⚡ electrical engineering

Parameter optimization, shaking table test, and intelligent response prediction of adjacent structures connected by viscous dampers

This study proposes and validates an integrated framework combining a genetic algorithm-optimized SAP2000 model, shaking table tests, and an SSA-BP neural network to optimize viscous damper parameters for adjacent structures, demonstrating that multi-location damper deployment significantly enhances seismic vibration control and enables accurate response prediction.

Original authors: Jiyun Zou, Chaoxian Wang, Haode Cheng, Yingxiong Wu

Published 2026-07-29

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Jiyun Zou, Chaoxian Wang, Haode Cheng, Yingxiong Wu

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine two tall buildings standing right next to each other in a crowded city. When the ground starts to shake during an earthquake, these buildings don't just wiggle; they dance to their own unique rhythms. If one building is tall and stiff while its neighbor is shorter and more flexible, they might swing out of sync. This is a bit like two dancers trying to move together but stepping on each other's toes, which can lead to a painful collision or "pounding" that damages both structures. Engineers have long known that adding a shock absorber—like a giant, high-tech car damper—between the buildings can help them move in harmony and soak up the shaking energy. But here's the tricky part: figuring out exactly where to put these dampers and how strong they need to be is a massive puzzle. If you guess wrong, the dampers might not help at all, or they could even make things worse. This is the challenge researchers face: how to tune these "seismic shock absorbers" perfectly so that neighboring buildings stay safe and steady during a quake.

In this study, a team of researchers from Fuzhou University decided to solve this puzzle using a mix of super-smart computer math, physical experiments, and artificial intelligence. They started by building a digital twin of two adjacent towers—one 15 stories high and the other 7 stories high—and used a "genetic algorithm" (think of it as a digital evolution process that tries thousands of random damper setups and keeps only the best ones) to find the perfect recipe. They discovered that placing dampers at specific floors and tuning their strength could significantly reduce the buildings' shaking. To make sure their computer math wasn't just a fantasy, they built a tiny, 1/20th-scale model of these towers and tested it on a giant shaking table in a lab. The results were promising: the real-world model behaved almost exactly like the computer predicted, and adding dampers cut the shaking of the main tower's top by up to 35% and the smaller tower's top by nearly 49%.

But the researchers didn't stop there. They realized that running a complex computer simulation every time a new earthquake hits is too slow for real-world emergencies. So, they taught an "intelligent" computer brain (a neural network) to predict how the buildings would react in a flash. By feeding this AI thousands of scenarios from their experiments, they created a model that could guess the maximum shaking and swaying of the buildings with over 92% accuracy, almost instantly. Their work suggests that using multiple dampers in different spots works better than just one, and that this smart combination of optimization, physical testing, and AI prediction offers a powerful new way to keep our cities safe when the ground starts to move.

The Digital Evolution and the Physical Test

The researchers approached the problem of connecting two different-sized buildings (a 15-story "main" structure and a 7-story "substructure") with viscous dampers. They knew that simply guessing where to put the dampers wasn't enough. Instead, they built a sophisticated system that combined MATLAB (a math software) with SAP2000 (a structural engineering program). They used a Genetic Algorithm (GA), which works a bit like natural selection. Imagine a computer that generates thousands of random ideas for where to place dampers and how strong they should be. It then "tests" these ideas in a virtual world. The ideas that work best—meaning they reduce the shaking the most—are kept and "bred" together to create even better ideas for the next round. This process repeats until the computer finds the absolute best setup.

They tested two main scenarios:

ASVD-1: A system with a single set of dampers.
ASVD-2: A system with two sets of dampers placed at different locations.

The computer determined that for the specific towers they were studying, the best single damper location was on the 7th floor, while the best two-damper setup involved placing them on the 1st and 7th floors. The optimal "strength" (damping coefficient) varied depending on the type of earthquake, ranging from 260 Ns/mm to 340 Ns/mm.

The Real-World Shake-Up

To prove their computer wasn't just dreaming, the team built a physical model. They constructed a 15-story main tower and a 7-story substructure using micro-concrete shear walls and copper frames, scaled down to 1/20th of a real building's size. They placed this model on a massive shaking table capable of simulating real earthquakes. They tested the model with three different famous earthquake records: El Centro, San Fernando, and SHW2.

