Novel machine learning approach for shape-finding design of tree-like structures

With the rapid development of machine learning (ML), it is regarded as an effective approach for the shape-finding method of tree-like structures, for optimizing their structural forms. Considering that the existing shape-finding methods are do not easily solve the shape-finding problem of tree-like structures under a non-uniform load or supporting an irregular surface roof, the combination of a shape-finding concept and ML opens a new direction for shape-finding design. This paper presents a novel ML approach by combining a new shape-finding concept with a backpropagation and particle swarm optimization neural network. The core concept is to locate the load-bearing centre of each shape-finding unit by executing a positioning program, which finds the coordinates of the optimized nodes to achieve shape-finding design. Taking a tree-like structure with bottom two-bifurcate and upper four-bifurcate types as a shape-finding example, this paper elaborates the ML shape-finding method and specific steps in detail. Moreover, different shape-finding cases are analysed and compared with the inverse-hang recursive method. Results show that the ML shape-finding method can not only solve the unsolvable problem of shape-finding design of tree-like structures under a non-uniform load or supporting an irregular surface roof but also has higher efficiency than the compared method.