Automated system form concrete damage classification identification using pretrained deep learning model

Abstract

The main objective of this project is to create a machine learning-based model for detecting cracks in concrete surfaces. In terms of inspection, the proposed model is meant to assess the percentage of automation in identifying and classifying on concrete surfaces. A deep learning convolutional neural network (CNN) image classification algorithm is used in the proposed crack detection model. The image dataset was collected by the search engine (Google) which consists of corrosion, cracks, honeycomb and non-damage concrete. The images on surface concrete defects were selected, and divided into a training set and testing set, and preprocessed through the transfer learning using the deep learning approach. Deep learning allows for the creation of a concrete crack detecting system that can account for a variety of situations. In particular, the type of deep learning model used was 3 types which is GoogLeNet, ResNet-50 and AlexNet as the basic development of the model. The function of model parameters including learning rate, max epochs, validation frequency, and training dataset size was studied. The validation accuracy was measured in each experiment to determine the best outcome. ResNet-50 outscored the AlexNet and GoogLeNet networks in terms of accuracy, according to the results of the comparison. The best experiment for the dataset utilized in this study provided a model with an accuracy of l00%, demonstrating the promise of deep learning for concrete defects identification. The development of machine learning for an automated system to inspect concrete flaws will improve the engineering scope, economics, and environment of the construction industry. As a result, the use of an automated system might lower the cost of maintenance and rehabilitation. During the inspection, this technology might help minimize the quantity of hazard and unsafe approaches