1 Introduction

New technologies are advancing at an unprecedentedly accelerating pace over the years. The distance humanity has covered in 2200 years, from the Antikythera mechanism of ancient Greek world, the oldest known analogue computer, to the 4-bit first microprocessor in 1971, is not even comparable to the advancement of technology in the last 50 years. This dazzling journey of technological development has impacted all aspects of modern life, including industry.

Earthquake engineering is one of the disciplines that has embraced new technologies. Earthquake engineers, accustomed to dealing with highly nonlinear and dynamic problems that require complex mathematical and often iterative approaches, are called nowadays to summon dexterity on advanced coding, and masteries on statistics and handling of large amount of data. Artificial Intelligence, Sensing Technologies of all sorts, and Big Data Analytics emerge as essential tools for reducing uncertainty, facilitating engineering process and enhancing knowledge. This Special Issue is a manifestation of the fact that the new technologies can be useful for the most challenging problems of earthquake engineering, opening new prospects in the field.

1.1 Objectives of the special issue

This Special Issue is designed to present a selection of articles on “Artificial Intelligence, Sensing and Big Data Analytics in Earthquake Engineering” to the readers of the Bulletin of Earthquake Engineering. There are two basic objectives of this Special Issue. First, it serves as a platform to present applications of new technologies to practical earthquake engineering problems. Second, it creates awareness for the applicability of such technologies and encourages researchers to make use of these emerging methods more intensively.

When using the new technologies, especially those related to artificial intelligence and its derivatives, it is important to present the results in an explainable way. The physical basis of the addressed problem should be traceable throughout the presented work. This was one important criterion when evaluating the papers included in this Special Issue. Furthermore, a variety of problems have been chosen so that the readers can see a wide range of possible and successful applications of the new technologies and get inspired.

In this Special Issue, it was highlighted that the new technologies either largely improve the existing processes or radically alter the classical approaches used in earthquake engineering shifting the paradigm from “less data– brute computational force” towards “more data– smart algorithms”.

2 Overview of the contributions

An interesting application of artificial neural networks was presented by Kalakonas et al. (2022a and 2022b) where earthquake scenarios for building portfolios were analysed. In an earthquake loss estimation process, three basic components are amalgamated to estimate the risk; hazard, exposure and vulnerability. The paper by Kalakonas et al. (2022a) first addresses the hazard component, by trying to replace the classical multi-parameter regression models used for develo** ground motion models with an artificial neural network (ANN) ground motion model. In order to achieve that, they used a compiled database from two subsets of the Pan-European strong motion database (Bindi et al., 2014) and the NGA-West2 database (Ancheta et al. 2014). Their ANN model uses five input parameters as moment magnitude (Mw), depth, Joyner–Boore distance (Rjb), shear wave velocity in the top 30 m (Vs30), and the faulting type. Their model provides as output the RotD50 horizontal components of common intensity measures at 27 periods from 0.01 to 4.0 s. After establishing the ground motion model, Kalakonas et al. (2022b) presents results of their ANN model for risk estimation for a case study in Balkan region, following the 2019 Durres Earthquake Mw 6.4. They concluded that in overall, the ANNs led to damage and economic loss estimates closer to the observations.

Earthquake risk models are used for estimating the effects of earthquakes on a portfolio of buildings geographically distributed in a region of interest. Response of a single building to earthquakes may also be of interest sometimes. If this is the case, structural response reconstruction (SRR) would be one of the possible approaches, where the response of a structure to a specific earthquake motion is not calculated but instead predicted. Abdelmalek‑Lee and Burton (2023) introduced a dual model that uses kriging combined with the extreme gradient boosting (XGBoost) algorithm. Their aim was to reconstruct the seismic response demands in 207 buildings from 35 real earthquakes. They first used kriging for predicting peak ground accelerations (PGAs) at the location of the buildings, which were not instrumented. Then, they used measured PGAs from instrumented building sites with similar features, such as distance, soil type and magnitude. In the second phase, they used the XGBoost algorithm for reconstructing the maximum peak story drift ratio and peak floor acceleration in the buildings that are not instrumented.

Similar to estimating the building response, prediction of rocking blocks is also an important and challenging theme in earthquake engineering, since it concerns issues such as simple structures, heritage buildings, standing industrial objects or non-structural elements in buildings. Achmet et al. (2023) proposed a supervised machine learning (ML) model for quickly predicting the seismic response of rocking systems. They compared different supervised ML algorithms, such as the k-Nearest Neighbor (k-NN) and the Support Vector Machine (SVM). They compared the performances of different models by using sine pulses and different sets of natural ground motion records. They presented that the prediction accuracy is practically perfect for sine pulses. They also showed that, as far as prediction for natural records is concerned, accurate results were obtained.

