温度影响下的锆合金轴向、环向和内压试验OA

The mechanical response of thin-walled zirconium alloy cladding at different temperatures is essential for constitutive description and integrity assessment.The temperature points of existing cladding mechanics data are discrete and often obtained through a single loading method.This makes it difficult to quantitatively compare different loading methods and makes temperature domain inference significantly sensitive to boundary conditions.In this work,we established stress-strain relationships under different temperature and stress conditions,and adopted a unified empirical constitutive relationship to achieve curve parameterization description.On this basis,probability regression and distribution independent uncertainty quantification were introduced to study the predictability of temperature dependence and its boundary risks. This article conducts axial tensile,circumferential tensile,and internal pressure burst tests on unirradiated zirconium alloy cladding tubes within the temperature range of room temperature to 400℃.A multi condition stress-strain dataset was constructed.The unified empirical constitutive relationship was obtained through data fitting,achieving parameterized characterization of the entire curve trajectory.Four characteristic quantities capturing the dominant transitions and curve morphology were extracted,and reinserting them into the fitted relation enables curve reconstruction and curve-level consistency checks. Then,Gaussian process regression(GPR)was used to predict the coupled response of temperature and load characteristic quantities.To characterize the differences in operating conditions,temperature single input models were trained for each of the three loading conditions.Two types of strain targets were modeled using logarithmic transformation,and the original scale after prediction was restored through exponential inverse transformation to ensure that the strain values are positive and improve numerical stability.The generalization performance was assessed via leave-one-out cross-validation(LOOCV).To distinguish interpolation fidelity from boundary sensitivity,the temperature domain was divided into a central interpolation zone and an endpoint zone for partition evaluation.The residual-temperature diagnostics showed markedly amplified and more dispersed residuals at the endpoints,indicating heteroscedastic near-extrapolation behavior. In terms of uncertainty quantification,a conformal prediction based on residual calibration was introduced,and a distribution independent prediction interval with nominal coverage semantics was constructed.Calibration characteristics were tested at multiple coverage levels.The model provided stable characteristic-quantity predictions and accurate curve reconstruction in the interpolation region,whereas both errors and interval widths increased substantially at the endpoints,with deviations propagating into the plastic and peak segments.This framework delineated temperature-domain applicability boundaries and provided an uncertainty-aware,mechanistically consistent basis for inferring cladding mechanical response.