The results from the shaking table were a major success. The physical model's behavior matched the computer's predictions incredibly well. The natural "vibration frequencies" of the real model were less than 5% different from the computer model, and the actual shaking responses (how much the floors moved) were within 10% of the simulation. This gave the researchers high confidence that their digital optimization method was reliable.

When they compared the dampened buildings to the undamped ones, the improvements were clear:

Under the El Centro earthquake: The single-damper system (ASVD-1) reduced the main tower's top displacement by 27.33% and acceleration by 24.01%. The double-damper system (ASVD-2) did even better, cutting displacement by 35.69% and acceleration by 26.98%. The smaller tower saw even bigger gains, with displacement dropping by nearly 49% with the double-damper setup.
Under the San Fernando earthquake: The double-damper system reduced the main tower's displacement by 47.68% and acceleration by 19.14%.
Under the SHW2 earthquake: The double-damper system reduced the main tower's displacement by 22.19% and acceleration by 20.03%.

The data consistently showed that ASVD-2 (the two-damper setup) was the winner. It provided better control over both the main and substructures than the single-damper setup, suggesting that spreading out the dampers to different floors helps the buildings work together more effectively to fight the shaking.

The Crystal Ball: AI Prediction

While the genetic algorithm found the best settings, running those complex simulations every time a new earthquake scenario arises is time-consuming. To speed things up, the researchers built an SSA-BP neural network. Think of this as a "crystal ball" trained on the data from their experiments and simulations.

They fed this AI model a massive dataset containing 100 different earthquake records (including far-field, near-field, and pulse-like waves) and various damper settings. The AI learned to predict four key outcomes:

Maximum displacement of the main structure (MSD).
Maximum acceleration of the main structure (MSA).
Maximum displacement of the substructure (SD).
Maximum acceleration of the substructure (SA).

The results were impressive. The AI model achieved a coefficient of determination (R²) greater than 0.92 for all four outcomes on both the training and testing data. In plain English, this means the AI's predictions were extremely close to the actual results, with very little error. For example, the Mean Absolute Percentage Error (MAPE) for predicting the main structure's acceleration was around 8-9%. This suggests that in the future, engineers could use this AI tool to instantly predict how well a damper system will work without needing to run hours of heavy computer simulations.

The Bottom Line

This study suggests that connecting adjacent buildings with viscous dampers is a highly effective way to reduce earthquake damage, provided the dampers are placed and tuned correctly. The researchers found that using a genetic algorithm to optimize the damper locations and strength works well, and that using two dampers at strategic locations (like the 1st and 7th floors in their model) is superior to using just one. They verified these findings with real-world shaking table tests, confirming that their computer models are accurate. Finally, they demonstrated that an AI model can learn from these results to predict future earthquake responses quickly and accurately, offering a promising tool for safer, more resilient city planning.

Technical Summary: Parameter Optimization, Shaking Table Test, and Intelligent Response Prediction of Adjacent Structures Connected by Viscous Dampers

Problem Statement
Rapid urbanization has led to reduced spacing between buildings, increasing the risk of pounding and detrimental dynamic interactions between adjacent structures during seismic events. This risk is particularly acute for asymmetric twin-tower systems with differing heights, stiffness distributions, and dynamic characteristics, which often exhibit poorly coordinated seismic responses. While viscous dampers are a recognized solution for suppressing pounding and dissipating energy, existing research faces three primary limitations: (1) reliance on simplified dynamic models that fail to capture complex spatial mechanical behaviors; (2) a scarcity of experimental validation for theoretically optimized parameters; and (3) a lack of systematic investigation into the applicability of optimized results across varying earthquake inputs and rapid prediction methods for structural responses.

Methodology
This study proposes a comprehensive framework integrating parameter optimization, experimental verification, and intelligent prediction to address the control of adjacent asymmetric twin-tower structures (a 15-story main structure and a 7-story substructure).

