A high-rise building can appear perfectly still while the ground beneath it is constantly moving. Traffic, trains, construction, ocean waves, atmospheric pressure and distant earthquakes generate a nearly continuous background of tiny vibrations known as seismic ambient noise. Individually, these signals are usually too weak to attract attention. But a new study presents them as a powerful, always-on source of information about the health of tall buildings, especially those standing on thick layers of soft sediment. By combining ambient-noise analysis with machine learning, Qin, Guo, Yuan and colleagues describe a new route toward monitoring skyscrapers without waiting for a major earthquake or relying only on expensive, controlled inspections.

The research, published in Communications Engineering, focuses on a central challenge in structural engineering: detecting small changes before they become visible damage. Buildings are designed to flex in the wind, respond to daily temperature changes and absorb vibrations from human activity. These normal movements can be much larger than the subtle signals associated with early deterioration. A monitoring system must therefore distinguish harmless environmental variation from changes in the building’s mechanical behavior. The problem becomes even more complicated when a high-rise is constructed on thick sediments, because the ground can amplify, prolong or reshape seismic motion before it reaches the structure.

Traditional structural health monitoring systems often depend on dedicated sensors that record acceleration, displacement or strain. Engineers then examine characteristics such as a building’s natural frequencies, damping ratios and mode shapes. A natural frequency is the rate at which a structure tends to vibrate, much like the tone produced by a musical instrument. If stiffness is reduced by cracking, connection failure or other damage, that frequency may shift. Yet frequency changes can also be caused by temperature, occupancy, wind intensity or variations in the surrounding soil. The scientific challenge is not simply to measure vibration, but to identify which changes carry meaningful evidence about structural condition.

Ambient seismic noise offers an unusual advantage because it is available continuously. Instead of waiting for a rare earthquake, researchers can use the weak vibrations generated during ordinary days and nights. Sensors installed in or around a building record these signals, and advanced processing can extract repeating patterns from the noise. One important technique is seismic interferometry, in which recordings from different sensors are cross-correlated to estimate how vibrations travel between them. Repeating this process over time can reveal whether the building’s response is stable or whether its internal dynamics are gradually changing.

The machine-learning component is designed to handle the enormous complexity of that data. A high-rise may generate millions of measurements across multiple floors, directions and time windows. Algorithms can learn the statistical fingerprint of normal behavior and then flag observations that fall outside the established pattern. Depending on the model, the input may include spectral peaks, changes in vibration amplitude, correlations between sensors, apparent wave speeds and the relationship between building motion and ground motion. Rather than relying on one threshold, machine learning can combine these features to identify subtle, multidimensional anomalies that conventional monitoring might overlook.

The thick sediment beneath a building is not a minor detail. Sedimentary layers can behave differently from the underlying bedrock, particularly when they are soft, deep or irregularly structured. They may amplify certain frequencies, trap seismic energy and cause motion to persist after the strongest part of an earthquake has passed. The building and the sediment can also interact, producing a coupled system in which the structure modifies the motion of the ground while the ground influences the building’s vibration. A monitoring approach that ignores this interaction could mistake a site effect for structural damage. By incorporating the seismic environment into the analysis, the study addresses a key source of uncertainty in urban earthquake engineering.

The researchers’ approach is especially relevant to cities filled with tall buildings but limited opportunities for intrusive inspection. A skyscraper cannot easily be shut down for repeated testing, and visual surveys may miss damage hidden inside columns, beams, foundations or joints. Ambient-noise monitoring, by contrast, can operate while occupants continue their daily routines. Continuous records could allow engineers to compare a building with its own historical baseline, examine how it responds to storms or earthquakes and prioritize inspections when the data indicate an unusual change. In a future emergency, such systems could help authorities rapidly identify structures that require closer examination before people are allowed to return.

The appeal of the method is not that machine learning replaces engineers, but that it can act as an early-warning layer between raw measurements and expert decisions. Algorithms are highly effective at finding patterns, but they can also be misled by changes in sensor calibration, renovations, new mechanical equipment or unusual occupancy. A reliable system must therefore be trained on long records, tested against independent observations and designed to explain why an alert was issued. Engineers still need to determine whether an anomaly reflects actual damage, a temporary environmental effect or a problem with the monitoring equipment itself.

The study also points toward a broader transformation in how cities may manage aging infrastructure. Bridges, tunnels, dams and towers are increasingly being equipped with networks of low-cost sensors, creating large streams of information that were previously unavailable. Machine learning can help convert those streams into estimates of structural condition, while seismic methods provide a way to interrogate buildings using vibrations that are already present in the environment. The combination could be particularly valuable in dense urban regions where earthquakes, soft ground and rapid construction growth overlap.

For the public, the most striking implication is simple: a building may be capable of revealing its condition through the faint tremors of the city around it. Every passing vehicle, distant storm and small ground vibration contributes to a complex but potentially informative signal. The work by Qin, Guo, Yuan and colleagues suggests that, with the right sensors and carefully trained algorithms, that background noise can become a continuous health record for a high-rise. The technology is not a crystal ball, and it cannot eliminate the need for inspections or sound construction. But it could give engineers something they have never had at this scale: a persistent, data-driven view of how a building’s behavior evolves long before damage becomes obvious.

Subject of Research: Machine learning–driven structural health monitoring of a high-rise building on thick sediments using seismic ambient noise.

Article Title: Machine learning–driven structural health monitoring of a high-rise building on thick sediments via seismic ambient noise.

Article References: Qin, L., Guo, Z., Yuan, T. et al. “Machine learning–driven structural health monitoring of a high-rise building on thick sediments via seismic ambient noise.” Communications Engineering (2026). https://doi.org/10.1038/s44172-026-00756-8

Image Credits: AI Generated

DOI: 10.1038/s44172-026-00756-8

Keywords: structural health monitoring, machine learning, seismic ambient noise, high-rise buildings, thick sediments, earthquake engineering, building dynamics, seismic interferometry, infrastructure safety

Tags: AI-driven vibration analysis in civil engineeringambient noise-based structural integrity assessmentearly damage detection in tall buildingshigh-rise building vibration detectionmachine learning for structural health monitoringmonitoring building health without major earthquakesnon-invasive seismic monitoring techniquesseismic ambient noise analysisseismic data analysis for urban infrastructureseismic noise amplification on sedimentary foundationssoft sediment ground impact on skyscraper stabilitysustainable building maintenance using ambient seismic data