The work, carried out by Shandy Rianto and Xinzheng Lu of the Department of Civil Engineering at Tsinghua University in Beijing, addresses a stubborn problem in regional seismic damage assessment. Traditional regional models typically describe buildings using a handful of coarse attributes—height, construction era, occupancy class—and then map those attributes to damage states through simplified relationships. The approach is computationally cheap, but it flattens away the very details that govern how a structure actually behaves under strong shaking: the arrangement of its frames, the distribution of member stiffness, the topology of its lateral-force-resisting system.

At the other extreme sit multi-degree-of-freedom, or MDOF, models, which condense a building into a stack of lumped masses connected by nonlinear shear or flexural springs. These models can reproduce hysteretic behavior—strength and stiffness deterioration under cyclic loading—reasonably well, but they must be parameterized from limited structural data. When those parameters are assigned from generic statistical rules rather than from the building’s actual design, the simulated response can drift far from reality. Meanwhile, machine learning surrogates trained to imitate finite-element analysis have grown popular, but most of them ingest buildings as fixed-length feature vectors. A vector, however long, cannot natively capture the fact that a structure is a network of connected members whose behavior depends on adjacency, hierarchy, and load paths.

The Tsinghua team’s insight was to represent each building as a graph. In this representation, nodes correspond to structural components or stories, and edges encode the physical connections between them—beam-to-column joints, story-to-story coupling, lateral load transfer. Node features carry engineering attributes such as member dimensions, material properties, and story masses, while edge features describe connectivity and geometry. Graph neural networks, or GNNs, are uniquely suited to such data because their message-passing architecture propagates information along the actual edges of the structure. During each layer of the network, a node aggregates features from its neighbors, transforms them, and passes the result onward. After several rounds of message passing, each node’s embedding reflects not just its own properties but the structural context in which it sits—exactly the kind of relational reasoning that fixed vectors struggle to supply.

The framework the researchers developed is not a single model but a spectrum of data-physics integration strategies, offering three distinct surrogate modeling approaches. Two of them are hybrid schemes that retain an MDOF model as the physics backbone and use graph-based learning to correct or enrich its predictions. In effect, the MDOF simulation provides a physically consistent first estimate of the seismic response, and the GNN learns the systematic discrepancies between that estimate and the true nonlinear finite-element response, using the graph-structured description of each building to explain why the simplified model falls short for a particular structure. This division of labor matters: the physics component constrains the prediction to plausible dynamic behavior, while the data component supplies building-specific nuance that generic parameterization cannot capture.

The third approach is more ambitious—a physics-aware, end-to-end GNN that estimates seismic responses directly from the graph representation, embedding structural mechanics knowledge into the network’s architecture and training rather than relying on an intermediate simulation step. This end-to-end route eliminates the need to run even a simplified MDOF simulation at prediction time, promising the greatest speed, while the embedded physics keeps the network from learning spurious correlations that would fail under earthquakes unlike those in its training set.

To train and validate these models, the researchers drew on established datasets of seismic designs, nonlinear models, and response simulations for steel moment-resisting frame buildings—a class of structure whose ductile steel connections and frame-based lateral systems make them a natural testbed for graph methods. Steel moment frames were modeled with nonlinear hysteretic behavior consistent with the deterioration phenomena documented in the earthquake engineering literature, and the ground-motion demand was characterized through intensity measures that capture spectral shape, not just peak acceleration. The graph-based structural representation dataset compiled for the study has been deposited in the Science Data Bank, making both the dataset and the GNN algorithms available to other researchers—a transparency move that matters in a field where reproducibility of machine learning benchmarks has often lagged behind headline accuracy claims.

The benchmark comparisons are where the framework makes its case. Against traditional parameterization-based MDOF modeling—the standard practice when only coarse building data are available—the proposed approaches reduced response estimation errors by more than 50 percent. They also outperformed commonly used data-driven surrogates that rely on fixed-length vector features, confirming the central hypothesis: the accuracy bottleneck in regional machine learning for earthquakes is not merely the amount of training data but the adequacy of the structural representation itself. When the model can see the building as a connected system rather than as a list of attributes, it learns correspondences that map far more faithfully onto actual structural behavior.

The timing of this work is no accident. Cities worldwide are generating increasingly rich digital descriptions of their building stock—digital twins, semantic city information models, and detailed urban survey databases that record geometry, materials, and structural systems at a fidelity unimaginable when the first regional loss estimation tools were assembled in the 1980s. The missing ingredient has been a computational method that can exploit this detail at scale. Full finite-element analysis of a million-building portfolio remains impractical even with modern GPU clusters, but a trained GNN performs inference in a fraction of a second per building, once trained. The hybrid framework thus occupies a sweet spot: fidelity approaching refined structural simulation, speed approaching that of the crudest empirical tools.

