Correct and complete observability of structures that may suffer damage during work operations is not easy to achieve in practice. Most methods only highlight the possibility of the existence of damage in the structure. Determining the correct location of this damage is more difficult. Several localization methods have already been presented by the author in previous research. However, the proper quantification of the size of a defect is much more difficult to evaluate. The observability of a mechanical system is given by the ability of the monitoring system to detect changes in the internal states of the observed system. Unobservability implies the existence of states that cannot be traced at the system's output. The selection of parameters to be monitored for a correct evaluation depends on the shape of the structure, its complexity, the materials used, the proper evaluation of the modal characteristics, the sensitivity of the sensors used, the nonlinearity of the observed processes, the data acquisition and post-processing tools, and last but not least, the researcher's ability to process huge experimental databases with many parameters and sub-structural model tree, etc. This paper presents an innovative method for determining, localizing, and severity assessment of damages in complex structures subjected to vibrations. The method is based on using AI in monitoring certain parameters after a proper development of a database for training the ANN model, the implementation of machine learning techniques, and the evaluation methods for assessing the confidence in the predictions provided by the neural network computing model.