Crack detection poses a common challenge in structural health monitoring (SHM). Often conducted manually, this process is prone to errors. The use of images offers significant advantages by eliminating the need for direct contact with the structure and providing more comprehensive coverage. Furthermore, machine learning techniques have demonstrated effectiveness in image classification, identifying the presence or absence of damage. This study implements a machine learning pipeline consisting of preprocessing, feature extraction using Principal Component Analysis, model creation for Support Vector Machine, Decision Tree, and K-nearest neighbors, hyperparameter optimization, and results analysis in the task of classifying images of concrete with and without cracks. The best performance achieved in this study was using the SVM model, which exhibited an accuracy of 98.18%, precision of 98.70%, recall of 97.60%, and F1- Score of 98.15%.