Tire cracks pose a significant safety risk, as undetected defects can lead to severe accidents. Traditional inspection methods rely on manual visual assessments, which are prone to human error. This study proposes an automated tire crack detection system using Convolutional Neural Networks (CNN), leveraging transfer learning techniques to improve accuracy and generalization. A dataset of 600 tire images was collected and preprocessed, including augmentation techniques such as rotation, flipping, and brightness adjustments. The CNN model was trained with different optimizers, including Adam and Stochastic Gradient Descent (SGD), to compare their performance. Experimental results indicate that Adam achieved the highest test accuracy of 78.3% with the lowest test loss of 53%, while SGD required more epochs to reach optimal performance. This study demonstrates the feasibility of deep learning-based automated tire inspection, providing a scalable alternative to traditional methods. Future research should focus on optimizing model architectures, expanding datasets, and integrating real-time detection for industrial applications.