Abstract:

It has long been a grand goal for researchers and industry professionals in the chemical engineering community to revolutionize the paradigm of chemical product development and shorten the time from product discovery to application. However, chemical product design is a complex process involving multiple components, scales, and physical fields. It is difficult for existing experimental research models to reveal the relevant physical and chemical mechanisms in depth and efficiently. Therefore, it is necessary to use multi-scale computer simulation technology to predict the properties of chemical products by coupling multi-scale simulation methods starting from the chemical structure at the micro-molecular level. Along with the increasing computing power, "artificial intelligence (AI)-driven" approaches are becoming a significant promise in the pursuit of this objective, where AI is being organically integrated with established multi-scale simulation techniques for efficient and high-fidelity modeling framework with potential for transformative impact on chemical design. For instance, machine learning models trained on high-fidelity multi-scale simulation data can accelerate the prediction of chemical structure-property relationship by orders of magnitude. However, the chemical industry, particularly, the development of new chemical products, presents many unique challenges. The crude application of AI to existing problems and data to construct some predictive models can hardly break the existing bottlenecks fundamentally. Hence, it is imperative to consider how we can integrate AI techniques more effectively and comprehensively with innovative chemical product design. We envision this can be achieved through, e.g., using AI to optimize existing physics-based simulation techniques and efficiently explore hundreds of millions of design parameters to find the best design solutions. Here we discuss the recent development of AI-driven chemical innovation design from three aspects: multi-scale simulation, material design framework, and scientific software development, with an emphasis on the important role of AI technology in achieving the innovation pathway of chemical products. At last, we present our perspective on the current efforts to embrace AI techniques in the engineering of novel chemical product, with the goal of providing a strong foundation to support the advancement of domestic chemical industry.

Key words: artificial intelligence, chemical products design, computer modeling

摘要：

改进化学品研发模式，缩短化学品从发现到应用的时间是化工行业中所有科学研究者和产业人员的最终目标。本文提出：化学品设计是一个涉及多组分、多尺度和多物理场的复杂过程，现有的实验研究模式难以深入高效地揭示相关的物理化学机制；因此，需要借助多尺度计算机模拟技术，从微观分子层面的化学结构出发，耦合多种模拟方法来预测宏观产品的性能；同时，随着计算机算力的提升，将基于物化机制的多尺度计算机模拟方法与数据驱动的人工智能相结合的研发模式，具有广阔的应用前景，例如基于高精度多尺度模拟数据训练的机器学习模型能够指数级地缩短化学品结构-性质的预测。尽管如此，由于广阔的分子结构空间和复杂的分子作用力关系，新型化学品研发面临着众多独特的挑战。如何借助人工智能提高现有模拟技术的准确性与速度，更好地理解和预测材料的性质和特点，并将人工智能引入材料设计算法，以实现更高效地探索和优化复杂的化工设计参数，使其更适应实际需求，是化学品设计研究的前沿方向。本文从多尺度模拟、材料设计框架和科学计算软件开发三个方面，分析讨论了人工智能驱动化学品创新设计的发展现状，阐述了人工智能技术在实现化学品设计创新途径中所起的重要作用，并对人工智能驱动在化学品设计的研究方向和发展目标进行了展望，以助力实现新型化学品的设计，为我国化工产业发展提供坚实的技术支撑。

关键词: 人工智能, 化学品设计, 计算机模拟