Machine Learning Model for Wall Crack Trend Analysis

Ensuring structural safety necessitates the timely detection of cracks in building walls. However, traditional crack detection methods are often labor-intensive and subjective, yielding inconsistent results. In this research, we propose an automated approach for wall crack detection using machine learning techniques. Our method involves image acquisition, pre-processing, and crack detection, employing supervised and unsupervised machine learning algorithms, specifically Convolutional Neural Networks (CNNs). This approach offers heightened efficiency, objectivity, and consistency in detecting wall cracks and can be seamlessly integrated into existing structural health monitoring systems for continuous and early detection. Furthermore, our paper provides a thorough review of recent machine-learning studies dedicated to wall crack trend analysis. This review encompasses diverse model types, features, and evaluation metrics, addressing challenges and suggesting future research directions. The combined insights from our proposed approach and the reviewed studies present a promising solution for efficient and accurate wall crack detection and trend analysis, contributing to advancements in structural health monitoring. In our evaluation, three prominent CNN architectures—ResNet50, InceptionV3, and VGG16— were tested on a crack detection dataset. Results revealed that ResNet50 achieved the highest accuracy at 99.9%, establishing it as the most suitable model for our crack detection and monitoring. This study not only enhances the understanding of effective crack detection methodologies but also lays the groundwork for further exploration and optimization in the field of structural health monitoring.