Hydrogen induced cracking (HIC) is a common form of failure in steel exposed to hydrogen rich environments, which always occurs inside material making their detection difficult. Acoustic Emission (AE) is a promising Nondestructive Evaluation technique for crack detection. However, it is challenging to distinguish the signals generated from HIC, corrosions and bubbles. In this study, these mixed AE signals were obtained by electrochemical hydrogen charging on A516 carbon steel with the solution of 0.5 mol/L H2SO4 and 0.5 g/L NaAsO2. Microstructures of the sample’s cross-section were observed to verify the existence of HIC. moreover, separate signals of bubbles and corrosions were acquired from just exposures to 0.5 mol/L H2SO4. After analyzing, key parameters from the AE signals, i.e. the peak frequency, the spectrum gravity, the duration and energy, were identified to differentiate the source mechanisms. Results show that the peak frequency and the spectrum gravity of corrosion signals are higher than those of HIC and bubbles, while the duration and energy of HIC signals are higher than those of the other two kinds of signals. To reduce manual operations, the unsupervised learning algorithms, Principal Components Analysis (PCA) and K-means, were used for clustering, for further understanding of these clusters and their evolution with time.