Photovoltaic (PV) systems hold great promise for reducing greenhouse gas emissions and diversifying electricity generation sources. However, faults and damages in PV modules can significantly reduce energy output. Therefore, the development of automatic and reliable techniques for fault detection and classification is crucial to enhancing the efficiency, reliability, and lifespan of PV systems, while also minimizing operational and maintenance costs. Despite growing interest, few studies have thoroughly examined the classification of a wide range of PV module defects. In this study, we evaluate the effectiveness of data augmentation techniques in improving the performance of a convolutional neural network (CNN) designed to classify six categories, including five defect types and one class representing clean PV modules. This clarification ensures consistency across the manuscript. The variability across and within these categories poses a challenge for automated classification. To evaluate performance, confusion matrices are used to analyze classification accuracy. Using cross-validation, the proposed CNN model achieved an accuracy of 91.46% in anomaly detection, highlighting its potential to support intelligent fault detection in PV systems.