Accurate 3D reconstruction of nanomaterials is essential for studying their physical properties. Focused Ion Beam (FIB) tomography is a preferred method for creating 3D image stacks of micrometer-sized material volumes at nanometer resolution. To achieve valid 3D reconstructions, it is crucial to segment these images using machine learning-based methods, as they help mitigate artifacts in the data. However, supervised machine learning requires a large amount of training data and ground truth, which is challenging because FIB tomography is a destructive technique. While training machine learning models on synthetic data and applying this to real data is possible, it is only partially accurate due to differences in data distributions. Moreover, generating synthetic training data is time-consuming, even with modern computing, because of the complex physical Monte Carlo modeling. This study proposes a machine learning pipeline that reduces the difference in FIB tomography data distribution using domain adaptation techniques and introduces a novel method for quickly generating synthetic data by considering physical effects without Monte Carlo simulations.