CausalPD: Joint Causal Discovery and Intervention for Large-Scale Pavement Distress Distribution Data.

Area of Science:

• Civil Engineering
• Data Science
• Infrastructure Management

Background:

• Pavement distress modeling is crucial for infrastructure resilience and maintenance planning.
• Existing methods struggle with sparse, non-stationary data and lack daily granularity.
• Infrequent inspections limit the effectiveness of current pavement inspection systems (PISs).

Purpose of the Study:

• To introduce CausalPD, a novel framework for predicting pavement distress distributions.
• To leverage joint causal discovery and intervention for enhanced pavement health monitoring.
• To improve the accuracy and interpretability of pavement degradation predictions.

Main Methods:

• Utilizing a transformer-based architecture to extract causal patches from historical pavement data.
• Integrating joint causal discovery and intervention to model pavement degradation.
• Mitigating confounding influences to isolate genuine causal signals from noise.

Main Results:

• CausalPD demonstrates superior performance compared to state-of-the-art methods on real-world datasets.
• The framework achieves high accuracy across various scenarios, forecasting horizons, and network configurations.
• Validation confirms robust inductive biases and generalization capabilities of the CausalPD model.

Conclusions:

• CausalPD establishes a promising causal modeling paradigm for infrastructure health monitoring.
• The framework supports proactive maintenance planning by providing accurate distress predictions.
• CausalPD enhances the optimization of pavement inspection systems and preventive maintenance strategies.