Self-adjusting configuration control method for diagonal cable truss structures using deep learning technology

Abstract

Cable truss structures are composed of edge cables and internal connection members. A common design involves internal members arranged in a continuous oblique pattern, which is efficiently employed in long-span space structures. This configuration primarily governs the equilibrium of free nodes through the curve of the edge cables. However, manually adjusting the prestress to align with the edge cable shape is challenging due to the variable skew angles of the internal members. To address this, this paper proposes a deep learn method to establish a mapping relationship among the edge cable curve, the coordinates of the free nodes, and the prestress distribution. The force density distribution is automatically adjusted by comparing the coordinates of the free nodes with the edge cable shape. Based on the curve characteristics of the edge cable, the polynomial power function is fitted using the least squares method. A deep neural network model with four hidden layers and Adam optimizer, using constrained node coordinates as input and force density distribution as output, successfully achieved the rational configuration of several typical diagonal cable truss structures. The edge cable shape of each cable truss can be controlled by automatically adjusting its configuration. This method is suitable for cable trusses with complex space topology and achieves a high degree of automation.