Digital Twin and Data-Driven Remaining Useful Life Prediction of Gearbox

Traditional approaches for predicting the remaining useful life (RUL) of gearboxes often face challenges in integrating physical and virtual data, leading to reduced prediction accuracy and an increased risk of system failure. To realize reliable RUL prediction of gearbox, this paper proposes a novel approach called DT-SegRNN, which combines Digital Twin (DT) simulations with the data processing power of a Segment Recurrent Neural Network (SegRNN) to enhance RUL prediction. The key innovation of the DT-SegRNN model lies in its use of a DT system to simulate the gearbox’s lifespan under various operating conditions, generating a theoretical RUL. This theoretical RUL is then integrated with actual RUL data derived from experimental observations using a SegRNN model. To further improve prediction accuracy, the paper employs the Central Particle Swarm Optimization algorithm to merge both theoretical and actual RUL values. The DT-SegRNN model achieves a prediction accuracy exceeding 99.5%, outperforming traditional methods by more than 10.5%. In addition to its superior accuracy, the DT-SegRNN model is highly adaptable to a wide range of industrial machinery, offering a robust framework for proactive maintenance and minimizing unplanned downtime.