Development of an Intelligent Tablet Press Machine for the In-Line Detection of Defective Tablets Using Machine Learning and Deep Learning Models.

This study aims to develop a tablet press machine (TPM) integrated with machine learning (ML) and deep learning (DL) models for in-line detection of defective tablets as a Process Analytical Technology (PAT) tool. This study aimed to predict tablet defects, including capping occurrence and inappropriate tablet breaking force (TBF), using real-time processing data. Free-flowing metformin HCl (MF) granules produced using the granulation method were compressed into tablets using a TPM. Commercial-scale experiments were conducted to determine the MF tablets' defect criteria. Random Forest (RF) and Artificial Neural Network (ANN) models were designed and trained using sensed in-line data, including compression force, ejection force, and compression speed, to predict tablet quality defects. Subsequently, the TPM was designed and manufactured for in-line PAT using an RF model. The TPM was verified by sorting defective tablets in-line using a pretrained defect-detection algorithm. The RF model demonstrated the highest predictive accuracy at 93.7% with an Area Under the Curve (AUC) of 0.895, while the ANN model achieved an accuracy of 92.6% with an AUC of 0.878. The TPM successfully sorted defective tablets in real time, achieving 99.43% sorting accuracy and a defective tablet detection accuracy of 93.71%. These results suggest that a ML-based TPM applied during the tableting process can detect defects non-destructively during the scale-up of wet granulation. In particular, it can serve as the base TPM model for an in-line PAT process during a scale-up process that produces small batches of multiple products, thereby reducing additional labor, time, and API consumption, and decreasing environmental pollution.