A generic physics-informed neural network-based framework for reliability assessment of multi-state systems

In this paper, we develop a generic physics-informed neural network (PINN)-based framework to assess the reliability of multi-state systems (MSSs). The proposed follows two-step procedure. first step, recast assessment MSS as machine learning problem using PINN. A feedforward with two individual loss groups is constructed encode initial condition and state transitions governed by ordinary differential equations in MSS, respectively. Next, tackle high imbalance magnitudes back-propagated gradients from multi-task perspective establish continuous latent function for system assessment. Particularly, regard each element an task project task’s gradient onto norm plane any other conflicting taking projecting (PCGrad) method. We demonstrate applications variety scenarios, including time-independent or dependent transitions, where scales increase small medium. computational results indicate that PINN-based reveals promising performance incorporation PCGrad into PINN substantially improves solution quality convergence speed algorithm. • Provide novel on deep learning. Develop Mitigate instability training perspective. Compare PINN’s solutions several state-of-the-art methods different contexts.