Crack detection and categorisation on steel surfaces using machine learning techniques

The improved methods for fracture localization and detection in civil construction and other industries have led to an increase in the prevalence of crack detection. It is crucial to identify and maintain the integrity of cracks on steel surfaces to ensure structural safety. Conventional gradient-based and evolutionary algorithmic methods are crucial for detecting and assessing damage. Deep learning methodologies are being utilized more frequently in the field of structural damage identification. We train and evaluate the photos at varying ratios, utilizing CNN-based ResNet-50 and AlexNet algorithms. Initially, we constructed the training dataset for the model and classified the damage into three categories: steel beam, steel plate, and corroded steel. This study employed two neural networks, ResNet-50 and AlexNet, to classify crack images and identify damages. Additionally, train the constructed CNN using images with a resolution of 224 × 224 pixels for ResNet-50 and 227 × 227 pixels for AlexNet. Upon completion of the training and validation processes for ResNet-50, the peak average accuracy was attained utilizing 80% of the training dataset. Similarly, we achieved the highest accuracy with 80% of the training data after conducting training for AlexNet.