SEM Image Characterization Dataset of Hydrogen-Charged High-Entropy Alloys for Artificial Intelligence and Machine Learning Applications
Description
Hydrogen embrittlement presents a critical challenge in the deployment of metals in hydrogen-exposed environments, where the interaction between hydrogen and the metal gives rise to complex failure mechanisms. High-entropy alloys (HEAs) have been identified as an alloy arrangement with the potential to reduce the impact of embrittlement. However, studies are yet to unravel optimisation mechanisms to achieve a fully resistant alloy architecture. Traditional assessment of hydrogen-assisted fracture relies on manual expert interpretation, which is time-consuming, subjective and potentially difficult to standardize. This dataset presents a multimodal Vision-Language Model based for fractographic analysis of HEAs and interpretable assessment of hydrogen embrittlement. The dataset contains 237 High Entropy Alloy (HEA) experimental samples, with 1,343 Large Language Model instructions. For Artificial intelligence and machine learning training and validation, the dataset was divided into training (80%), validation (15%), and test (5%) folders. In each folder, the authors included metadata (in .jsonl) describing the "Instruction", "Condition", "Input", "Output" and "Response" for every SEM print Image ID and Image Type (Hydrogen-Charged under Gaseous and Electrochemical Conditions). The visual HEA prints and corresponding metadata were read as the characterization of each HEA scan in natural language for utilization with any language model.
Files
Steps to reproduce
The dataset contains a structured experimental sample detailing hydrogen exposure conditions, chemical compositions, mechanical properties, microstructural prints, fracture morphology, microscopic techniques, and scan resolutions. The dataset contains hierarchical headers and multi-unit formats. Structured records were unpivoted and transposed from the horizontal study-by-study experimental samples into a long-form relational format where each sample vector contains linked experimental metadata and is ordered by image ID. • Hydrogen exposure conditions: Charging type, concentration, pressure, current density, temperature, duration and desorbed H2. • Mechanical properties: Yield strength, ultimate tensile strength, elongation and other mechanical properties. • Chemical composition: HEA and percentage ratios. • Pre-processing history: Manufacturing method, heat treatment type, cooling rate and prior deformation. • Microstructural features: Grain size distribution, average grain size, phase identification, dislocation density, Kernel Average Misorientation, and other microstructural features. • Mechanical testing: Test type and initial strain rate. • Fracture morphologies: Fracture mode, fracture characteristics and other morphologies. • Microstructural Prints: Images of post-hydrogenation, microscopic techniques and scan resolutions. The dataset contains post-hydrogenation micrographs with different microscopic techniques. As this research is primarily focused on fracture surfaces after hydrogen exposure, the post-hydrogenation micrographs were filtered to obtain only Scanning Electron Microscopy micrographs. The characterization of each HEA scan in natural language was then organised into "Instruction", "Condition", "Input", "Output" and "Response" in an Excel format and transformed into .jsonl to be easily read by a large language model.