AI predicts underground cracks using real physics rules.

Engineers need to monitor hidden damage inside bridges and roads before dangerous failures happen. They often use ground-penetrating radar, which sends radar waves underground to spot hidden defects, but forecasting how that subsurface damage will grow is challenging. Researchers built a specialized physics-informed neural network to interpret and predict this radar data. Instead of acting like a standard computer program that only guesses patterns from raw images, this model builds the actual physical laws of electromagnetic wave propagation directly into its code. It treats underground structures similarly to how medical scans monitor tumors in human tissue. The system uses a convolutional neural network, specialized long short-term memory units, and self-adapting attention tools to focus on the most important visual and time-based clues. The team found that adding physical laws gave their model enhanced accuracy in forecasting radar signals, providing more precise predictions for bridge decks and civil infrastructure.

Why it matters

This helps civil engineers spot hidden damage in bridges and other structures before they fail. Safer infrastructure means fewer accidents and lower repair costs.

Questions answered