Wavelet Packet-Based Diffusion Model for Ground Motion Generation with Multi-Conditional Energy and Spectral Matching
summary
The gist
Temporal energy distribution strongly affects nonlinear structural response and cumulative damage <ref:2602.07405#pg2> The authors propose a multi-conditional diffusion framework for ground motion…
In short
The authors developed a diffusion model framework to create realistic ground motions that simultaneously match target response spectra and temporal energy evolution. By using wavelet packet decomposition for signal representation and a Transformer encoder with multiple conditions like Arias intensity and Husid curves, the method significantly improved control over the onset and duration of energy release compared to simpler spectral matching methods.
Key concepts
- Multi-conditional Diffusion Framework
- This is a generative AI approach that uses multiple constraints—such as response spectra, Arias intensity, and temporal parameters—to guide a diffusion model. Instead of just matching the overall shape (spectrum), it forces the generated motion to also follow specific rules about when energy starts and how it changes over time.
- Wavelet Packet Decomposition
- This technique breaks down complex ground motion signals into different frequency subbands across various time scales. It provides a detailed, multiscale representation of the waveform, which is then used to reconstruct the signal directly from generated coefficients using orthogonal filter banks without needing iterative phase retrieval.
- Elucidating Diffusion Model (EDM)
- The EDM is a specific design strategy for diffusion models that systematically decouples different parts of the model, like noise scheduling and loss weighting. This allows researchers to precisely control how the model learns from noisy data, making it more efficient and effective for generating high-quality synthetic data.
- Conditional Diversity Sampling
- This technique is used to quantify uncertainty in the generated ground motions. By sampling different versions of a motion under similar constraints, the framework can show how much variability exists in the output, helping engineers understand the reliability of their synthesized seismic inputs.
Terminology used across episodes
This episode discusses
- Wavelet Packet-Based Diffusion Model for Ground Motion Generation with Multi-Conditional Energy and Spectral Matching · Paper Radio
- Auto-Encoding Variational Bayes
- Generative Adversarial Networks
- Learning Physics for Unveiling Hidden Earthquake Ground Motions via Conditional Generative Modeling
- Denoising Diffusion Implicit Models
- High-Resolution Image Synthesis with Latent Diffusion Models
- Adding Conditional Control to Text-to-Image Diffusion Models
- High Resolution Seismic Waveform Generation using Denoising Diffusion
- Broadband Ground Motion Synthesis by Diffusion Model with Minimal Condition
- Elucidating the Design Space of Diffusion-Based Generative Models
The paper
Wavelet Packet-Based Diffusion Model for Ground Motion Generation with Multi-Conditional Energy and Spectral Matching · Read on arXiv
Yi Ding, Su Chen, Jinjun Hu, Xiaohu Hu, Qingxu Zhao, Xiaojun Li
National Key Laboratory of Precision Blasting, Jianghan University · Key Laboratory of Urban Security and Disaster Engineering of the Ministry of Education, Beijing University of Technology, Beijing · Key Laboratory of Earthquake Engineering and Engineering Vibration, China Earthquake Administration · Institute of Geophysics, China Earthquake Administration
Temporal energy distribution strongly affects nonlinear structural response and cumulative damage. We propose a multi-conditional diffusion framework for ground motion synthesis that simultaneously matches temporal energy evolution and target response spectra. Wavelet packet decomposition provides the signal representation and enables direct waveform reconstruction via orthogonal filter banks. A Transformer-based conditional encoder with cross-attention integrates heterogeneous conditions, including spectral ordinates, Arias intensity, temporal parameters, and Husid curves. The framework adopts the Elucidating Diffusion Model (EDM) with second-order Heun sampling to improve inference efficiency without sacrificing quality. Tests on the NGA-West2 database show that explicit temporal-energy constraints markedly improve control of energy onset and significant duration while preserving spectrum matching and maintaining stable diversity sampling. The framework yields spectrum-compatible motions with realistic energy evolution and supports uncertainty quantification via conditional diversity sampling.
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: We're looking at the paper: "Wavelet Packet-Based Diffusion Model for Ground Motion Generation with Multi-Conditional Energy and Spectral Matching".
