Evaluating and Explaining Earthquake-Induced Liquefaction Potential through Multi-Modal Transformers

2025-02-11

AI Summary

This study addresses the challenge of predicting seismic soil liquefaction risk by proposing the first interpretable multimodal Transformer model specifically designed for geotechnical engineering. Methodologically, it integrates earthquake ground-motion spectra (encoded via FFT), stratigraphic layering (represented through an LLM-inspired hierarchical tokenization scheme), and site-specific geotechnical features; cross-modal interactions are modeled via parallel multi-head self-attention, and SHAP is incorporated to ensure physically meaningful local interpretability. Key contributions include: (1) the first domain-specific multimodal architecture for geotechnical applications; (2) a 93.75% accuracy under cross-regional k-fold cross-validation; (3) successful out-of-distribution prediction of liquefaction spatial distribution during the 2024 Noto Peninsula earthquake; and (4) real-time, multi-site web deployment capability. The model achieves high predictive accuracy, strong generalizability across diverse geological settings, and practical engineering interpretability.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

This study presents an explainable parallel transformer architecture for soil liquefaction prediction that integrates three distinct data streams: spectral seismic encoding, soil stratigraphy tokenization, and site-specific features. The architecture processes data from 165 case histories across 11 major earthquakes, employing Fast Fourier Transform for seismic waveform encoding and principles from large language models for soil layer tokenization. Interpretability is achieved through SHapley Additive exPlanations (SHAP), which decompose predictions into individual contributions from seismic characteristics, soil properties, and site conditions. The model achieves 93.75% prediction accuracy on cross-regional validation sets and demonstrates robust performance through sensitivity analysis of ground motion intensity and soil resistance parameters. Notably, validation against previously unseen ground motion data from the 2024 Noto Peninsula earthquake confirms the model's generalization capabilities and practical utility. Implementation as a publicly accessible web application enables rapid assessment of multiple sites simultaneously. This approach establishes a new framework in geotechnical deep learning where sophisticated multi-modal analysis meets practical engineering requirements through quantitative interpretation and accessible deployment.

Problem

• Predict earthquake-induced soil liquefaction potential
• Integrate multi-modal data for geotechnical analysis
• Achieve interpretability in deep learning models

Innovation

• Multi-Modal Transformers for soil analysis
• SHAP for prediction interpretability
• Web application for rapid site assessment