The practical stakes extend well beyond academic benchmarks. Rapid post-earthquake damage assessment determines where rescue teams are sent first, which hospitals are presumed operational, and how emergency budgets are allocated in the hours and days after a major event. Pre-earthquake risk mapping, in turn, drives retrofit prioritization and insurance pricing for entire metropolitan building stocks. In both scenarios, systematic over- or under-estimation of damage has real consequences: misallocated resources, mispriced risk, and misplaced confidence in buildings that may not perform as the regional model assumes. A surrogate that halves the estimation error without slowing the analysis down changes the reliability of every downstream decision.

There are, of course, caveats inherent to the approach. Surrogate models inherit the distributional boundaries of their training data; a GNN trained on code-compliant steel moment frames cannot be trusted, without retraining, to predict the response of unreinforced masonry or irregular older buildings that behave in fundamentally different ways. The physics-aware design mitigates but does not eliminate extrapolation risk, which is precisely why the authors emphasize the hybrid tiered structure—users can choose the level of data-physics integration appropriate to their data availability and confidence requirements. The authors also note the framework’s applicability hinges on the quality of the urban-scale databases feeding it, though the trend line for such data availability is clearly pointing upward.

What the study ultimately demonstrates is a conceptual shift in how the field thinks about structural representation for machine learning. For a decade, progress in seismic surrogates came largely from better regression architectures stacked on top of the same hand-engineered feature vectors. This work argues that the representation itself—whether the model perceives a building as a graph of connected physical components—is a first-order determinant of accuracy. Given the field’s trajectory, graph-based hybrid surrogates are likely to spread quickly beyond steel moment frames to shear-wall buildings, reinforced concrete frames, and increasingly to million-scale urban clusters, where related AI frameworks are already being developed for city-scale simulation.

The research was supported by the National Natural Science Foundation of China. As urban populations in seismically active regions continue to grow, tools that can predict, quickly and accurately, how an entire city’s buildings will respond to shaking are moving from academic aspiration to operational necessity. This study offers a concrete, open, and demonstrably accurate step in that direction—teaching machines to see buildings the way structural engineers do: as systems, not statistics.

The work, published in Earthquake Engineering and Engineering Vibration, addresses a computational bottleneck that has long frustrated the seismic assessment of nonstructural components, the category that includes furniture, mechanical equipment, storage racks, and countless other objects whose failure during earthquakes is often more dangerous than damage to the buildings themselves. The problem is fundamentally one of scale and interaction. To predict whether a bookshelf on the fifth floor of a fifteen-story apartment building will topple, an engineer cannot simply analyze the bookshelf in isolation. The bookshelf sits atop a floor that amplifies and distorts the ground motion beneath the building. Floors high up in tall structures experience accelerations several times larger than those at ground level, and the character of that motion—its frequency content, its duration, its peaks—changes in ways that depend on both the building’s dynamic properties and the earthquake’s own signature.

The classical approach to this problem requires two nested analyses. First, engineers must perform a nonlinear time-history analysis of the entire building, feeding it a recorded or synthetic ground motion and computing how every floor responds second by second. Then, they must take the resulting floor motion and use it as input to a separate rocking-body dynamic analysis, tracking the object’s rotation angle as it pivots, impacts, and—if things go badly—tips past its critical angle. Each of these analyses is computationally demanding in its own right, and the rocking problem is notoriously ill-conditioned. Unlike linear structural systems, a rocking body’s response is extremely sensitive to tiny changes in initial conditions and input motion: two nearly identical earthquakes can produce wildly different outcomes for the same block, a chaotic quality that has made rocking motion famously difficult to predict since George Housner’s foundational work on inverted pendulum structures in 1963.

Compounding the difficulty, the interactions among the three governing factor families—seismic motion characteristics, structural amplification effects, and the geometric properties of the rocking body—are deeply nonlinear and resist the simple empirical equations that engineers have traditionally relied upon. A taller, slenderer block has a lower critical rocking angle and a longer characteristic rocking frequency; a squat block is more stable but slides more readily. Meanwhile, floor motions embed the amplified response of the building’s higher modes, and strong-motion duration influences how many rocking cycles an object experiences before shaking subsides. Capturing all of this with a formula has proven elusive.

