High-resolution ultrasonic waveform analysis for decoupling physical effects in multi-damaged and healing in concrete...

Characterizing complex microstructural changes in heterogeneous media like concrete presents a significant challenge: multiple physical effects (e.g., thermal microcracking, material porosity, moisture content, and subsequent repair mechanisms) generate overlapping acoustic phenomena that conventional analysis methods fail to distinguish and quantify. Specifically, relying on single-descriptor metrics like Time-of-Flight (TOF) inherently ignores the full information capacity of the propagating ultrasonic waveform. This study introduces an advanced ultrasonic signal processing methodology designed to achieve a high-resolution, quantitative assessment of these multi-parametric effects, including the evaluation of thermal damage, moisture effects, and material healing. The core innovation lies in Vibro-Acoustic Feature Engineering, where a specialized script extracts 79 unique signal features from the time, frequency, and time-frequency domains to capture subtle variations within the full 55 kHz waveform. To interpret the underlying physical effects driving these signal changes, an XGBoost model is coupled with SHapley Additive exPlanations (SHAP). This approach enables quantitative decoupling, defined as the ability to distinguish the relative influence of multiple physical parameters on the observed signal characteristics, for inherent material porosity, thermal damage, moisture content, and healing progression. The SHAP analysis not only confirms the method's analytical power but also provides deeper physico-acoustical insight by quantifying the specific influence of each parameter on key signal features, such as high-frequency attenuation, time-frequency energy distribution, and signal entropy. Furthermore, the integration of deep learning-based crack quantification (Adaptive Crack Profiling Network) with machine learning-based feature interpretation creates a comprehensive framework for damage assessment and healing validation. This work represents a substantial contribution to data-driven ultrasonics and physical acoustics, enhancing the sensitivity and interpretive power of ultrasonic tools for complex material characterization and monitoring repair efficacy.