The analysis of microstructures in multiphase steels is highly complex but it is essential to achieve better process control or to optimize their properties. EBSD is a well-suited technique for this application since it provides much more information than classical imaging. Even if martensite, bainite, and ferrite, have similar crystal structures, they can be distinguished on EBSD maps by their local misorientation and packets/blocks/sub-blocks arrangement. However, a quantitative characterization of such complex steels microstructures requires a considerable amount of time, effort and expertise. Therefore, there is a need to develop a reliable tool to accelerate and automates the phase recognition of multi-phase microstructures. In previous studies, we have demonstrated the input of Convolutional Neural Networks (CNNs) for this task [1,2]. The developed CNN with the UNET architecture shows the ability to automatically distinguish martensite, bainite, and ferrite in a low carbon industrial steel with an accuracy of over 90%. The success of using artificial neural networks is strongly dependent on enough and relevant database. In this contribution, we assess the robustness of our model against the variability of the input (steel grade, influence of sample preparation and EBSD acquisition set-up). Additionally, we discuss the amount of data needed to train an accurate model, including the contribution of simulated EBSD microstructures and data augmentation. [1] Martinez Ostormujof T., Purushottam Raj Purohit R., Breumier S., Gey N., Salib M., Germain L. (2021). Deep Learning for automated phase segmentation in EBSD maps. A case study in Dual Phase steel microstructures. Materials Characterization, 184:111638. [2] Breumier S., Martinez Ostormujof T., B. Frincu, Gey N., P.E. Aba-Perea, A. Couturier, N. Loukachenko, Germain L. (2022). Leveraging EBSD data by deep learning for bainite, ferrite and martensite segmentation. Materials Characterization, 186:111805.