Keynote Lectures

Keynote Lecture 1

The New Zealand Geotechnical Database (NZGD): Foundations and Future Vision

Abstract: 

The New Zealand Geotechnical Database (NZGD) is internationally renowned as one of the most successful data-sharing platforms in the global geotechnical community.  The success of the NZGD is unique in many ways, not least due to its collaborative data-sharing model that is free to users and spans both public and private sectors.

The NZGD originated as a regional initiative in response to the Canterbury Earthquake Sequence in 2010 and 2011. Eleven years after its inception as the regional Canterbury Geotechnical Database, the nationwide NZGD now contains around 200,000 geotechnical records including Cone Penetration Tests (CPTs), Standard Penetration Tests (SPTs), shear vane tests, hand augers, and laboratory characterisation data.  The NZGD has served around 11,000 unique users in the science, engineering, council, insurance, and research sectors.  The database is used extensively by engineering consultants, researchers, and insurers to support development, hazard and risk assessments, and planning decisions.

This presentation will cover the unique origins of the NZGD and discuss factors that have led to the success of the NZGD, along with data insights regarding its current use and impact.  Looking ahead, a Future Vision project is underway in order to support stable governance, funding, and functionality, and ultimately ensure that the benefits of the NZGD are better realised across New Zealand. The Future Vision project identifies a wide range of aligned stakeholder groups, and outlines future benefits associated with interoperable data schemas. The project also describes opportunities, threats, and challenges that emerging digital technologies pose to the NZGD.  Emerging digital technologies are considered in terms of: (1) database structure and user model, and (2) data quality, transformation, and use.

Dr. Kaley Crawford-Flett is a geotechnical engineering specialist who promotes the advancement of engineering practice through industry-focused research, technical project management, and independent consultancy. She is presently employed by Engineering New Zealand as part of the Ministry of Building, Innovation, and Employment (MBIE) Building System Performance (BSP) partnership where she manages the technical delivery of BSP projects, including the New Zealand Geotechnical Database (NZGD) Future Vision Project. 

More broadly, Dr Crawford-Flett is the Chair of the New Zealand Society on Large Dams (NZSOLD) and holds an Honorary Academic position within the Civil and Environmental Engineering Department at the University of Auckland.  Dr Crawford-Flett’s research interests focus on managing internal erosion risks and geotechnical hazards, particularly for aging soil structures. Her professional research outcomes improve the understanding and management of aging infrastructure in New Zealand’s challenging geological and geomorphological environment. Dr Crawford-Flett is an appointed member of the Technical Working Group for New Zealand Building (Dam Safety) Regulation and serves as a reviewer for various academic journals.

Keynote Lecture 2

Adapting Database and Machine Learning to Reduce Uncertainty Associated with Geophysical Investigation Methods

Abstract: 

Geophysical investigation methods are increasingly getting popular as engineers come to pay more attention to 2D or 3D underground structures. Evaluating uncertainty or reliability of geophysical methods draws more attention compared with conventional in-situ geotechnical methods since the geophysical methods generally indirectly delineate underground structures from surface measurements in terms of inversion. Most geophysical inversions are essentially non-unique and it is very difficult to obtain unique, reliable, and high-resolution underground structure for engineering purposes without uncertainty. The presentation discusses the application of geophysical methods to engineering and environmental investigations from “Uncertainty” point of view and introduces the use of database and machine learning to reduce the uncertainty and increase reliability and resolution. We classify the cause of the uncertainty to several issues and discuss how to improve the reliability. Several blind tests to evaluate the uncertainty of geophysical methods are introduced. We developed relational database for geophysical investigation results to reduce uncertainty. The geophysical investigation results stored in the database were used as training data for machine learning. The presentation introduces the application example of supervised machine learning to seismic ambient noise data to estimate site class and amplification from an earthquake engineering point of view.    

Dr. Koichi Hayashi is presently a Senior Researcher at Geometrics, Inc. in San Jose, California and OYO Corporation in Japan.  Over the past 32 years, he has worked as a research geophysicist focusing on providing better tools and algorithms for near-surface geophysical methods.  He earned a M.S. degree in Earth Sciences from the Massachusetts Institute of Technology, and a Ph.D. in Earth Resources Engineering from Kyoto University in Japan.  His main research areas are seismic refraction, active and passive surface waves, and traveltime inversion. He is the author of the SeisImager one of the premier active and passive surface waves, refraction, and downhole data processing packages. In 2014, he was selected as the SEG Near-Surface Honorary Lecturer with his talk entitled “Integrated Geophysical Methods Applied to Geotechnical and Geohazard Engineering: From Qualitative to Quantitative Analysis and Interpretation”. Most recently, he was a contributing author to the textbook entitled “Seismic Ambient Noise”. He is currently in charge of a lecturer for a SEG training course “Passive Surface Wave Methods Using Ambient Noise: from Basic 1D Soundings to High-resolution 3D Imaging”.

