SCCB-U-Net: Convolutional neural network for real-time analysis of 3D mechanical properties of umbilical

The mechanical properties of umbilical are mostly analyzed by traditional numerical simulation, which is time-consuming and not easy to converge, seriously affecting the analysis efficiency of mechanical properties, which hardly guarantees its reliability and safety. Inspired by the idea of deep learning in solid mechanics computation, a novel 3D convolutional neural network model, Spatial and Channel Convolutional Block U-Net, (SCCB-U-Net), is proposed in this paper to realize real-time and efficient mechanical properties analysis of umbilical. Simultaneously, the Spatial Channel Convolutional Block (SCCB) is proposed to be embedded layer by layer into the network, consisting of the Spatial Convolutional Unit (SpU) and the Channel Convolutional Unit (ChU), which performs convolutional feature compression for spatial redundancy and channel redundancy, respectively. Taking the mechanical properties analysis of umbilical under tension-bending coupling as an example, the effect of hyperparameters such as optimizer, learning rate etc. on the performance of the proposed model are discussed in detail in this paper. From the experimental analysis, it can be seen that the proposed method can not only obtain the required mechanical properties results of umbilical in about 50s, less than the computation time of the traditional numerical simulation, but also has an accuracy of up to 89.97% compared with other deep learning models, and the prediction results are in good agreement with the error of 5.96%, which has demonstrated the accuracy and real-time performance of the proposed SCCB-U-Net in the analysis of the mechanical properties. Consequently, the proposed method not only can expand the application scope of deep learning in mechanical properties analysis, but also can provide a novel thought and technical tool for the safety monitoring and maintenance of umbilical.