Adaptive Dosing Control Strategies for the Improvement of Dosing Accuracy of Peristaltic Pumps in Pharmaceutical Manufacturing

Peristaltic pumps are widely employed in pharmaceutical manufacturing for precise liquid dosing, yet they face inherent accuracy limitations, particularly in low-volume applications where deviations can reach 4.0%. Traditional approaches rely on costly mechanical enhancements that provide inconsistent improvements across different volume ranges. This research focuses on investigating systems for improving dosing precision in pharmaceutical applications. The presented Adaptive Dosing Control System (ADCS) operates on the fundamental principle of predicting the subsequent dose based on historical data and implementing corrective adjustments that counteract the anticipated deviation, enabling proactive error compensation. We investigated both traditional statistical and machine learning approaches, analyzing and fine-tuning ARIMA and Recurrent Neural Network models chosen for their ability to capture temporal dependencies in sequential data, alongside ensemble methods that combine multiple predictive approaches. Through extensive experimental validation processing over 100,000 individual doses across volumes from 0.1 to 2.0 ml, we demonstrate significant accuracy improvements: statistical models achieving up to 49.4% improvement for standard volumes, neural networks excelling in micro-volumes with 47.0% improvement, and ensemble approaches reaching 53.9% improvement, surpassing state-of-the-art mechanical solutions. This offers key advantages: implementation on existing hardware without costly modifications, adaptive response to changing operational conditions, and superior performance in critical micro-volume applications. From analysis of experimental prediction data, we discovered that compensation effectiveness can also be estimated without actual machine testing and with a reasonable precision. We refined an offline performance indicator facilitating design space exploration and optimization, representing another novel outcome enabling rapid assessment of different configurations. The research establishes the complex relationship between temperature variations and dosing precision, revealing not only predictable thermal expansion effects but also non-trivial impacts on tube elasticity and mechanical components. Multiphysics FEM validation demonstrates this relationship can induce deviations up to 1.7% of nominal volume. This work demonstrates how algorithmic methods can enhance dosing system performance, providing an economical solution that can be used synergistically with traditional approaches in pharmaceutical manufacturing and other contexts where fluid dosing accuracy is critical.