Artificial Intelligence Improvement of Rate Theory Models

安大略省的CANDU核反应堆产生了超过60%的电力，而这个行业的二氧化碳排放量微不足道，为加拿大提供了76,000个工作岗位。CANDU核电站（NPPs）的主要财务和二氧化碳成本出现在其初始建设期间，因此通过实现准确的维护预测来最大化使用寿命，从而提供经济和环境效益。核电站的安全运行取决于结构材料在高温、腐蚀和辐射等恶劣环境下的性能。在役材料老化的主要机理之一是辐射诱导偏析（RIS）。RIS是溶质元素在材料中特定位置（如晶界）的空间再分布，这导致材料易受其他降解机制的影响，如应力腐蚀开裂。虽然在RIS上收集了实验数据，但数据的广度受到处理放射性物质的限制。因此，需要模型来弥补实验数据的局限性，并协助进行预测。然而，目前，我们的模型由于缺乏作为模型输入所需的基本材料参数的知识而受到限制。本提案的目标是开发一个物理知情的人工智能计算平台，用于RIS。结合使用基于物理的模型（速率理论扩散）和人工智能方法分析现有的实验数据，我们将确定模型输入的优化材料参数。这将直接支持改进核电站部件寿命预测的工作。

Ontario's CANDU nuclear reactors generate more than 60% of its electricity, with negligible CO2 production in an industry supporting 76,000 Canadian jobs. A major financial and CO2 cost of CANDU nuclear power plants (NPPs) arises during their initial build, hence maximizing lifetimes by enabling accurate maintenance predictions provides economic and environmental benefits. Safe operation of an NPP depends on the performance of structural materials in a harsh environment of high temperatures, corrosive conditions and radiation. One major aging mechanism of materials in-service is radiation induced segregation (RIS). RIS is the spatial redistribution of solute elements to specific locations in a material (e.g. grain boundaries) - this then leads to susceptibility to other degradation mechanisms such as stress corrosion cracking. While experimental data has been collected on RIS, the breadth of data is limited by the constraints of dealing with radioactive materials. Therefore models are required, to bridge the limits of experimental data and assist in making predictions. However at present, our models are limited by a lack of knowledge of fundamental material parameters that are required as inputs to the models. The goal of this proposal is to develop a physics-informed artificial-intelligence computational platform for application to RIS. Using a combination of a physics-based model (rate theory diffusion) and an Artificial Intelligence approach analysing existing experimental data we will determine optimised material parameters for model input. This will directly support efforts to improve lifetime prediction of nuclear power plant components.