Mira: Here's the gist of the paper. Temporal energy distribution strongly affects nonlinear structural response and cumulative damage <ref:2602.07405#pg2> The authors propose a multi-conditional diffusion framework for ground motion synthesis that simultaneously matches temporal energy evolution and target response spectra <ref:2602.07405#pg2>.
Kai: Framework Overview The proposed framework jointly constrains spectral and energy-evolution characteristics by using a comprehensive conditional set including the response spectrum, Arias intensity, energy time parameters, and the Husid curve <ref:2602.07405#pg2>.
Mira: Wavelet packet decomposition maps waveforms into multiscale time-frequency representations <ref:2602.07405#pg2>.
Kai: Based on orthogonal filter banks, time-domain waveforms are reconstructed directly from generated wavelet coefficients, avoiding iterative phase retrieval <ref:2602.07405#pg2>.
Paper summary: Kai: So, to summarize this paper, they are proposing a multi-conditional diffusion framework for ground motion synthesis that aims to simultaneously match temporal energy evolution and target response spectra >
Mira: They start by saying that matching both is an inherently ill-posed inverse problem because multiple acceleration time histories can satisfy the same spectral target but have different physical characteristics like amplitude envelope or temporal evolution of energy >
Lev: And they address this by using wavelet packet decomposition to represent the signal, which lets them work in coefficient space, and then a Transformer-based conditional encoder to fuse all those different conditions into the U-Net denoiser via cross-attention >
Kai: They use a comprehensive set of conditions including spectral ordinates, Arias intensity IA, temporal parameters like T5 and D5-ninety-five the temporal centroid Eth, and the Husid curve H(t) > <ref:2602.07405#pg3>
Mira: The core idea is to start from pure Gaussian noise at maximum noise level sigma max and progressively denoise it to generate clean samples conforming to a target distribution pdata(xc) >
Lev: So they're using the Elucidating Diffusion Model EDM framework with Heun sampling for the sampling process, which they say keeps inference efficient while maintaining quality >
Kai: Their main finding is that explicitly including these temporal-energy constraints leads to spectrum-compatible motions with realistic energy evolution and supports uncertainty quantification via conditional diversity sampling >
Conclusion: Kai: So looking at the whole paper, "Wavelet Packet-Based Diffusion Model for Ground Motion Generation with Multi-Conditional Energy and Spectral Matching," it shows they managed to build a system that controls both the spectral matching aspect and the temporal energy evolution simultaneously >
Mira: What this means in plain terms is that we're moving past just generating motions that hit a certain frequency response; we can now generate motions that evolve physically in time, which is what really matters for damage assessment >
Lev: From my side, it’s interesting how they are using wavelet packet decomposition to handle the long-range temporal dependencies and transient local details in complex waveforms >
Kai: And from a hardware perspective, the way they structure the Transformer encoder to inject those multiple conditional inputs via cross-attention seems like a robust way to handle all that different information >
Mira: But they did flag something important, which is that the wavelet packet representation assumes a fixed time-frequency resolution across all scales, and they suggest that this might be suboptimal for ground motions with highly transient features >
Lev: And what about the relationship between how accurate those conditioning parameters are—like IA or T5—and what happens to the downstream structural response prediction if you get that wrong >
Kai: That's a point worth following up on, because if we can control the energy evolution this precisely, it should translate into better predictions for structural response >
More episodes
- 2611.14082-Two atoms with strong light-induced dipolar interactions
- 2611.14040-Characterization of BNLs 20 and 30 micrometer thick LGADs for the Electron Ion Collider
- 2611.13399-Lift-element-based dynamic network of vortex gust-airfoil interactions
- 2611.13476-Hopfion Superlattices in 3D Twisted Moir'e Light Fields
- 2611.13642-Investigation of the timing performance of double-sided p-type 3D silicon sensors produced by CNM
- 2611.13830-SiPM optical readout of a PICOSEC Micromegas precise timing detector
- 2611.14152-Where are the shortest-path nodes in spatial networks?
- 2611.14219-Sub-Ice: semi-automated mapping of basal channels from ice shelf surface elevation data
- 2611.16341-SUMNO: Spatially Unrolled Multigrid Neural Operator for Fluid Flow Surrogates
- 2611.16090-Multi-UTE Myelin Fraction Mapping: Repeatability and Cross-Field-Strength Reproducibility at 1.5T and 3T