Digital Twin Virtual-Real Synchronization for Aeroengine Gas Path System Based on Deep Reinforcement Learning

It is crucial to monitor the operational status of aeroengines by using the digital twin technology to realize virtual-real synchronization for the gas path system. The challenge is to accurately monitor deep parameters in real-time during synchronization, although existing digital twin technologies have made good progress in monitoring shallow parameters. This study proposes a virtual-real synchronization method for digital twins of an aeroengine gas path system (AGPS) based on deep reinforcement learning (RL). First, the parameters are divided into directly measurable parameters (DMP) and nondirectly measurable parameters (NDMP). Then, different algorithms are applied to different types of parameters. An unscented Kalman filtering algorithm is utilized to aid in the synchronization of the DMP. An RL approach is employed to train parameter inference models for the NDMP. By combining the two algorithms, synchronization between these two parameter classes is achieved. This method excels by integrating the NDMP into the virtual-real synchronization of the AGPS digital twin, concurrently reducing the inference time for this specific segment. Comparative experiments are conducted, and the results indicate an effective improvement in the accuracy of parameter inference with the proposed method. Simultaneously, it ensures real-time and robust parameter inference.