Optimization Framework (GA-SAP2000):
The authors developed an integrated optimization method combining the MATLAB Genetic Algorithm (GA) with the SAP2000 Open Application Programming Interface (OAPI). The objective was to determine the optimal installation locations and damping coefficients for viscous dampers. The process involved:
- Generating an initial population of candidate damper layouts encoded as chromosomes.
- Linking MATLAB to SAP2000 to perform dynamic time-history analysis for each candidate.
- Evaluating fitness based on structural responses (displacement, acceleration, and damping efficiency).
- Iteratively refining the population through selection, crossover, and mutation until convergence.
  This framework was applied to two configurations: a single-damper system (ASVD-1) and a double-damper system (ASVD-2).
Experimental Verification (Shaking Table Test):
To validate the finite element models and optimization schemes, a 1:20 scale shaking table test was conducted.
- Model Fabrication: A 15-story main structure and a 7-story substructure were constructed using micro-particle concrete shear walls and copper frames. Counterweights were added to simulate live loads and ensure dynamic similarity with the prototype.
- Similarity: The model adhered to specific geometric, material, and dynamic similarity coefficients (e.g., length scale 1/20, acceleration scale 3).
- Testing: The models were subjected to three distinct seismic records (El Centro, San Fernando, and SHW2) to evaluate the performance of uncontrolled (AS), single-damper (ASVD-1), and double-damper (ASVD-2) systems.
Intelligent Prediction (SSA-BP Neural Network):
To overcome the computational cost of repeated simulations and the limitations of physical testing, a hybrid neural network model was developed.
- Architecture: A Back-Propagation (BP) neural network optimized by the Sparrow Search Algorithm (SSA). The SSA was used to globally optimize the initial weights and biases of the BP network to avoid local optima and improve convergence.
- Inputs/Outputs: The model utilized inputs including damping coefficients, damping exponents, damper numbers, spatial configurations, and peak ground accelerations (scaled from 0.1g to 0.6g) across 100 seismic records. It predicted four key outputs: maximum displacement and acceleration for both the main and substructures.
- Dataset: A dataset generated from the validated finite element model was split into 80% training and 20% testing sets.

Key Results

Model Validation: The numerical finite element model demonstrated high reliability. The discrepancies in natural frequencies between the experimental and numerical models for the first two modes were within 5%, and errors in floor dynamic responses were less than 10%.
Vibration Control Performance:
- Compared to the uncontrolled system, the optimized viscous damper systems significantly reduced both displacement and acceleration responses.
- Under El Centro excitation, the double-damper system (ASVD-2) reduced the main structure's maximum displacement by 35.69% and the substructure's by 48.51%.
- The ASVD-2 configuration consistently outperformed the single-damper configuration (ASVD-1) across all seismic inputs and response metrics, indicating that multi-location collaborative deployment enhances vibration mitigation.
- Time-history analysis confirmed that ASVD-2 achieved greater reductions in Root Mean Square (RMS) values for displacement and acceleration compared to ASVD-1.
Prediction Accuracy: The SSA-BP neural network model achieved high prediction accuracy. The coefficient of determination ( $R^2$ ) for all four response indices exceeded 0.92 for both training and testing datasets, with the highest $R^2$ reaching 0.9634 for the main structure's acceleration during training. The Mean Absolute Percentage Error (MAPE) ranged from 7.32% to 12.31%.

Significance and Claims
The paper claims that its integrated approach effectively addresses the limitations of previous studies by combining rigorous optimization with experimental validation and intelligent prediction. Specifically:

The GA-SAP2000 framework is demonstrated to be a robust tool for coordinating damper location and damping coefficient optimization for complex adjacent structures.
The shaking table tests provide critical experimental evidence that the optimized parameters are not only theoretically sound but practically effective in reducing seismic responses.
The study highlights that installing multiple dampers at critical floors (multi-location deployment) offers superior control performance compared to single-damper systems.
The developed SSA-BP model offers a rapid and accurate method for predicting seismic responses under various parameter combinations and excitations, facilitating efficient design and assessment without the need for extensive new simulations or physical tests.

The authors conclude that this multi-faceted methodology provides both theoretical insights and practical guidance for the optimization design of viscous damper systems in adjacent structural engineering.

The Digital Evolution and the Physical Test

The Real-World Shake-Up

The Crystal Ball: AI Prediction

The Bottom Line

Technical Summary: Parameter Optimization, Shaking Table Test, and Intelligent Response Prediction of Adjacent Structures Connected by Viscous Dampers

More like this