An official website of the United States government
Abstract Clustering analysis of sequence data continues to address many applications in engineering design, aided with the rapid growth of machine learning in applied science. This paper presents an unsupervised machine learning algorithm to extract defining characteristics of earthquake ground‐motion spectra, also called latent features, to aid in ground‐motion selection (GMS). In this context, a latent feature is a low‐dimensional machine‐discovered spectral characteristic learned through nonlinear relationships of a neural network autoencoder. Machine discovered latent features can be combined with traditionally defined intensity measures and clustering can be performed to select a representative subgroup from a large ground‐motion suite. The objective of efficient GMS is to choose characteristic records representative of what the structure will probabilistically experience in its lifetime. Three examples are presented to validate this approach, including the use of synthetic and field recorded ground‐motion datasets. The presented deep embedding clustering of ground‐motion spectra has three main advantages: (1) defining characteristics that represent the sparse spectral content of ground motions are discovered efficiently through training of the autoencoder, (2) domain knowledge is incorporated into the machine learning framework with conditional variables in the deep embedding scheme, and (3) the method results in a ground‐motion subgroup that is more representative of the original ground‐motion suite compared to traditional GMS techniques.
more »
« less
This content will become publicly available on April 1, 2026
Convolutional autoencoder-based ground motion clustering and selection
Ground motion selection has become increasingly central to the assessment of earthquake resilience. The selection of ground motion records for use in nonlinear dynamic analysis significantly affects structural response. This, in turn, will impact the outcomes of earthquake resilience analysis. This paper presents a new ground motion clustering algorithm, which can be embedded in current ground motion selection methods to properly select representative ground motion records that a structure of interest will probabilistically experience. The proposed clustering-based ground motion selection method includes four main steps: 1) leveraging domain-specific knowledge to pre-select candidate ground motions; 2) using a convolutional autoencoder to learn low-dimensional underlying characteristics of candidate ground motions’ response spectra – i.e., latent features; 3) performing k-means clustering to classify the learned latent features, equivalent to cluster the response spectra of candidate ground motions; and 4) embedding the clusters in the conditional spectra-based ground motion selection. The selected ground motions can represent a given hazard level well (by matching conditional spectra) and fully describe the complete set of candidate ground motions. Three case studies for modified, pulse-type, and non-pulse-type ground motions are designed to evaluate the performance of the proposed ground motion clustering algorithm (convolutional autoencoder + k-means). Considering the limited number of pre-selected candidate ground motions in the last two case studies, the response spectra simulation and transfer learning are used to improve the stability and reproducibility of the proposed ground motion clustering algorithm. The results of the three case studies demonstrate that the convolutional autoencoder + k-means can 1) achieve 100% accuracy in classifying ground motion response spectra, 2) correctly determine the optimal number of clusters, and 3) outperform established clustering algorithms (i.e., autoencoder + k-means, time series k-means, spectral clustering, and k-means on ground motion influence factors). Using the proposed clustering-based ground motion selection method, an application is performed to select ground motions for a structure in San Francisco, California. The developed user-friendly codes are published for practical use.
