DDF-LSTM Model for Efficient Reliability Analysis

Ensuring the long-term safety and performance of engineering systems under various uncertainties requires robust time-dependent reliability analysis. Traditional methods often struggle to effectively combine time-independent random variables with time-dependent stochastic processes, leading to limitations in capturing their complex interactions. This gap can hinder accurate predictions of system reliability over time. To overcome these challenges, researchers have developed the Dual-domain Fused Long Short-Term Memory (DDF-LSTM) model. This novel network architecture is designed to jointly process information from both time-dependent and time-independent domains. It embeds time-independent variables into the initial hidden states of the LSTM and uses a fully connected layer to map both LSTM outputs and time-independent variables to the final output. Furthermore, the DDF-LSTM model incorporates an improved loss function that prioritizes sensitivity to minimum responses, thereby enhancing the precision of failure probability estimations. Once trained, the model allows for efficient Monte Carlo simulations, significantly reducing the computational cost of estimating time-dependent failure probabilities. Case studies validate its superior computational efficiency and predictive accuracy.