A specific application on model updating at a cultural heritage building was presented in this Special Issue by Salachoris et al. (2023). They take the civic Clock Tower of Rotella as a case study, a structure damaged during the 2016 Central Italy seismic sequence and use a detailed numerical model of the tower. The paper presents a fully automated finite element model (FEM) updating procedure based on genetic algorithms and global optimization. Their results show that it is possible to automatically create a reference numerical model of the structure in its actual health, which then can be used to assess the dynamic performance of the building.

Marano et al. (2023) present a thorough review on generative adversarial networks (GANs), specifically for earthquake-related engineering fields. GAN models are particularly useful for generating reliable synthetic data that represent the characteristics of the actual sample set. The amount, variety and quality of data play a tremendous role in the success of prediction models. Their study presents a critical state-of-the-art review of GANs, explaining the most recent research into AI-based GAN synthetic generation of ground motion signals and seismic events. They also offer a comprehensive insight into seismic-related geophysical studies and possible use areas of GANs in that field.

When building a numerical model, calibration of the modelling parameters is sometimes possible if field or experimental data are available. This is a particularly useful approach for calibrating numerical models of heritage structures, since collection of data, especially using destructive methods, is limited or sometimes not even possible. Monchetti et al. (2023) uses historic masonry towers, where experimental data for model updating were available. They use Bayesian inference, which is frequently employed for addressing parameter uncertainty, observation errors and model inadequacy. Comparison of two different methods, namely Bayesian Model Updating (BMU) and Approximate Bayesian Computation (ABC), results in the conclusion that both methods provide similar results, with the ABC method being more flexible. Their study bridges a gap from field data to accurate numerical modelling, which can be helpful in numerous applications such as digital twinning and automated numerical modelling of structures.

Liu et al. (2023) used knowledge-enhanced neural networks (KENNs) to tackle the seismic analysis of bridges incorporating simplified component elements. The KENN is used for calibrating parameters of a lump plasticity model in their work. They used a long list of experimental results for identifying the key characteristics of reinforced concrete columns and their seismic response. These key parameters were then employed for supervising KENN model. They demonstrated the accuracy and efficiency of the proposed methodology by applying it to rapid seismic response analysis of typical bridges.

Remote sensing is one of the fastest develo** fields in the recent years. It is particularly useful for earthquake-induced structural damage on geographically distributed building stocks and cascading geological hazards. In particular, space-born remote sensing can complement field missions for providing valuable estimates of the spread and degree of damage for better coordinated response and recovery operations.

Giardina et al. (2023) present application of a novel method on integrating very high-resolution Synthetic Aperture Radar (SAR) data with building survey data for estimating building damages, as well as intensity-based detection of landslides. They presented a successful implementation of the proposed methods for 2021 Haiti Earthquake and Tropical Cyclone Grace.

The new computer technologies, both in the hardware and in the software side, are extremely powerful for image processing. One way of using the image-based methods is for collecting geometrical characteristics of a building using only photogrammetric data and automating the processing of this data for a specific purpose. Pantoja-Rosero et al. (2023) implement automation of the finite element modelling process using image data and machine learning methods. They present an image-based process where finite element model and mesh are automatically generated, equivalent-frame models of the outer walls of free-standing historical masonry buildings are created. They used and processed RGB (red-green-blue) images of the building in question using structure-from-motion algorithms. They created 3D geometries using convolutional neural networks (CNNs), by segmenting the openings and corners. They then used the generated layers for automatically producing a detailed model. They successfully tested their method on several masonry structures with irregular façades and surfaces.

3 Concluding remarks

Earthquake engineering is an inter-disciplinary engineering branch, the primary purpose of which is protecting human lives against catastrophic earthquakes. Whether designing new buildings and analysing existing structures for future earthquakes or organizing post-earthquake response and recovery, complex, often imperative, problems need to be faced. The challenging nature of these problems stem for the randomness of the earthquakes, as well as from the simple fact that there are earthquakes which can push structures up to and beyond their material limits.

New technologies can play a crucial role in advancing earthquake engineering by providing innovative tools and methods. Advanced simulation and modelling, remote sensing and geospatial analyses, sensors and sensor networks, structural health monitoring and many other technologies, which are not presented in this Special Issue, can assist earthquake engineering in tackling the huge societal responsibility of protecting human lives, and in reducing the likelihood of serious injury or property damage.

The Guest Editors hope that this Special Issue will be inspirational for other researchers who are willing to accept the challenge of improving the earthquake engineering field using the new technologies holistically and more intensively.