The work presented hereafter is about the development of a diagnostic system for damage detection, localization and quantification using transmissibility functions. The structural diagnosis is carried out by dedicated deep learning algorithms, which are convolutional neural network (CNN) and autoencoders (AEs). They work together to predict the system structural status and to filter the perturbations introduced by the temperature. The considered Neural Networks (NNs) have been trained thanks to a dataset which has been generated starting from a Finite Element Model of the system. Results have shown good performances of the entire diagnostic algorithm, even under a generical temperature field, providing an accurate predictor at low overall cost.