more »
« less
- Award ID(s):
- 2053741
- PAR ID:
- 10573397
- Publisher / Repository:
- Soil Dynamics and Earthquake Engineering
- Date Published:
- Journal Name:
- Soil Dynamics and Earthquake Engineering
- Volume:
- 191
- Issue:
- C
- ISSN:
- 0267-7261
- Page Range / eLocation ID:
- 109240
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Interest in earthquake resilience has increased in recent years, and the use of building cluster performance objectives has been shown to be an effective method for evaluating the resilience of built environment. A building cluster is a portfolio of buildings that share the same role in a community; its performance objectives are defined by considering earthquake scenarios, hazard levels, and individual building performance. The methodology presented in this paper employs performance-based assessments to estimate the probability of achieving building cluster performance objectives immediately following a seismic event. It can be used to assess the immediate post-earthquake community resilience in five steps: 1) hazard analysis, 2) conditional assessment of individual building performance, 3) conditional assessment of building cluster performance, 4) building cluster performance assessment by aggregation, and 5) earthquake resilience assessment of building clusters considering all hazard levels of interest. The design and extreme hazard levels are formulated using ground motion records selected based on the conditional spectra considering characteristics of earthquake scenarios and spatial correlation. Three performance objectives are defined for both individual buildings and building clusters: functionality, safe and usable during repair, and collapse prevention. Two engineering demand parameters – the maximum transient and the permanent interstory drift indices – are used to estimate individual building performance. The probability of achieving building cluster performance objective is calculated using the total probability theorem. The application of the proposed methodology is demonstrated using two clusters of reinforced concrete buildings, corresponding to ASCE 7 Risk Category II and IV structures, in San Francisco, CA.more » « less
-
ABSTRACT Earthquake-induced landslides can record information about the seismic shaking that generated them. In this study, we present new mapping, Light Detection and Ranging-derived roughness dating, and analysis of over 1000 deep-seated landslides from the Puget Lowlands of Washington, U.S.A., to probe the landscape for past Seattle fault earthquake information. With this new landslide inventory, we observe spatial and temporal evidence of landsliding related to the last major earthquake on the Seattle fault ∼1100 yr before present. We find spatial clusters of landslides that correlate with ground motions from recent 3D kinematic models of Seattle fault earthquakes. We also find temporal patterns in the landslide inventory that suggest earthquake-driven increases in landsliding. We compare the spatial and temporal landslide data with scenario-based ground motion models and find stronger evidence of the last major Seattle fault earthquake from this combined analysis than from spatial or temporal patterns alone. We also compare the landslide inventory with ground motions from different Seattle fault earthquake scenarios to determine the ground motion distributions that are most consistent with the landslide record. We find that earthquake scenarios that best match the clustering of ∼1100-year-old landslides produce the strongest shaking within a band that stretches from west to east across central Seattle as well as along the bluffs bordering the broader Puget Sound. Finally, we identify other landslide clusters (at 4.6–4.2 ka, 4.0–3.8 ka, 2.8–2.6 ka, and 2.2–2.0 ka) in the inventory which let us infer potential ground motions that may correspond to older Seattle fault earthquakes. Our method, which combines hindcasting of the surface response to the last major Seattle fault earthquake, using a roughness-aged landslide inventory with forecasts of modeled ground shaking from 3D seismic scenarios, showcases a powerful new approach to gleaning paleoseismic information from landscapes.more » « less
-
Abstract Post‐earthquake reconnaissance survey of structural damage is an effective way of documenting and understanding the impact of earthquakes on structures. This article aims at providing an efficient data‐based framework that reduces the required time for reconnaissance missions and predicts the damage intensities for every building in the affected region. We hypothesize that a joint selection of necessary structural and earthquake parameters along with sparse damage observations are sufficient to train a supervised learning algorithm and accurately infer the damage for other buildings in the region. Gaussian process regression is employed to prove the hypothesis for probabilistic inference of different damage indices. The algorithm performs efficiently by selecting a set of diverse and representative buildings for damage observations using K‐medoids clustering. To validate the hypothesis and the proposed method, the algorithm framework is implemented on two severe earthquake simulation testbeds. The impacts of different building and ground motion variables on the damage inference performance are discussed. Furthermore, the effectiveness of observation sampling by clustering in the post‐earthquake damage inference is compared with random sampling.more » « less
-
This paper investigates the effects of near-fault pulse-type ground motions on the structural response of a 3-story steel structure with nonlinear viscous dampers using the real-time hybrid simulation (RTHS) testing method. The real time loop of action and reaction between the experimental and numerical partitions executed in the RTHS enabled the accurate capturing of the velocity pulse effects of pulse-type ground motions. An ensemble of 10 natural pulse-type ground motions at the design basis earthquake (DBE) level is used for the RTHS. The accuracy of RTHS under high velocity loading is demonstrated, and thereby, is a validated method for experimentally investigation of the complicated structural behavior of structures with rate-dependent damping devices. The test results showed that the dampers are essentially effective in earthquake hazard mitigation effects involving pulse-type ground motions. The average peak story drift ratio under the set of pulse-type ground motions is 1.08% radians with a COV value less than 0.3, which indicates that the investigated structure would achieve the ASCE 7-10 seismic performance objective for Occupancy Category III structures under the DBE level pulse-type ground motions.more » « less
