Roman Gr. Maev is a distinguished university professor of the University of Windsor, Ontario and founding director-general of The Institute for Diagnostic Imaging Research, Canada - a multi-disciplinary, collaborative research institute. The diverse range of disciplines encompassed by Dr. Maev includes theoretical physical acoustics, ultrasonic and nonlinear acoustical imaging, nanostructural properties of advanced materials and its analysis. He has published over 600 peer-reviewed items and holds 44 international patents. Dr. Maev is a Life Fellow of IEEE, and a Fellow of ASNT, BINDT, CINDE, and RSNTTD. Dr. Maev was the recipient of various Fellowships and Awards, including the Roy Sharpe Award, UK, and the ASNT Mentoring Award, USA. He has Chaired numerous US, Canadian and International Symposia and Conferences, also Dr. Maev is a Distinguished Lecturer of IEEE and has presented a number of Keynote and Invited lectures worldwide.

NDE AND DEEP LEARNING: FASHION TREND OR THE FUTURE?

The future of NDE in light of the increasing use of artificial intelligence (AI) - particularly deep learning - is at least uncertain if not controversial. In many ways, deep learning has brought ground-breaking advancement in NDE data interpretation and downstream decision-making - and promises to continue to do so. On the other hand, widespread use of deep learning artificial intelligence in NDE systems is impeded due to some major challenges and concerns regarding its role, development, standardization, consistency, validation, generality, trustworthiness, coexistence with human experts, and ethical use - among others.

In this presentation, we will discuss some of these major challenges and concerns regarding the use of deep learning in NDE, along with case studies from our work and examples from works of others. In particular, we highlight results from our developments of advanced NDE technologies which enable zero-defect mass production of bonded joints through the integration of AI into real-time ultrasonic process monitoring systems. We will discuss use cases, including the implementation of resistance spot weld process monitoring using a deep learning approach which analyzes ultrasonic B-scans via semantic segmentation in real time. Our AI can assess throughout the weld process various weld properties including e.g. the amount of nugget penetration into each sheet in the stack up. These assessments are fed back to an adaptive weld controller so it can always produce high-quality welds that match production-level requirements. A key aspect of this problem is that spot weld cycle times are fast, so real-time assessments are crucial. The required performance, generality, and speed necessitate deep learning. Fast inference and communications are vital to feedback actionability; it must be fast enough that the weld controller can suitably adapt its parameters to produce high-quality welds.

Our approach is a great example of advancement in NDE 4.0. and is applicable beyond spot welding and ultrasonic NDE. We are confident that successful implementation of such technologies will revolutionize manufacturing including automotive and aerospace. These technologies have the potential to bring big savings in production, reduce labor costs, and eliminate unnecessary destructive tests which are still part of today's quality inspection process. Our goal is to achieve zero-defect mass production and our results demonstrate that this is achievable today.

We will discuss what is still needed from the NDE community to further facilitate widespread use of modern-day artificial intelligence in NDE. We will also highlight potential ways in which some challenges and concerns surrounding widespread use of artificial intelligence in NDE might be alleviated considering the broader ongoing artificial intelligence research and development, for example with respect to emerging learning paradigms, generative modeling, interpretability in artificial intelligence, and emerging model architectures. Finally, we will discuss what the future of NDE may hold, and whether deep learning artificial intelligence will have any part of it.

Nico Avdelidis is a professor and head of the Integrated Vehicle Health Management (IVHM) Centre at Cranfield University since July 2020. The Centre is funded by a number of industrial partners such as Boeing, BAe Systems, etc. Nico is also a professeur associe (visiting professor) at Universite Laval in Quebec, Canada, where he does a lot of his research activities there in collaboration also with other organizations in Canada including the National Research Council (NRC Canada). He has contributed extensively to several research areas, such as non-destructive monitoring and diagnostics, robotic and autonomous systems in MRO, advanced sensing technologies, advanced IR imaging and image fusion using non-invasive techniques, and aircraft structures and/or systems monitoring.

IR THERMOGRAPHY AS A TOOL IN MRO ACTIVITIES FOR A SUSTAINABLE AIRCRAFT

IR Thermography (IRT) is a technique that could be applied efficiently for the inspection of composite materials and an emerging NDT tool in the aviation sector with specific characteristics that make it attractive, such as being fast and effective. Thermography is effective in imaging a wide area in a single acquisition and detecting the presence of defects. Therefore, the technique can be applied as a fast full-scale inspection technique, by identifying promptly deficiencies like delamination, cracks, voids, and impact damage. In this talk, aircraft inspections for maintenance operations using IRT are discussed in order to determine the reliability and repeatability of the technique as a Non-Destructive Testing (NDT) tool. The advantages of IRT for fully automated and/or autonomous assessments in aircraft structures are presented. Furthermore, the limitations of the technique are pointed out and suggestions to overcome such challenges are conferred. Specific examples – applications discussed include hangar and UAV inspections for automated damage assessment for the future of Maintenance, Repair and Overhaul (MRO), the use of Machine Learning (ML) for damage assessment and decision making, and in parallel the aim to improve operational efficiency by eliminating unplanned maintenance and reduced time in hangar using such tool is also defined.

Željko Ivezić is a professor of astronomy at the University of Washington. He obtained his undergraduate degrees in physics and mechanical engineering from the University of Zagreb, Croatia and his PhD in physics from the University of Kentucky. Željko is serving as the Rubin Observatory Construction Director and Head of Science for Rubin's Legacy Survey of Space and Time. Željko's scientific interests include detection, analysis and interpretation of electromagnetic radiation from astronomical sources, with emphasis on Big Data analysis and modeling of infrared radiative transfer in stellar envelopes and quasars.

ASTRONOMERS AND AI IN THE BIG DATA ERA

Astronomical sky surveying is experiencing a bonanza as detectors, telescopes and computers become ever more powerful. At the same time, increased data rates, volume and complexity lead to new challenges for data analysis. I will first motivate discussion by introducing Rubin Observatory and Legacy Survey of Space and Time, which will deliver about 100 PB of astronomical imaging data and catalogs including over 20 billion stars, galaxies and other objects, and then I will discuss a few analysis challenges using examples ranging from studies of dark matter and dark energy to search for hazardous near-Earth asteroids.