Failure prediction in advanced materials using unsupervised translation with inheritance from the microscopic images target domains

Unsupervised image-to-image translation emerges as a revolutionary tool in materials science, particularly for studying damage and fracture phenomena without destructive experimentation. In this work, we present a framework based on CycleGAN enhanced with attention mechanisms, specifically designed to translate images of tantalum carbide (TaC) obtained through focused ion beam scanning electron microscopy (FIB-SEM), a material renowned for its thermal and mechanical resistance, from one state to another with simulated structural damage. Our method generates realistic representations of cracks, fissures, and fracture patterns directly from microscopic images, preserving the base microstructure while adapting textures and morphologies to the "damaged" domain. The architecture incorporates attention modules that identify and highlight critical regions prone to failure, demonstrating that Cycle-GAN significantly improves the quality and physical coherence of translated images. This enables the simulation of material behavior under extreme conditions, such as mechanical loads or fatigue cycles, which are crucial factors in high-demand applications like aerospace, nuclear, and advanced engineering industries. By eliminating the need for repetitive destructive testing, this approach not only preserves valuable samples but also allows for the exploration of hypothetical damage and failure scenarios without altering the original material. Furthermore, by simulating phenomena such as crack propagation, the model facilitates the identification of weak points in the microstructure, thereby optimizing the design of stronger and more efficient materials. The adaptability of the framework makes it a versatile tool for analyzing a wide range of material systems and microscopy techniques, solidifying the role of machine learning as a key tool in advanced materials characterization. This work establishes a paradigm for data-driven simulation, integrating artificial intelligence with advanced microscopy to accelerate the development of materials optimized for critical applications. It represents a crucial step toward a predictive, sustainable, and globally oriented materials science, aimed at addressing pressing technological challenges.