Data Science and Composite Materials Manufacturing

不列颠哥伦比亚省大学的两个研究集群--复合研究网络和数据科学研究所--与他们的工业合作伙伴Convergent制造技术公司正在交织两条互补的、迄今连接不畅的数字线索：数据科学与工程科学，以提升加拿大在先进材料制造领域的地位。过去二十年，我们见证了所谓的“大数据”时代的出现。这是由从大量丰富、复杂的数据学习到构建复杂的模型并从中提取有洞察力的模式方面的突破推动的。许多应用领域，如金融、医学等，都受益于数据科学、机器学习、人工智能和计算统计中的这些新方法。然而，先进制造的应用领域，特别是在复合材料领域，尚未利用和利用上述技术的进步。特别是，我们的动机是航空航天工程中复合材料(例如碳纤维)部件高压罐固化的问题，以及应用数据科学、机器学习和概率编程的进步来更好地理解和预测复合材料制造过程中的物理过程。最终结果是减少复合材料设计和制造中的昂贵决策(即时间、财务)--最终导致提高效率和更好的产品。将统计机器学习工具应用于复合材料制造的关键挑战是：小而昂贵的数据；缓慢而复杂的正向生成；需要指定控制输入；昂贵的特征；考虑模拟器误差；以及不确定性量化。在这个提案中，我们将解决所有这些挑战，以解决物理过程的数据驱动模拟器中的推理、预测和优化问题，尤其是复合材料固化过程。

The consortium of two research clusters at the University of British Columbia--the Composite Research Network and Data Science Institute -- and their industrial partner, Convergent Manufacturing Technologies Inc., are interweaving two complementary, yet to-date, poorly connected digital threads: data science with engineering science to advance Canada's position in advanced materials manufacturing. Over the past two decades, we have witnessed the emergence of the so-called "Big Data" era. This is driven by breakthroughs in learning from massive quantities of rich, complex data, to building sophisticated models and extracting insightful patterns from them. Numerous application domains, such as in finance, medicine, and so on, have benefited from these new methods in data science, machine learning, artificial intelligence, and computational statistics. However, the application domain of advanced manufacturing, particularly in composite materials, has yet to harness and leverage the advances in the aforementioned technologies. In particular, we are motivated by the problem of composite (e.g., carbon fibre) part autoclave curing in aerospace engineering, and applying advances in data science, machine learning, and probabilistic programming to better understand and predict the physical processes during composites manufacturing. The ultimate outcome is to reduce costly decisions (i.e., time, financial) in the design and manufacturing of composites--ultimately leading to increased efficiency and better products. The key challenges in applying statistical machine learning tools to composites manufacturing are: small, expensive data; slow, complex forward generation; need to specify control inputs; expensive features; accounting for simulator errors; and uncertainty quantification. In this proposal we will tackle all these challenges to address the problems of inference, prediction, and optimization in data-driven simulators of physical processes, with a particular focus on composite curing processes.