Reliability-adaptive probabilistic imaging and severity assessment of axial cracks in steel pipelines

Abstract

Axial crack detection in steel pipelines remains challenging because the crack-reflected guided-wave response is usually weak, frequency-dependent, and susceptible to dispersion, modal interference, sensor-coupling variation, and environmental noise. To address this issue, a reliability-adaptive probabilistic imaging and semi-quantitative severity-assessment framework is proposed using multi-frequency ultrasonic guided-wave signals. A T(0,1)-dominated torsional guided-wave packet is selected as the target response, and dispersion analysis is used to support mode selection and reflection-window prediction. Time-domain, frequency-domain, and time-frequency-domain features are extracted from the crack-reflection window to characterize the defect response from complementary perspectives. For each excitation frequency, a single-frequency probability image is constructed by combining reflected-response strength and travel-time consistency. A reliability-adaptive fusion strategy is then developed to integrate the probability images obtained at different frequencies, where the contribution of each frequency is determined by echo quality, inter-sensor consistency, and modal purity. Equal-weight fusion and SNR-weighted fusion are introduced as baseline methods to evaluate the contribution of the proposed reliability-adaptive weighting strategy. A confidence-enhancement step is further applied to suppress uncertain secondary hotspots and improve noise-interference stability under the tested controlled condition. Experiments were conducted on a Q235 steel pipe specimen containing three axial surface crack depths. The results show that amplitude- and energy-related features increase progressively with crack depth, indicating their sensitivity to severity variation under the present configuration. Compared with equal-weight and SNR-weighted fusion, the proposed method produces more compact probability distributions, lower background interference, and improved localization concentration around the actual crack region. The posterior-distribution-based severity index provides effective semi-quantitative grading of mild, moderate, and severe crack cases. The present study should be interpreted as controlled-condition verification of the proposed framework, while broader validation with different crack geometries, crack locations, sensor-coupling states, and environmental conditions is required in future work.