Digital Twin Framework for Interpretable Aero-Engine RUL Prediction via Multibranch LSTM and Unity Visualization

Aero-engines, as pivotal components in aviation, demand robust prognostics and health management (PHM) to mitigate degradation risks and optimize maintenance amid complex operational stresses. Conventional scheduled strategies often yield inefficiencies, prompting the integration of digital twins (DTs) for enhanced predictive capabilities. This study introduces a comprehensive DT-based aero-engine health management framework, stratified into physical, data, model, and application layers, augmented by a Unity-driven 3D visualization module for immersive virtual-real interactions. The physical layer captures real-time sensor data from turbofan components, feeding into the data layer for preprocessing, fusion, and lifecycle storage. The model layer employs a novel physically decoupled multi- branch long short-term memory (LSTM) network with attention fusion, segmenting sensor features (e.g., temperature, pressure) into specialized branches to discern degradation channels while dynamically weighting salient contributions. Outputs manifest in the application layer via interactive dashboards for RUL fore- casting, anomaly alerts, and maintenance simulations. Validated on the N-CMAPSS DS01 dataset, the model attains an R2 of 0.9376 and RMSE of 5.76 cycles, outperforming single-branch LSTM (R2 = 0.7306) and Transformer (R2 = 0.7810) baselines by leveraging physical priors for superior trend capture and interpretability. Unity integration enables real-time 3D mapping of predictions, fostering user-centric diagnostics. This framework advances PHM by harmonizing deep learning with DTs, boosting accuracy, usability, and decision-making in civil aviation. Limitations in multi-condition scenarios underscore avenues for future uncertainty modeling and lightweight adaptations. Note to Practitioners This digital twin (DT) framework advances aero-engine prognostics and health management (PHM) by delivering interpretable remaining useful life (RUL) predictions to combat inefficiencies in traditional scheduled maintenance, which costs aviation billions in downtime and overhauls. Structured across physical, data, model, and application layers, it processes real- time turbofan sensor data (e.g., temperature, pressure) to support proactive decision-making. The core multi-branch LSTM with attention fusion decouples features by physics (e.g., temperature branches), dynamically weighting degradation signals, such as turbine exhaust, for superior accuracy and transparency. On the N-C MAPSS DS01 dataset, it reaches R2 = 0.9376 and RMSE = 5.76 cycles, surpassing the single-branch LSTM (R2 = 0.7306) and Transformer (R2 = 0.7810) baselines by 20–28%. Benefits include 10-15% longer on-the-wing times and reduced inspections. Unity 3D visualization maps RUL onto interactive models, enabling engineers to identify anomalies, simulate maintenance, and receive alerts, bridging AI to actionable insights without requiring expertise. Deployable via edge cloud for avionics, it scales to variable flight sizes; future work will add uncertainty modeling for noisy data. This empowers predictive stewardship, boosting safety and economics.