Characterizing the steady-state creep properties of materials using a curved specimen and AI-driven inverse finite element analysis

• An AI-assisted approach was proposed for steady-state creep characterization. • Inverse FEM and AI algorithm minimized experimental costs and time. • A curved specimen design was utilized to reduce the number of required creep tests. • A single test covered multiple stress levels for accurate creep characterization. • The results from the hybrid approach matched the results from standard creep tests. Characterizing the creep properties of materials is a time-consuming and costly procedure. Efforts have been made to develop methods that reduce experimental costs and enhance the accuracy of creep property determination. Artificial intelligence (AI) has emerged as an efficient tool in various fields, including material property characterization. This study introduces a method for determining the steady-state creep properties of materials using a curved specimen design compatible with uniaxial creep test machines. Unlike standard specimens, this specimen design accounts for the effect of stress level range on test outcomes rather than a single stress level, minimizing the number of tests required for creep characterization. An inverse finite element method, integrated with a particle swarm optimization algorithm, is applied to the proposed curved specimen to address the challenge of attributing varying stress levels to a single test output. Using this hybrid approach, creep testing of only two curved specimens yielded a stress exponent ( n ) of 9.1 and a creep coefficient ( K ) of 4.9 × 10 - 23 in the Norton-Bailey constitutive model—showing excellent agreement with reference values obtained from five standard uniaxial tests ( n = 8.1, K = 5.2 × 10 −23 ).