Abstarct: Structural Health Monitoring (SHM) plays a crucial role in the early detection of damages and the enhancement of safety and durability in civil engineering structures. Shell structures, due to their geometric complexity and diverse applications, require precise and efficient damage detection approaches to accurately assess their performance under various loading conditions. In this study, a comprehensive frxamework for damage detection in shell structures was developed, integrating numerical modeling, signal processing, and deep learning. To this end, the vibrational responses of the structures were obtained in the form of acceleration data and initially transformed into the frequency domain using the Fast Fourier Transform (FFT). Subsequently, the data were preprocessed using standard normalization techniques and data augmentation with Gaussian noise. The processed data were then fed into a convolutional neural network (CNN) with an optimized architecture, including convolutional laxyers, batch normalization, max pooling, flattening, dropout, fully connected laxyers, and a softmax output, to extract relevant features and accurately identify damages. To evaluate the proposed frxamework, three numerical models including a cantilever beam, a stiffened plate, and a truncated cone were modeled as shell structures in ABAQUS, and white noise was applied as a random excitation. Damage scenarios were defined by reducing density and Young’s modulus, and acceleration data in both healthy and damaged states were extracted and fed into the proposed algorithm. The model’s performance was assessed using common evaluation metrics, and the results from confusion matrices, learning curves, and performance tables indicated that the proposed frxamework is capable of detecting damages at various scales, from large damages in the cantilever beam to small and complex damages in the truncated cone, with complete success. Furthermore, transforming the data into the frequency domain provided benefits such as noise reduction, enhancement of dynamic features, and improved model performance in damage detection.