Keynote Lecture 3

Learning About the Potential Directions of Data-centric Geotechnics from Materials Science

Abstract: 

In today's era of rapid technological advancement, the influence of machine learning and data-driven methodologies has left an indelible mark on various scientific disciplines. However, some fields have embraced this transformative wave more readily than others, and geotechnical engineering, as a critical pillar of societal infrastructure, has understandably adopted a cautious stance. The presenter, bridging the realms of data science and diverse fields such as materials science, neuroscience, and earthquake engineering, offers a unique perspective on the journey towards data-centric geotechnics. 

This presentation aims to elucidate the potential directions for shaping the future of geotechnical engineering by harnessing the capabilities of modern machine learning and big data analytics. By exploring the strategies and successes of materials informatics to accelerate materials discovery by leveraged data-driven techniques, we hope to lay the foundation for a data-driven paradigm in geotechnical problem-solving. We will discuss the issue from three perspectives: (1) new statistical theories on deep learning models, (2) similarities between materials informatics and geotechnics informatics, and (3) unique aspects of data-centric geotechnics.

Dr. Stephen Wu is an Associate Professor at The Institute of Statistical Mathematics (ISM), Tokyo, Japan. His primary research interest focuses on machine learning and Bayesian modeling for a wide range of applications, including materials informatics, bioinformatics, pharmacokinetics, molecular dynamics, geotechnical engineering, structural health monitoring and earthquake engineering. He received his MS and PhD degree from the California Institute of Technology, with a thesis studying engineering applications of Earthquake Early Warning (EEW) system by Bayesian inference. His postdoctoral research on hierarchical Bayesian models for molecular dynamics and pharmacokinetics problems was conducted at ETH-Zurich, Switzerland. Afterward, he began his academic career at ISM studying polymer design based on a machine learning framework. He shifted his focus to geotechnics after the collaborative research with Prof. Jianye Ching and Prof. Kok-Kwang Phoon on quasi-site-specific modeling for soil/rock properties starting in 2020. 

Keynote Lecture 4

Data-driven Predictive Railway Maintenance for Preventing Track Failures

Abstract:

Railway organizations typically employ various maintenance and repair techniques based on their existing infrastructure. Broadly, there are three types of interventions throughout the lifespan of a railway system: corrective, preventive, and predictive maintenance activities. Corrective maintenance is carried out in response to failures, such as replacing a broken railway track component. Preventive maintenance follows a predetermined schedule, involving tasks like ballast cleaning or integral renewal. Predictive maintenance is executed as required, relying on real-time collection and analysis of machine operation data to detect issues in their early stages before they can disrupt operations. With predictive maintenance, repairs are conducted while the machinery is operational, directly addressing existing problems. If a shutdown is necessary, it is shorter and more targeted.

In recent years, the Australian Railway Track Corporation (ARTC) has amassed extensive dynamic and static datasets from various sources, including service failure data, signal data, ballast history, grinding history, remedial action history, traffic data, and inspection data obtained from instrumented revenue vehicles, track geometry cars, Ground-Penetrating Radar, and ultrasonic testing. These datasets exhibit different characteristics, including discreteness or continuity, spatial or temporal attributes, and a range of signal and image data. Moreover, the database generated from continuous monitoring has grown significantly over time. However, these datasets have primarily been used for corrective and preventive maintenance purposes, relying on simplistic rules that are not entirely effective.

To manage and process such vast volumes of data and provide efficient maintenance planning, the utilization of Machine Learning techniques becomes essential. This keynote presentation discusses the development of a data-driven and Machine Learning-based approach for predictive railway maintenance, which has been implemented by ARTC in its Decision Support System. The following topics will be explored:

Dr. Jinsong Huang is a professor at the Priority Research Centre for Geotechnical Science and Engineering, Discipline of Civil, Surveying and Environmental Engineering, the University of Newcastle. His research interests include risk assessment in geotechnical engineering and computational geomechanics. He has published over 100 journal papers on the risk assessment of slope stability and landslides, the modelling of spatial variability, stress integration techniques for elastoplastic models, the contact dynamics of granular media, the analysis of hydraulic fracturing and the predictive maintenance of railway tracks. He received a Regional Contribution Award from the International Association of Computer Methods and Advances in Geomechanics at its international conference in Kyoto in 2014 and the GEOSNet Award from the Geotechnical Safety Network in 2017. He is an editorial board member for Georisk, Canadian Geotechnical Journal and Computers and Geotechnics. He has been invited to deliver several lectures at international conferences, universities and industry, and is a committee member on the ASCE Geo‐Institute’s Technical Committee on Risk Assessment and Management (RAM) and the ISSMGE Technical Committee (TC304) on Engineering Practice of Risk Assessment & Management. He is the chair of the Executive Board of the Geotechnical Safety Network. He served as the conference chair of the 8th International Symposium on Geotechnical Safety and Risk held at the University of Newcastle in December 2022.