基于神经网络代理模型的板式无砟轨道CA砂浆层脱空损伤识别OA

Void damage identification of cement-emulsified asphalt(CA)mortar layer in the slab track system is crucial for ensuring track safety.A time-domain sparse Bayesian learning method based on a neural network surrogate model was proposed for void damage identification of CA mortar layers.The surrogate model was constructed by the convolutional neural network and long short-term memory network,employing a dual-channel feature mechanism,positional encoding,and residual learning strategy to predict the track slab acceleration responses.In the damage identification process,the surrogate model replaced finite element simu-lation in the model updating.The results demonstrate that the surrogate model achieves the response prediction with a average mean squared error of 0.007 and a average determination coefficient of 0.889.For damage identification,the proposed method can simultaneously identify the location and severity of void damage and quantify the uncertainties of identification results.The damage identification time based on the surrogate model is only 2.2%of that based on finite element model updating.The proposed method significantly im-proves computational efficiency while successfully identifying damage,providing a new technical pathway for real-time health monitoring of the slab track system.