Development of Advanced AI-assisted Computational Routines to Infer Quantum Technology Sensors Data

“如果没有适当和强大的后处理分析和算法来解释原始数据，从（QT）重力传感器获得的数据几乎没有价值。通常，来自传感器的数据通过反演过程转化为有意义的信息。现有的反演算法在处理大数据和复杂数据模式时存在精度和效率上的不足。特别是，现有的反演方法只能用于某些规则的密度异常，即复杂的几何形状通常在反演过程中被忽略，或者计算成本很高，因此没有考虑。此外，即使对于简单的密度异常，目前的反演程序计算效率也很低，在大多数情况下，可能需要几天才能返回有用的结果。这个博士项目打算通过开发一种变革性的方法来克服上述问题，该方法使用人工智能（AI）支持的新颖和先进的计算建模工具来解释（QT）重力传感器数据。特别是，本博士将专注于通过将数学上传统的反演工具与机器学习（ML）算法集成来开发有效的反演程序，从而从根本上减少反演过程。在机器学习方法中，将特别关注进化计算技术（即遗传算法，GA），因为当使用合适的数值建模策略时，它们具有鲁棒性和效率，可以找到复杂问题的全局最优解。在iFEEL合作项目中使用重力数据，在GUIDE （EP/P010415/1）项目中使用振动数据探测埋地管道，遗传算法的使用已经显示出有希望的初步结果。此外，将使用监督深度学习（ML技术的一个子类别）来研究高级建模和反演程序的使用。这种方法对于提高反演过程的计算效率至关重要，因为它将受益于（在可能的情况下）先前训练的模型，以减少计算时间。建议的处理算法的主要应用将是识别市区地下物体的位置（例如污水渠和隧道）或地面交通基础设施下的密度对比（例如识别道路和铁路轨道下是否存在冲蚀和天坑）。提名候选人（温纳奥尼先生）有数学背景，特别是在数值建模技术方面表现出色。他将非常适合推动合适的反演过程向前发展。因为他可以利用他在数学和统计学方面的知识和背景，为QT传感器数据开发新的数据推断工具。虽然在基于ML的重力数据反演领域存在着零星的（很少的）工作，但它们都没有提供一个统一的框架或关于如何使用ML来解决这个问题的一般指导。他们更倾向于关注一些具体问题，而没有提供任何有用的通用指导，因此，他们的作品对其他人没有多大用处。具体来说，我们设想这个博士学位将有助于首次开发一个合适的基于ml的工具，并为地球物理学家和相关行业推断QT重力数据提供必要的相关指导。除了提出的工具外，还将制作一个关于使用基于人工智能的QT传感器数据分析的“如何”包，该包可作为未来适应该方法的基准，并确保该研究的未来持续和可持续的过程。我们计划通过各种方式展示这种工具的实用性，包括通过我们与RSK的密切关系，以及在公共和重点活动中传播结果。在数据建模和数据处理中使用人工智能可以为alm提供重大进步

"The data obtained from (QT) gravity sensors have very little value without appropriate and robust post-processing analysis and algorithms to interpret the raw data. Usually the data from sensors are converted into meaningful information using an inversion process. The existing inversion algorithms have several limitations including the lack of accuracy and efficiency especially when dealing with large and complex data patterns. In particular, the existing inversion approaches, can only be used for certain regular density anomalies - i.e. complex geometries are usually overlooked in the inversion process or it becomes computationally expensive hence not considered. Additionally, even for simple density anomalies, current inversion procedures are computationally inefficient and in majority of scenarios can take days before they can return useful results. This PhD project intends to overcome the above issues by developing a transformative approach that uses novel, and advanced computational modelling tools supported by artificial intelligence (AI) to interpret (QT) gravity sensor data. In particular, this PhD will focus on developing efficient inversion procedures by integrating mathematically conventional inversion tools with machine learning (ML) algorithms to radically reduce the inversion process. Within ML approaches, a special attention will be given to evolutionary computing techniques (namely genetic algorithm, GA) due to their robustness and efficiency to find global optimum of complex problems when suitable numerical modelling strategies are used. The use of GA has shown promising initial results in the iFEEL partnership project using gravity data and, in the GUIDE, (EP/P010415/1) project using vibrational data for the detection of buried pipes. Furthermore, the use of advanced modelling and inversion procedures will be investigated using supervised deep learning (a sub-category of ML techniques). This approach will be crucial in improving the computational efficiency of the inversion process as it will benefit (where possible) from a previously trained model to reduce the computational time. The main application of the proposed processing algorithms will be to identify the location of underground objects in urban areas (e.g. sewers and tunnels) or density contrast under surface transportation infrastructure (e.g. to identify the presence of wash-out and sinkhole under roads and railway tracks).The nominated candidate (Mr Winner Oni) has a background in mathematics, particularly demonstrating a strong performance in numerical modelling techniques. He would be ideally suited to drive the development of suitable inversion processes forwards. as he can use his knowledge and background in mathematics and statistics to develop novel data inferring tools for the QT sensor data.Whilst scattered (and very few) works exist in the field of ML-based inversion of gravity data, none of them have offered a unified framework or a general guidance on how ML can be used to tackle the problem. They rather have focused on a few specific problems without offering any useful general instructions and therefore, their works have not been of much use to others. Specifically, we envisage that this PhD facilitates, for the first time, development of a suitable ML-based tool and necessary associated guidance for geophysicists and relevant industries to infer QT gravity data. In addition to the proposed tool, a 'how to' package on the use of AI-based analysis for QT sensor data will be produced that can be used as a benchmark for future adaptation of this approach and ensuring a continuous and sustainable process for future of this research. We plan to demonstrate the usefulness of such tool via various means including through our very close relationship with RSK, as well as dissemination of the results in both public and focused events. The use of AI in data modelling and data processing can offer significant advances to alm