Data-driven approaches for in-line monitoring of particle attributes in chemical and pharmaceutical manufacturing processes.

颗粒处理广泛应用于化学和制药工业。在这种情况下，颗粒属性影响加工性能，是优化产品质量的关键。然而，尽管涉及高昂的材料成本，但这些部门仍然普遍存在显著的工艺效率低下问题。改善颗粒属性监测的新技术对于改变理解和控制制药过程的能力以及实现航空航天和汽车等其他行业的可靠性和稳定运行至关重要。目前，颗粒属性主要使用离线技术来表征，这些技术在采样，运输和分析过程中容易发生颗粒变化。在线测量作为一种快速的替代方法正在迅速发展，以克服这些限制，并有可能提供更具代表性的现场颗粒群视图。然而，由于在线测量环境的复杂性，颗粒定量属性的提取仍然存在未解决的挑战。该项目将使用实验，数据分析和模拟的组合，以提供更准确的在线测量颗粒属性的表示。将使用连续制造和结晶中心（CMAC）提供的最先进的过程分析技术（PAT）采集数据，包括粒子观察显微镜（PVM）、聚焦光束反射测量（FBRM）和拉曼光谱。这些数据流将为机器学习和深度学习模型的开发提供信息，以提取更具代表性的粒度、形状和形态分布以及溶液固体含量。测量环境的模拟将有助于识别与理想情况的偏差，并为这些异常提供物理意义。虽然从单个传感器中提取信息本身就是一个挑战，但该项目将致力于实施数据融合方法，以进一步增强颗粒属性的在线量化，并为更先进的过程控制策略提供信息。

Particle processing is widely used in chemical and pharmaceutical manufacturing industries. In this context, particle attributes influence processability and are key to the optimisation of product quality. However, despite the high material costs involved, significant process inefficiencies are still common in these sectors. New technologies that improve the monitoring of particle attributes are essential to transform the ability to understand and control pharmaceutical processes and to achieve the reliability and stable operation of other sectors such as aerospace and automotive.Currently, particle attributes are mainly characterised using off-line techniques that are prone to particle alteration during sampling, transport and analysis. In-line measurements are quickly developing as a fast alternative to overcome these limitations and have the potential to provide a more representative view of the particle population in-situ. However, unsolved challenges still remain in the extraction of quantitative particle attributes due to the complex in-line measurement environment.The project will use a combination of experiments, data analytics and simulation to provide more accurate representation of particle attributes from in-line measurements. Data will be captured using state-of-the-art Process Analytical Technologies (PAT) available at the Centre for Continuous Manufacturing and Crystallisation (CMAC), including Particle View Microscopy (PVM), Focused-Beam Reflectance Measurement (FBRM) and Raman spectroscopy. These data streams will inform the development of Machine Learning and Deep Learning models to extract more representative particle size, shape and morphology distributions, as well as solution solid loading. Simulations of the measurement environment will contribute to identifying deviations from ideal scenarios and to providing physical meaning to these anomalies. While extracting information from individual sensors is a challenge in itself, the project will aim to implement data fusion approaches to further enhance in-line quantification of particle attributes and inform more advanced process control strategies.