The Athena X-ray Observatory, to be launched in the early 2030s, is designed to implement the Hot and Energetic Universe science theme. The X-ray Integral Field Unit (X-IFU) proposed for the Athena mission is a cryogenic imaging spectrometer, offering spatially-resolved high-spectral resolution X-ray spectroscopy over a 5 arcminute equivalent diameter field of view. In this project, we aim to detect and reconstruct the X-ray pulses that will be detected by the X-IFU instrument using Deep Learning techniques over a dataset of simulations with SIXTE (a software package for X-ray telescope observation simulations). We construct and train a Convolution Neural Network (CNN) to differentiate between single, double and triple pulses (i.e., detections with one, two or three pulses in the same record, respectively). We use a hyper-parameter bayesian optimization to select the best CNN architecture. Finally, we present the results of our CNN classification on a test set of simulated pulses, covering all the possible cases in terms of energy ranges and pulse distances, achieving excellent performance metrics.