Health monitoring of structures is essential to maintain optimal performance during their service life. The main classical structural health monitoring techniques are visual inspections and non-destructive testing. As classical methods have limitations, vibration-based damage detection has become popular among researchers. However, vibration-based techniques have challenges arising from data inaccuracies due to noise and other environmental factors. Advances in computational power have enabled the use of various techniques to overcome these challenges. In this context, vibration-based damage detection of simply supported concrete beams using measured acceleration responses is proposed by incorporating Artificial Intelligence (AI) techniques. The beam was numerically simulated using ABAQUS. The acceleration responses were obtained through dynamic implicit time history analyses. Frequency responses of the beam were extracted using the Fast Fourier Transform (FFT) of the numerical acceleration responses. 3500 multiple damage scenarios were numerically simulated by altering the elastic modulus to create a database of acceleration responses. The damage locations and the damage severity of the beam were identified using Artificial Neural Networks (ANN), particularly Convolutional Neural Networks (CNN) and Multi-Layer Perceptron (MLP). The inputs to the ANN consisted of Frequency Response Functions (FRFs) measured at eleven locations on the top surface of the beam. The damage severity, the target output of the ANN model, was computed as the percentage deviation of the elastic modulus of the damaged beam compared to the undamaged beam. The coefficient of determination (R2) for training, testing and validation data sets of CNN and MLP models were 0.99 and 0.95 respectively. According to the R2 values of the machine learning models, the CNN using the Tanh activation function demonstrated higher accuracy across the training, testing, and validation sets compared to the MLP model.