A novel principal components analysis (PCA) method for energy absorbing structural design enhanced by data mining

A PCA based structural design methodology is proposed and applied to the vehicle structure crashworthiness design. The aim of this approach is to develop structures with complex geometry to satisfy the energy absorbing design requirements using reduced number of design variables. This method is assisted by data mining technique to discover the implicit interrelationship of these variables and generate corresponding design rules. In this approach, the surface of structure is described by the control points, in the form of non-uniform rational basis spline (NURBS). The large control point dataset is compressed using PCA technique, which is able to reduce high-dimensional data by expressing them with a set of linearly uncorrelated orthogonal basis, i.e. Principal Components (PCs), together with corresponding Principal Components Scores (PCSs). By changing the value of the PCSs, the geometry of the part can be modified. Through this process, instead of directly handling a huge number of geometry control points, one can perform the design by adjusting the values of a small number of PCSs, and thus the computational cost is significantly reduced. As a case study, the vehicle frontal side rail (known as an S-shaped beam) is designed using this method. After the design is complete, a data mining process is performed, to explore the implicit interrelationship between design variables (i.e. PCSs), and generate design rules to guide the design procedure. The results suggest that the PCA approach can be used to design a complicated structure with irregular shape effectively and efficiently, and overcome the weakness of conventional design methods, i.e. limited capability of handling high-dimensional design variables and big design datasets. The subsequent data mining process enhances the design procedure by revealing some critical interrelations between parameters and generating design rules for practical applications.