.png){kind=icon}
.png){kind=icon}
Return
Modeling Continuous Direct Compression processes with inherent delay using SINDYc and Bayesian inference
.png){kind=icon}
PRE
AI
DOI：10.1016/j.compchemeng.2026.109646.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Developing robust dynamic models for Continuous Direct Compression (CDC) tableting lines is crucial for process understanding, control, and quality assurance. This study develops a novel methodology for modeling CDC dynamics using the Sparse Identification of Nonlinear Dynamics with control (SINDYc) framework, adapted to capture the time-delayed responses of Blend Uniformity (BU) to changes in API mass flow. We introduce a tailored candidate function library incorporating lagged and moving-averaged control inputs, constructed from historical data. By integrating this tailored formulation with Bayesian inference, we obtain compact dynamic models together with quantified uncertainty in both coefficients and predictions. Using simulated CDC datasets with varying sampling rates (1 s, 10 s, 60 s) and noise levels, model performance was systematically evaluated. Results indicate that sampling resolution is the dominant factor: very sparse data (60 s) reduces predictive accuracy and calibration, while intermediate sampling (10 s) achieves a practical balance between data acquisition demands, accuracy, and uncertainty reliability, even under noisy conditions. The interaction between moving average pre-processing and sampling frequency was found to influence the selection of active delay terms, highlighting the role of feature construction in SINDy models. Overall, the proposed framework provides a data-efficient and uncertainty-aware approach for modeling CDC processes and similar systems with inherent delays, offering a practical basis for model-based monitoring and control in pharmaceutical manufacturing.
Keywords:
Sparse identification
Nonlinear process
Bayesian inference
Continuous direct compression
Transport delay
Uncertainty quantification
Sampling frequency
Journal
C
IF:
3.9
Papers:
184
Citations:
0