This project delivers a complete AI solution for detecting structural defects in building façades. The client needed a reliable system to replace manual inspections, reduce processing time and improve documentation quality. The final model achieved over 95% accuracy, enabling fast and consistent defect identification across thousands of images.
Traditional façade inspections were slow, inconsistent and required manual review of large image sets. The client needed an automated pipeline capable of organizing data, detecting multiple defect types and generating standardized reports for engineering teams.
The project provides an end to end workflow: dataset preparation, YOLO based model training and a Windows desktop application for uploading images, previewing detections, editing results and exporting automatic Word reports. The system delivers high accuracy, improved productivity and fully traceable inspection documentation.
Creation of a clean, balanced and fully annotated dataset for the five defect categories. Includes data review, normalization, image standardization, manual labeling of ~1,000 images, YOLO/COCO file generation and train/val/test split.
Training a custom YOLO model on the annotated dataset to detect and classify structural defects. Includes preprocessing, hyperparameter tuning, performance evaluation and sample predictions. Deliverable: model weights, metrics report and training scripts.
Fine tuning the model with additional images and integrating it into a Windows desktop application. The app allows image upload, detection preview, editing and automatic Word report generation. Deliverable: optimized model, full .exe app and documentation.
The project delivers a complete, production ready AI inspection pipeline capable of processing large volumes of façade images with over 95% detection accuracy. The client now benefits from faster inspections, consistent defect classification and fully automated reporting.
• Significant reduction in manual review time
• Standardized documentation for engineering teams
• Higher reliability and repeatability compared to human inspection
• Scalable system ready for future expansion and new defect categories