Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø

Past Seminars:

Seminar ArchiveÌý

Seminars 2022

13-Jan-2022: Chu Lun Alex LeungÌý(Department of Mechanical Engineering, Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø & Rutherford Appleton Laboratory, HarwellÌý),ÌýA Journey ToÌýData-DrivenÌýReliableÌýEfficientÌýAdditiveÌýManufacturing (DREAM)Ìý

Powder bed fusion (PBF) additive manufacturing (AM) produces complex net-shape parts from powder feedstock, in a layer-by-layer manner. This emerging technology produces functional products that serve a wide range of industrial sectors, including architecture, aerospace, automotive, biomedical, energy, etc. However, AM parts are not often used as safety-critical components,Ìýe.g.Ìýturbine blades or propellers, owing to the presence of imperfections,Ìýe.g.Ìýpores and cracks. This presentation will first review the history of using 3D and 4D X-ray imaging to examine AM parts. I will then deep-dive into my journey of the development and application of a ‘physical twin’ of the AM process, ultrafast synchrotron X-ray imaging, machine-learning image processing, and high-fidelity simulation to monitor and elucidate the process dynamics during PBF. We use these tools to quantify the process dynamics,Ìýe.g.Ìýchanges in keyhole geometry, porosity, and remelting zone, as a function of time, layer number, and local layer thickness. After that, we compare our data with a multiphase and multiphysics simulation to reveal the solid-liquid-gas-metal vapour interaction, evolution mechanisms of the keyhole, melt pool, and porosity. This talk highlights the importance of imaging and data analytics in gaining novel insights into the AM process and possible ways to make manufacturing technologies more reliable and efficient.Ìý

---

10-Feb-2022: Anna Scaife (Professor of Radio Astronomy at the University ofÌýManchester, where she Heads the Jodrell Bank Interferometry Centre ofÌýExcellence and an Academic Co-Director of Policy at Manchester),ÌýAI in the SKA Era: Challenges for Recovering Well-Calibrated Uncertainties from Bayesian Deep-Learning

The expected volume of data from the new generation of scientific facilitiesÌýsuch as the Square Kilometre Array (SKA) radio telescope has motivated theÌýexpanded use of semi-automatic and automatic machine learning algorithmsÌýfor scientific discovery in astronomy. In this field, the robust and systematicÌýuse of machine learning faces a number of specific challenges, including aÌýpaucity of labelled data for training (paradoxically, although we have too muchÌýdata, we don't have enough), a clear understanding of the effect of biasesÌýintroduced due to observational and intrinsic astrophysical selection effects inÌýthe training data, and motivating a quantitative statistical representation ofÌýoutcomes from decisive AI applications. In this seminar, I will talk specificallyÌýabout the challenge of recovering well-calibrated uncertainties from BayesianÌýneural networks whenÌýclassifying radio galaxies, a canonical example of aÌýradio astronomy AI application. I will discuss how both model and likelihoodÌýmisspecification can affect this calibration, how these effects potentiallyÌýcontribute to the cold posterior effect seen when building models using realÌýastronomical data, and how error calibration can be affected by domain shift between labelled and unlabelled datasets.

---

10-Mar-2022:James Hetherington (Director of ARC, Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø)ÌýÌýIntroducing Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø ARC, A New Kind of e-Science Research and Innovation CentreÌý

In this talk, ProfessorÌýJamesÌýHetherington, Director of Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø’s new Advanced Research Computing Centre, will talk about the vision forÌýthe department as an unusual hybrid: a support centre providing supercomputing platforms that enable computational science in Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø and beyond, a home for pools of research technology professionals (research software engineers, data scientists, informaticians and data stewards) contributing to Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø research teams, and a research institute in the methods of digital research. He will look at factors which limit the progress of digital transformation in research, consider how the new model proposed for the centre may help, and review the risks that may prevent success.Ìý

---

7-Apr-2022:ÌýGabriel Facini (Data-Intensive Science and Industry [DISI], Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø),ÌýA View of the ATLAS Experiment in Preparation for Run 3 and Beyond

ATLAS,Ìýa flagshipÌýdata-intensive experiment, is in preparation for Run 3 of the Large Hadron Collider. The seminar will review recent results in the hunt for beyond the Standard Model physics, challenges of the charged particle reconstruction, prospects for innovative use of advanced DIS techniques, and long-term challenges.

---

12-May-2022:ÌýAdhamÌýHashibon (Data-driven innovation, Materials Modelling, Ïã¸ÛÁùºÏ²ÊÖÐÌØÍø),ÌýMaterials Digitalisation: what is an ontology and how it can be utilised in data intensive domainsÌý

In this presentation, an introduction to foundations of semantic interoperability and ontology is given. Specifically, the Elementary Multi Perspective Materials Ontology (EMMO) as basis for digitalisation frameworks is introduced along with a short survey of existing digitalisation tool, including the SimPhoNy – Integrated Simulation Framework.Ìý

Ìý