The Chinese team’s solution is to sidestep the physics simulation entirely, at least at prediction time, by letting a deep neural network learn the mapping directly from data. The researchers began with city-scale nonlinear time-history analyses—the same simulation technology used to model how thousands of buildings across an entire urban area respond simultaneously to a scenario earthquake. From these analyses, they extracted the seismic responses at building floors, which then served as excitation for rocking-body dynamics calculations. For every combination of ground motion, building, and floor level, they computed the corresponding rocking response and recorded whether the body overturned. This process generated a multidimensional database spanning a wide range of ground-motion intensity measures, building heights, and rigid-body geometries.

Ground-motion intensity measures deserve a brief explanation, because they are the language in which engineers compress a complicated earthquake into a handful of numbers. Quantities such as peak ground acceleration, peak ground velocity, spectral accelerations at various periods, Arias intensity, and cumulative absolute velocity each summarize different aspects of a record—its amplitude, its energy content, its duration—and decades of research have explored which of these measures best correlates with damage of various kinds. For rocking bodies, the choice is particularly delicate, since overturning depends on a subtle interplay of all of these factors rather than on any single one. By incorporating a rich suite of intensity measures into the input features, the researchers gave their neural network access to the multiple dimensions along which earthquake danger manifests.

On top of this database, the team trained a deep neural network to perform what is essentially a binary classification task: given the intensity measures characterizing the ground motion, the height of the building, the floor level, and the geometric parameters of the rigid body, predict whether the object will overturn. Deep neural networks excel precisely where conventional regression struggles, because their layered architecture can represent highly nonlinear, high-dimensional relationships among inputs. Each layer of the network transforms its inputs through learned weights and nonlinear activation functions, allowing successive layers to build up increasingly abstract representations of the underlying physics—representations that no human analyst would need to specify by hand.

The results are striking. The trained model achieved 94.37 percent accuracy on its test set, exceeding the performance of conventional machine learning approaches such as random forests, support vector machines, and decision trees, which the authors used as benchmarks. Just as importantly for practical deployment, the network achieves high computational efficiency: once trained, it produces predictions essentially instantaneously, whereas a full coupled building-and-rocking-body analysis can take minutes to hours per case. For risk analysts who need to estimate, across an entire city, how many thousands of cabinets, servers, and shelves will topple in a given earthquake scenario, that difference is transformative.

The team also explored dimensionality reduction—mathematical techniques that compress the input feature space while discarding redundant information. Remarkably, reducing the number of input features decreased both training time and model complexity while preserving strong predictive performance. This finding matters for two reasons. Practically, it means the method can run on more modest computing resources and be more easily embedded in rapid assessment workflows. Scientifically, it suggests that a relatively small subset of intensity measures and geometric parameters captures most of the information relevant to overturning—a valuable insight for future fragility studies of building contents.

The work builds on a growing body of research applying artificial intelligence to seismic response prediction, and it arrives at a moment when the field of urban earthquake simulation is maturing rapidly. City-scale nonlinear time-history analysis, which underpins the new study’s training database, has itself only become feasible in recent years through advances in computational structural engineering, and it now serves as the engine for post-earthquake damage assessment systems that can estimate regional losses within hours of a real event. Extending that simulation capability to nonstructural contents through machine learning closes an important gap, because previous rapid assessment frameworks focused almost exclusively on the buildings themselves.

The implications extend well beyond furniture. Critical facilities—hospitals, power substations, semiconductor fabs, emergency operations centers—are filled with free-standing and lightly anchored equipment whose overturning can trigger system-level failures far costlier than any structural damage. Museum collections, cultural heritage artifacts, and laboratory chemicals represent other categories where rocking and overturning carry severe consequences. A rapid, accurate overturning prediction tool enables engineers to generate fragility curves—probabilistic statements about the likelihood of overturning as a function of shaking intensity—for entire inventories of contents, which in turn feeds into loss estimation, retrofit prioritization, and the design of protective measures such as anchoring, base isolation pads, and tie-down systems.

The study was supported by the National Natural Science Foundation of China and the Institute of Engineering Mechanics of the China Earthquake Administration, among other funders. Its authors—Qingle Cheng, Zhengxuan Song, Linlin Xie, and Yuan Tian—position the approach as an efficient and intelligent framework for the seismic assessment and risk analysis of rocking components in buildings, and the broader trajectory of the research suggests a future in which earthquake early warning systems, city-scale simulators, and machine learning predictors work in concert to warn not only about collapsing structures but about the contents inside them. In an era when most earthquake deaths in countries like Japan and the United States stem from nonstructural failures rather than structural collapse, an algorithm that can instantly flag which objects in which buildings are likely to topple may prove to be one of the more consequential applications of deep learning in the earthquake engineering toolbox.