A Six-Layer Framework for RCC Crack Analysis: Engineering Judgment Meets Machine Learning

Abstract: This study presents a comprehensive, field-based diagnostic model for analysing cracks in reinforced cement concrete (RCC) structures, based on empirical observations across forty buildings in Tamil Nadu and Karnataka. Despite adherence to structural design codes, these structures continue to exhibit cracking due to a convergence of causes such as material shrinkage, poor detailing, substandard construction practices, and environmental stressors. Conventional diagnostic approaches, which rely heavily on permissible crack width thresholds, often neglect crucial indicators like orientation, spatial distribution, and progression behaviour. To overcome these diagnostic limitations, this study introduces a six-level analytical framework that integrates descriptive statistics, frequency pattern analysis, chi-square and correlation testing, risk profiling, and machine learning-based decision tree classification. Results demonstrate that crack orientation and structural location are significantly correlated with risk levels (χ² = 109.32 and 134.37, p < 0.001). The findings also reveal a weak inverse relationship between crack width and length (r = –0.13), which questions the assumption that wider or longer cracks necessarily indicate greater severity. Crack widths ranged from 0.4 mm to 20 mm (mean = 3.51 mm), and lengths ranged from 23 cm to 1000 cm (mean = 148.6 cm), covering a wide range of structural responses. High-risk cracks— defined as those exceeding 5 mm in width or 500 cm in length—were typically found in slabs and retaining walls, suggesting vulnerability to moisture ingress and flexural stresses. The predictive model achieved 75% accuracy and 100% precision in classifying high-risk cracks, highlighting the potential for integrating data-driven tools into structural diagnostics. This work presents more than just an improvement in inspection methodology; it represents a paradigm shift in understanding cracks as structural indicators rather than superficial defects. The study urges the revision of design codes like IS 456:2000 and ACI 224R-01 to accommodate multifactorial diagnostic criteria and recommends training for engineers in statistical and predictive diagnostic skills. The framework proposed here also lays the groundwork for further integration with sensor-based monitoring systems and geotechnical calibration models, particularly in the context of climate variability and rapid urban development.