| dc.description.abstract |
India requires frameworks for condition assessment and material design that are both infrastructure-scaled and budget-conscious while retaining engineering accountability. This study presents an artificial-intelligence and image-processing decision framework that connects road-damage monitoring, concrete compressive-strength modelling, and observed-space engineering optimisation at an analytical level. A field-scale proof-of-concept is not attempted; instead, three connected modules with demonstrated value are presented using Indian data. For the monitoring task, the India subset of the RDD2022 database is used, which consists of 9,665 road images; the training set contains 6,831 damage annotations. Of these annotations, potholes dominate the frequency at 46.7 percent compared with 1.0 percent for transverse cracks, a 46.9:1 ratio. Bounded region classification was piloted on the 1,221-image India data subset allocated with N-RDD2024. Histogram-of-oriented gradients, local texture, intensity, colour, and bounding-box descriptors yielded 1,826 features. An RBF support-vector machine, selected using five-fold stratified group validation, achieved 79.7 percent accuracy, 0.682 balanced accuracy, and 0.727 macro-F1 on the supplied held-out validation split. The recall for transverse cracking was low at 0.375. For mechanical performance, 188 experimental recycled-aggregate concrete records from IIT Bhubaneswar were grouped into 60 composition groups. With five-fold group-wise validation, Extra Trees were chosen as the regressor, with an out-of-fold R sup(2) of 0.893, an RMSE of 5.85 MPa and an MAE of 4.23 MPa. Curing age, fly ash, total cementitious material, and water-to-total-cementitious ratio were the most influential variables. For the optimisation task, a retrospective search was conducted within the subset of measured 28-day compressive strengths to screen conservative minimum-cement concrete recipes for 35, 45, and 55 MPa targets. It is concluded that a viable Indian framework would combine vision-based low-cost triage, material models with reduced leakage, observed-space optimisation with a transparent safety margin, and mandatory engineer acceptance, rather than relying on black-box AI to replace established inspection, testing, and design practices. |
en_US |