Mamba state space model and transformer model calibrated Raman spectroscopy for real-time multi-objective monitoring in ethanol precipitation process of Guanxinning tablets

Ethanol precipitation is a pivotal manufacturing step in the production of Guanxinning tablets (GXN). However, current quality control methods for this process remain inadequate. To address this gap, we propose a novel analytical framework that combines Raman spectroscopy with state-of-the-art deep learning architectures, specifically integrating a Mamba State Space model with a Transformer model, to enable non-destructive and real-time monitoring of seven critical quality attributes (CQAs) throughout the ethanol precipitation process of GXN. A total of 160 ethanol precipitation samples were collected, and their Raman spectra were acquired using a 785 nm laser (35 mW) across a wavenumber range of 200–3200 cm −1 , enabling rapid spectral acquisition within 1 s per sample. The content of five bioactive compounds (danshensu, ligustrazine, ferulic acid, rosmarinic acid, and salvianolic acid B) was quantified via HPLC, while solid content and alcohol content were determined using standard methods. Spectral outliers were removed using PCA-Mahalanobis distance analysis, and the Kennard-Stone algorithm was employed to partition datasets into training and test sets (4:1 ratio) to ensure model generalizability. Comparative model evaluation revealed that Mamba State Space model outperformed Transformer model and linear models, achieving prediction coefficients ( R p 2 ) of 0.9419–0.9978 and root mean square error of prediction (RMSEP) values of 0.0129–0.0501, with exceptional accuracy for solid content ( R p 2  > 0.99). This approach reduces reliance on time-consuming HPLC analytical method, offering a 99.9 % reduction in analysis time while maintaining predictive accuracy (RMSEP ≤0.0501) and robustness ( R p 2  > 0.94). By embedding the Mamba State Space model into Raman spectroscopy-based ethanol precipitation process monitoring, this study establishes a robust quality by artificial intelligence framework for real-time quality control in traditional Chinese medicine (TCM) manufacturing, paving the way for smart process analytical technology (PAT) applications.