about 2000 batches with a train–test gap below 0.02, showing that the ceiling is a property of the recorded features rather than of sample size. The strongest single predictor was total recovered yeast from tank 1 (Spearman ρ = −0.128), with a strictly monotonic dose–response across quintiles (2.016 → 1.505 NTU; Jonckheere–Terpstra p < 10 − 15 ), in the direction expected from a particle mass balance. Exceedance classification reached AUC 0.64–0.69 but F1 collapsed at high thresholds, supporting weak ranking only. The recorded process variables therefore carry at most 8–10% of the variance in final turbidity; usable soft sensing at this station requires additional measurement points, in particular pre-filtration online turbidity, transmembrane pressure and flux, and filtration duration and volume.

Research Square
Xiamen University Tan Kah Kee College (CN)
Clean water and sanitation
Fault Detection and Control Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon