Interval theory-embedded data-driven identification framework for uncertain thermo-elastic parameters

Since system parameters can reflect fluctuations in structural performance, identifying the thermo-elastic parameters based on measured responses is becoming increasingly important for health monitoring of thermo-mechanical systems. To avoid the drawback of traditional probabilistic methods in handling limited experimental samples, this paper proposes a novel interval theory-integrated computational framework for efficient and robust identification of uncertain thermo-elastic parameters. For the coupled thermo-mechanical problem, the thermo-elastic governing equation is derived and the thermal stress effect is discussed. In view of the limitation of extremum searching in capturing potential supplementary data, a confidence-based unbiased interval estimation method is introduced to quantify experimental response bounds of limited experimental samples. Subsequently, a gene expression programming support vector regression (GEP-SVR) metamodel is constructed to replace the full-scale finite element simulations, thereby alleviating the computational burden of the nested dual-loop optimization in interval parameter identification. The effectiveness of the proposed framework is demonstrated through three case studies. Numerical results show that the proposed method achieves identification errors below 3.0% while improving computational efficiency by 87.08% compared to full-scale finite element simulation, providing a practical and efficient tool for uncertainty-aware parameter identification of thermo-mechanical systems.