Probabilistic Machine Learning Driven Discovery and Design of New Materials for Sustainable Energy and Transport

提供清洁、可靠、环保的能源是一项重大的全球性挑战。为了克服这个问题，运输和能源行业正在通过采用更新更好的材料技术进行范式转变。为了加快材料开发的进程，传统的基于试错的实验方法正在被计算材料科学与目标实验的协同整合所取代。我的团队专注于使用这种综合计算材料工程方法：(i)为汽车和航空航天结构设计更轻、更强、更坚韧的材料，以提高燃油经济性，同时保持其安全性和性能；(2)发现新材料，使电池、催化剂和太阳能电池等可持续能源技术更高效、更经济。第一个主题是提高能源效率，而第二个主题是开发清洁能源生产的新技术。然而，设计新材料是相当复杂的，在这方面，机器学习（ML）的新兴领域可以帮助加速材料开发的步伐，通过从包含大量变量的数据中捕获模式，这些变量很难从人类直觉中捕获。******提出的研究计划的总体目标是通过有效地将数学上稳健的贝叶斯机器学习技术与物理上精确的原子建模相结合，设计和发现用于轻型运输和可持续能源的新材料。对于结构材料，目标是开发具有高保真度和效率的多尺度材料模型，能够预测包括失效在内的全局响应。在高通量密度泛函理论计算生成的数据集上使用ML，拟议的研究还将：(i)绘制出广泛的二维材料的结构-力学性能关系，（ii）具有最佳容量和寿命性能的金属-空气电池的筛选电极材料，（iii）设计用于CO2还原的气相催化剂，以及（iv）开发广泛用于结构应用的钢的强大的原子间电位。我们的长期愿景是与实验和工业伙伴密切合作，实现所提出的材料设计并将其商业化。******该计划将为NSERC的先进制造目标领域开发新的科学知识和材料技术，并培养六名博士生，成为能源、制造和运输行业的未来领导者。实际上，它将为加拿大制造业提供设计工具，以创造更坚固、更坚固的轻质材料、汽车用新电池材料和太阳能转换的新催化剂。*****

Providing clean, reliable and environment-friendly energy is a critical global challenge. To overcome this, the transportation and energy industries are undergoing a paradigm shift by adopting newer and better materials technologies. For speeding up the process of materials development, traditional trial-and-error based experimental approaches are being replaced by a synergistic integration of computational materials science with targeted experimentation. My group focuses on using this Integrated Computational Materials Engineering approach: (i) to design lighter, stronger, and tougher materials for automotive and aerospace structures to boost fuel economy while maintaining their safety and performance; and (ii) to discover novel materials to make sustainable energy technologies such as batteries, catalysts, and solar cells more efficient and cost-effective. The first theme caters towards improving energy efficiency while the latter towards developing new technologies for clean energy production. Designing new materials is, however, quite complex and in this respect, the emerging field of machine learning (ML) can help accelerate the pace of materials development by capturing patterns from data consisting of a multitude of variables that are difficult to capture from human intuition. ******The overarching goal of the proposed research program is to design and discover new materials for lightweight transportation and sustainable energy by effectively combining mathematically robust Bayesian machine learning techniques with physically accurate atomistic modeling. For structural materials, the aim is to develop multiscale material models with high fidelity and efficiency that are able to predict the global response including failure. Using ML on datasets generated by high-throughput density functional theory computations, the proposed research will also: (i) map out the structure-mechanical property relationships for a wide range of two dimensional materials, (ii) screen electrode materials for metal-air batteries with optimum capacity and life-time performance, (iii) design gas-phase catalysts for CO2 reduction, and (iv) develop robust interatomic potentials for steels widely used in structural applications. Our long-term vision is to physically realize proposed material designs and commercialize them in close collaboration with experimental and industry partners. ******The proposed program will contribute by developing new scientific knowledge and materials technologies for NSERC's target areas in Advanced Manufacturing and train six PhD students as future leaders in the energy, manufacturing and transportation industries. Practically, it will lead to design tools for the Canadian manufacturing industry to create stronger and tougher lightweight materials, new battery materials for automotives, and new catalysts for solar energy conversion.*****