Developing a Novel Data-driven Modelling Strategy for Process Flow Diagram Optimisation

该项目旨在为新配方创建可扩展的最佳工艺流程图(PFD)。配方化学品的一个独特特点是其产品寿命短，这意味着战术性和侵略性的产品创新至关重要。随着趋势和消费者需求的变化更加频繁和季节性，制造商必须适应和迅速响应客户不断变化的需求，同时能够生产改进的配方。配方化学工业中普遍存在的挑战涉及当前的工艺流程图(PFD)开发程序，该程序经历了几个基本但耗时的步骤，即从实验室规模的产品形成到中试规模的研发过程设计，最后转变为工厂制造。由于生产线变得更加复杂，分析和优化变得更加复杂，在扩大规模期间需要进行大量的设计迭代和实验。这些挑战导致PFD的开发周期很长，需要大量的劳动力和能源成本以及产生废物。例如，仅处理通过这些试验产生的废物就占个人护理产品制造厂总能源成本的10%。为了解决这一挑战，在这个项目中，我们将重点开发一个使用实时过程数据的高效反馈回路，以显著提高拟议的PFD的性能。该项目的创新之处在于研究如何使用最先进的可解释机器学习和混合建模技术来：(1)确定配方性质和PFD之间的关键关联；(2)为新配方创建可扩展和最佳的PFD；以及(3)利用实时过程数据开发有效的反馈回路，以进一步改进拟议的PFD的性能。它与EPSRC在制造未来和数字制造方面的优先事项非常一致。

This project aims to create scalable and optimal Process Flow Diagrams (PFDs) for new formulations. A unique characteristic associated with formulated chemicals is their short product life, meaning that tactical and aggressive product innovation is crucial. As both trends and consumers' desires change more frequently and seasonally, manufacturers must adapt and respond quickly to the everchanging needs of the customer, whilst being able to manufacture improved formulations. A prevalent challenge within the formulated chemical industries involves the current process flow diagram (PFD) development procedure which undergoes several essential, but time-consuming steps, i.e. starting from lab-scale product formation to pilot-scale R&D process design, and finally transformation to factory manufacturing. Numerous design iterations and experiments are required to be carried out during upscaling due to the production line becoming more complex to analyse and optimise. These challenges result in a long PFD development cycle with substantial labour and energy costs and waste generation. For example, disposing wastes generated through these trials alone accounts for 10% of total energy costs for a personal care product manufacturing plant. To resolve this challenge, in this project we will focus on developing an efficient feedback loop using real-time process data to significantly improve the performance of a proposed PFD. The innovation of this project is investigating how state-of-the-art interpretable machine learning and hybrid modelling techniques can be used to: (1) identify key correlations between formulation properties and PFDs; (2) create scalable and optimal PFDs for new formulations; and (3) develop an efficient feedback loop using real-time process data to further improve the performance of a proposed PFD. It aligns well with EPSRC's priority in Manufacturing the Future and Digital Manufacturing.