GPU-Accelerated Probabilistic Lifetime Analysis of High-Temperature Components Using Damage Mechanics Models

Abstract The lifetime analysis of high-temperature components using damage mechanics models is both computationally demanding and time-consuming, particularly when finite element results must be considered. The complexity increases further when probabilistic lifetime predictions are needed, as numerous simulations with slightly varied parameters or loading scenarios become necessary. In this work, we perform a high-temperature probabilistic lifetime analysis based on a damage mechanics model, formulated through a system of ordinary differential equations and accelerated using GPU parallelization. In our tests, this approach achieved about a 46-fold reduction in computation time, efficiently solving tens of thousands of differential equations in parallel and incorporating the results into the probabilistic evaluation. We demonstrate this GPU-accelerated method using service-like high-temperature loading profiles, accounting for statistical uncertainties in strain amplitudes and dwell times. Additionally, epistemic uncertainty in material damage evolution is modeled using a bootstrap parameter distribution. This method allows lifetime assessors to evaluate purely strain-controlled loading profiles, including the generation of probabilistic strain-life curves based on damage mechanics models, as well as arbitrary strain- and stress-time curves from thousands of nodes in finite element simulations, all within an acceptable timeframe and during the design stage of components. Finally, the reliability, expressed as the probability of crack initiation, is computed for various loading profiles, facilitating its application in risk analysis.