HIGH-SPEED PUNCHING DEFECTS: FORECASTING AND MITIGATING TABLET FRIABILITY BY USING AN AI-DRIVEN MODEL IN MANUFACTURING AND INTERPRETING THE TABLETING FORMULATION

Abstract

Background: Dwell time in a high speed rotary press (20-80 rpm) is decreased which leads to increased friability, capping and lamination. The conventional testing is destructive, and is not suitable for time-on-line monitoring.

Objective: High-speed punching defect on friability and identification and reduction of defects using AI.

Method: method used for preparation of granules was the wet granulation method and granules were compressed on Korsch XL 100 at compression forces between 5 kN and 25 kN and at press speeds between 20 rpm and 80 rpm (dwell time 19-193 ms). Friability, hardness and capping were tested. Five ML models (GPR, XGBoost, Random Forest, SVR, ANN) were trained using a dataset of 2500 observations to evaluate moisture, D50 and lubricant fraction as well as compression force, while the SHAP analysis was used for this.

Results: An increase in the speed of turrets from 20 to 80 rpm caused the friability to increase from 0.32% to 0.97% and the defects increased from 1.2% to 13% at 12kN. Compression force showed biphasic effect with optimum at 10-12 kN. GPR showed highest accuracy (R²=0.949, RMSE=0.081) followed by XGBoost (R²=0.941). The capping classification method had 97.8% accuracy. The reduced friability was 0.28% and capping was 2.1% with validation at optimized parameters, which is 68.4% defect reduction.

Conclusion: The use of AI in a model offers a real-time, non-destructive solution for quality control, and aids QbD implementation.