Graphical abstract
A diagram illustrating the process of predicting long-term polymer lifetime using physics-informed neural networks. Panel 1: Short-term aging and physics. A PEEK polymer slab is exposed to oxygen diffusion and thermal aging at elevated temperatures. The oxygen penetrates the slab, causing reaction-diffusion and thermal aging. Panel 2: Physics-informed neural network. A physics-informed digital twin (PINN) is trained with physics and data. The network includes physics constraints embedded in the governing equations, which relate time, temperature, depth, concentration, and property degradation. The loss function combines mean squared error from data and physics. Panel 3: Reliable long-term lifetime prediction. A graph shows degradation percentage over time for three different temperatures (T1, T2, T3). The data-driven artificial neural network (ANN) shows unstable extrapolation beyond 180 days, while the PINN forecast provides physics-consistent extrapolation for up to 10 years.