Generalised Time-Series Analysis of Fault Mechanics Using Explainable AI

AI can help predict earthquake precursors: The study used machine learning to identify three distinct warning phases before fault failure occurs—from initial microcracking through fault growth to final coalescence—suggesting that AI analysis of acoustic signals could potentially improve earthquake hazard forecasting.

Explainability matters for scientific understanding: Rather than treating the AI model as a "black box," the researchers used genetic algorithms to make the model's decision-making transparent, showing which physical parameters (like wave scattering and stress distribution) were most important for predicting fault behavior—demonstrating that responsible AI in science requires interpretable models.

Multiple data types strengthen predictions: The model combined different types of information (waveform characteristics, timing patterns, and spatial distributions) to achieve accurate fault stage characterization, illustrating that responsible AI systems benefit from diverse, complementary data sources rather than relying on single metrics.

Context-dependent predictions across conditions: The AI successfully predicted fault timing and magnitude across different pressure levels (5-40 MPa) and failure types (axial splitting versus shear), suggesting that well-designed AI models can generalize insights across varying real-world conditions rather than working in only narrow scenarios.