关键词: 数字孪生, 实时仿真, 代理模型, 神经网络, 风系统

Abstract: Digital twin technology has opened a new technical pathway for realizing virtual sensing in pneumatic systems. However, the simulation speed of complex system mechanism models is generally slow, making it difficult to meet the real-time requirements. A pipe-decoupling-based digital twin modeling method is proposed for pneumatic systems, which virtually cuts the pipelines between air supply and air-consuming devices, thereby decoupling the system into independent subsystems. Taking pressure and temperature at the cut interfaces as inputs, and mass flow rate and enthalpy flow rate of the air-consuming devices as outputs, neural network surrogate models are constructed. A typical pneumatic device operating in an “inflation-pressure maintenance-exhaust” cycle is used as a case study, and training data are generated through mechanism simulations using AMESim. Dynamic neural network surrogate models were obtained by training recurrent neural network(RNN) and long short-term memory(LSTM) networks, using two sets of training datasets exported from AMESim with time steps of 0.1 s and 0.01 s. The results demonstrate that the LSTM surrogate model trained on the dataset with a 0.01 s time step exhibits superior accuracy and effectively reproduces the pressure drop trajectory of the main air reservoir during the operation of the air supply equipment. In simulation scenarios involving multiple devices (120 case models), the total computation time of the surrogate models is significantly lower than that of mechanism-based simulations, demonstrating the capability for real-time simulation.

Key words: digital twin, real-time simulation, surrogate model, neural network, pneumatic system