ATLAS: Assurance through layer-wise anomaly sensing

我们将开发金属增材制造过程中缺陷的实时检测技术。我们设想未来，每一个零件都将有一个包含缺陷的3D地图的数字副本。这将使制造商能够通过虚拟测试加速高完整性零件的认证和质量保证，并为快速优化工艺参数提供在线反馈。该项目解决了数据分析、实时优化、虚拟测试和模型验证等多个数字制造研究挑战。金属材料的增材制造（AM），也称为3D打印，正在改变能源、运输、医疗保健和国防部门的制造供应链。它刺激设计创新，通过更轻、性能更好、更可靠的产品，它可以帮助我们实现未来的净零和可持续发展目标。然而，在安全关键行业中使用增材制造零件受到材料性能一致性的限制。这些问题对质量保证提出了相当大的挑战，减缓了增材制造工艺的进一步采用，并限制了急需的创新。我们的目标是使用过程传感来解决这一挑战，其中相机和其他传感器类型实时观察制造过程，结合数据驱动的机器学习模型来预测何时发生缺陷。为此，我们将设计和构建代表常见工业设计的零件几何形状，并从我们独特的过程监测平台收集多个传感器模式（即同轴熔池成像，表面温度，熔体轨迹形态传感器系统）的过程监测数据。然后，这些部件将在构建后进行微ct扫描，以建立孔隙度真实数据，创建一套原始的、空间注册的数据集。构建将涵盖各种工业相关的制造参数和常见的机器问题，如肮脏的镜头，堵塞的过滤器，污染的粉末，磨损的雨刷片等。这些数据集将用于训练和验证数据驱动的机器学习模型，以预测零件孔隙度。稳健的非破坏性评估方法将用于表征模型的性能。然后，我们将实现在线分层反馈，动态调整加工参数，并通过选择性重熔修复缺陷。这种方法将解决模型鲁棒性、数据缩减、实时处理、优化和反馈方面的基本挑战。最终，该项目将提高金属增材制造零件的质量，并加速虚拟认证。综合起来，这些产出将降低开发创新新产品的风险，消除广泛采用金属增材制造技术的重大障碍。

We will develop technology for the real-time detection of defects in metal additive manufacturing processes. We envision a future where every part made will come with a digital copy of itself containing a 3D map of defects. This will enable manufacturers to accelerate certification and quality assurance of high-integrity parts through virtual testing and also provide online feedback for the rapid optimisation of process parameters. This project addresses multiple digital manufacturing research challenges across data analytics, real-time optimisation, virtual testing, and model verification.Additive manufacture (AM), also known as 3D printing, of metallic materials is transforming manufacturing supply chains across the energy, transport, healthcare, and defence sectors. It stimulates design innovation and through lighter, better performing and more reliable products, it can help us meet our future net zero and sustainability goals. However, use of AM parts in safety critical industries is limited by concerns around material property consistency. These concerns present a considerable challenge for quality assurance, slowing further adoption of AM processes and constraining much needed innovation. We aim to solve this challenge using in-process sensing, where cameras and other sensor types observe the manufacturing process in real-time, in combination with data-driven machine learning models to predict when defects occur. To do this we will design and build part geometries representative of common industrial designs and collect in-processing monitoring data across several sensor modalities (i.e. co-axial melt pool imaging, surface temperature, melt track morphology sensor systems) from our unique in-process monitoring platform. The parts will then be micro-CT scanned post-build to establish porosity truth data, creating a suite of pristine, spatially registered, data sets. The builds will cover various industrially relevant manufacturing parameters and common machine issues such as dirty lens, clogged filter, contaminated powder, worn wiper blade, etc. These data sets will be used for the training and validation of data-driven machine learning models to predict part porosity. Robust non-destructive evaluation methodologies will be used to characterise model performance. We will then implement online layer-wise feedback to dynamically adjust processing parameters and repair defects through selective remelting. This approach will address fundamental challenges in model robustness, data reduction, real-time processing, optimisation, and feedback.Ultimately, this project will enhance metal additive manufacturing part quality and enable accelerated virtual certification. Combined, these outputs will reduce the risk involved in developing innovative new products, removing a significant barrier to the widespread adoption of metal additive manufacturing technology.