Fast, efficient and reliable: digital qualification of ultrasonic inspection for safety-critical components

在航空航天和核能等高价值制造业，安全至关重要。因此，安全关键部件的设计和检验是整个开发周期中至关重要的一部分。然而，目前的实践广泛使用物理组件和模型的实验试验，其中必须引入人工缺陷，限制在少数特定测试用例中，以证明它们可以被检测和表征。因此，检验资格是非常耗时和昂贵的（一些含有缺陷的部件的完整模型要花费数百万英镑），而且与敏捷、小批量、定制、数字化制造的普遍趋势不一致。我们建议用数字替代方法取代这些昂贵、浪费的物理测试样本，以提高制造效率。实现这一目标需要开发高速、代表性、逼真的数值模拟能力，并结合在参数空间的高维空间中可靠采样和插值的解决方案。这种虚拟测试能力将使高价值组件的检查能够在单个部件制造之前进行设计，优化和合格。它将为操作员培训提供模拟工具的基础，并能够以培训未来检测自动化机器学习解决方案所需的规模和保真度生成数据。超声波阵列检查将是一个示范案例，因为这是最广泛使用的评估安全关键部件内部完整性的方法，无论是在制造还是在使用中。要实现这一目标，需要经过验证的工具来大规模地综合真实的检测数据，并需要一种方法来稳健地探索可能缺陷的巨大参数空间，以确定检测性能。为了实现这一雄心勃勃的愿景，我们的想法是从两个互补的方向来解决问题。自下而上：我们将使原始数据的直接数值模拟更加高效。在申请者之前世界领先的研究的基础上，我们将展示如何更好地利用数值模拟工具来减少至少一个数量级的计算负担。自顶向下：我们将对多维参数空间进行定量表征，使检测性能更有效。利用我们的领域知识和与工业合作者（Rolls-Royce， EDF, Jacobs， Airbus和KANDE）的广泛讨论，我们将开发合适的代理建模，采样和集成策略，以便通过少量高保真度数值模拟准确表征参数空间。为了解决这个问题，我们将开发一套工具和技术，以确保检测资格的成本和复杂性降低几个数量级，使其适合未来的数字制造。

In high-value manufacturing sectors such as aerospace and nuclear, safety is paramount. For this reason, the design and qualification of inspection for safety-critical components is a crucial part of the overall development cycle. However, current practice makes extensive use of experimental trials on physical components and mock-ups, into which artificial defects, limited to small numbers of specific test cases, must be introduced to demonstrate that they can be detected and characterised. Inspection qualification is therefore extremely time-consuming and costly (with some full mock-ups of defect-containing components costing £millions), and at odds with the general move toward agile, small-batch, bespoke, digitally-enabled manufacturing. We propose replacing the use of these expensive, wasteful, physical test specimens with digital alternatives, to improve manufacturing efficiency. Delivering this will require high-speed, representative, realistic numerical simulation capabilities to be developed, in combination with solutions to reliably sample and interpolate across the high dimensionality of the parametric space. This virtual testing capability will enable the inspection of a high value component to be designed, optimised, and qualified before a single part has been manufactured. It will provide the basis of a simulation tool for operator training and be able to generate data at the scale and fidelity needed to train future machine learning solutions for inspection automation. Ultrasonic array inspection will be the demonstrator case as this is the most widely used method for assessing the internal integrity of safety-critical components, both at manufacture and in service. To achieve the goal requires validated tools to synthesise authentic inspection data at scale and a methodology to robustly explore the vast parameter space of possible defects to determine inspection performance. Our idea to achieve this ambitious vision is to approach the problem from two complimentary directions.Bottom-up: we will make the direct numerical simulation of raw data more efficient. Building on previous world-leading research by the applicants, we will show how numerical simulation tools can be better exploited to reduce the computational burden by at least one order of magnitude. Top-down: we will make the quantitative characterisation of the multi-dimensional parameter space to qualify inspection performance more efficient. Drawing on our domain knowledge and in extensive discussion with industrial collaborators (Rolls-Royce, EDF, Jacobs, Airbus, and KANDE), we will develop suitable surrogate modelling, sampling, and integration strategies for accurately characterising the parameter space with a small number of high-fidelity numerical simulations.In addressing this problem we will produce a set of tools and techniques that ensure that inspection qualification is reduced in cost and complexity by orders of magnitude, leaving it fit for the future of digital manufacturing.