Novel approach to rapid and accurate temperature-dependent mechanical testing using machine learning

• Introduced a novel test sample for multi-temperature mechanical testing of metallic alloys. • Leveraging full-field strain history and thermal gradient for inferring temperature dependent stresses via machine learning. • Achieved rapid, high-throughput derivation of stress–strain curves across a temperature range. • Validated approach with experimental data, showing high accuracy and efficiency. Understanding the temperature-dependent mechanical properties of metallic alloys is essential for informed engineering design. However, conventional elevated temperature tensile testing is costly, time-consuming, and inefficient. This paper introduces a novel, rapid, and accurate approach for characterizing these properties using a specially designed test sample − MAPS that works at the intersection of full-field optical strain and thermal measurements, and machine learning (ML). The MAPS sample is engineered to exhibit a controlled temperature gradient across its surface, enabling simultaneous acquisition of strain and thermal data at multiple temperatures in a single test. A finite element twin of the MAPS sample is used to generate synthetic training data based on known alloy properties. A Multi-Layer Perceptron (MLP) model is then trained to infer full-field stress distributions from experimental strain and temperature data, allowing the derivation of stress–strain curves across a range of temperatures for novel alloys. The proposed MAPS methodology was validated against conventional dogbone tensile tests conducted on four diverse materials − AA6061-T6, SS301-H12, SS304 and 15-5PH steel. The approach shows good generalizability across material families, showcasing its potential to revolutionize high-throughput temperature-dependent mechanical testing by enabling faster, more cost-effective material characterization for advanced engineering applications.