Research On Intelligent Shape-Finding Of Tree-Like Structures Based On Neural Network Algorithm

Considering that the mechanical performance of tree-like structures can be affected greatly by their shape,a reasonable shape-finding method is a key to realize elegant appearance design and ensure good load-bearing performance of tree-like structures.The traditional shape-finding methods can only achieve the similarity in appearance form of tree-like structures,which didn’t consider the influence of shape-finding on the mechanical properties and stability.Therefore,they will produce a large bending moment at the joint,and present stress concentration and tear failure,so that cause potential safety hazards to the overall structure.With the in-depth research on shape-finding of tree-like structures,although the shape-finding methods proposed in recent years consider partly the interaction between shape and force,there are still some shape-finding problems,including smaller application range,lower efficiency,complex operation,excessive uncontrollable factors,non-uniqueness and non-optimum results.Based on the above research background,this paper proposed a novel intelligent shape-finding method by combining the shape-finding concept and updated artificial intelligence neural network algorithm.Firstly,taking a tree-like structure with two-bifurcation type belonging to the single bifurcation structure as a research object,the intelligent form-finding and methodical verification process are completed.This paper generalized the shape-finding unit(namely,triangular unit)corresponding to this structure by researching and analysing the morphological characteristics and force-transfer path of two-bifurcation structures.Based on the mechanical theory analysis of the shape-finding unit and the idea of locating load-bearing center in inverse-hang recursive method,a novel shape-finding concept was proposed.In order to solve the shape-finding problem that is lower efficiency in processing data artificially by applying this shape-finding concept,BP neural network algorithm was introduced and combined with the shape-finding concept to form an intelligent shape-finding method for the tree-like structure with two bifurcation,which can transform the nodal positioning problem of shape-finding design to the adjustment problem of weights in neural network algorithm.Finally,from the perspective of structural geometrical shape and mechanical properties,the shape-finding results of intelligent shape-finding method for the tree-like structure with two bifurcation were compared with the inverse-hang recursive method,and the feasibility of this method was verified by designing the compared cases.Subsequently,considering that the tree-like structure with four bifurcation has unique shape-finding unit and three-dimensional extension,the intelligent shape-finding method for the tree-like structure with two bifurcation is difficult to achieve the shape-finding design for this structure.Taking the tree-like structure with four bifurcation as the research object,this paper determined the shape-finding unit of this structure(namely,four cone element)by analysing and discussing the morphological shape and force-transfer characteristics of the tree-like structure with four bifurcation,and completed the mechanical analysis of this shape-finding unit.Based on the above contents,this paper improved the shape-finding concept of the intelligent shape-finding method for the tree-like structure with two bifurcation by grouping the nodes,and also recompiled of intelligent location program for structural nodes,which formed the intelligent shape-finding method for the tree-like structure with four bifurcation.Finally,according to the shape-finding results under comparative cases,the feasibility of the intelligent shape-finding method was verified with the help of modeling and finite element analysis software.In addition,considering that the mixed bifurcation tree-like structure with bottom two-bifurcation and upper four-bifurcation types has the advantages of wide upper supporting space and large lower applied space compared to the tree-like structure with single bifurcation type,this paper chose the former as the research object.Based on the analysis of the morphological characteristics,force-transfer path and shape-finding units for this structure,this paper improved the shape-finding concept of tree-like structures with single bifurcation type by setting the switch of controlling nodes layer,which solved the problem of locating transformation between the shape-finding units under different bifurcation types.Considering the optimization deadlock problem of BP neural network algorithm in processing data,this paper proposed a new BP-PSO neural network model by introducing PSO algorithm,and combined it with the improved shape-finding concept to form a intelligent shape-finding method for the mixed bifurcation tree-like structure.Finally,the verification of this method was completed by the comparative analysis.The comparative analysis shows that,in terms of nodal zero moment,the intelligent shape-finding method can not only realize the nodal zero moment at the main trunk,but also achieve the zero moment requirements of nodes at all branches.In terms of the maximum equivalent stress at nodes,the equivalent stress analysis of the first-order joint for the tree-like structure with two bifurcation shows that the intelligent shape-finding method can reduce the maximum equivalent stress by 48.46% or29.01%,when the structure supports a horizontal or free-form surface roof under non-uniform load.In terms of the maximum bending stress at branches,the bending stress analysis of four bifurcation and mixed bifurcation tree-like structures shows that the intelligent shape-finding method can optimize the maximum bending stress by nearly 50% in the Z direction(i.e.perpendicular to the plane).Therefore,the intelligent shape-finding method not only has the advantages of universality,efficiency,intelligence and accuracy in shape-finding design,but also shows better optimization ability in shape-finding results.