Optimum design of a micro-positioning compliant ‎mechanism based ‎on neural network ‎metamodeling

This paper presents a comprehensive investigation of the optimization process of a ‎‎compliant nano-‎‎positioning mechanism based on a high-accuracy metamodel. Within ‎this ‎study, analytical approach, ‎finite ‎element analysis (FEA), and deep neural network ‎‎(DNN) ‎are integrated in order to achieve the ‎optimum ‎design of a parallel 2-degree-of-‎freedom‎ ‎compliant positioner while taking a broad range of ‎factors into ‎account. First, a ‎linear ‎regression analysis is performed on the primary finite element model ‎as a sensitivity ‎‎analysis. ‎Then an analytical model is established to express one of the objective ‎‎functions of ‎design, ‎namely the mechanism working range, as a function of ‎characteristic features: the ‎‎mechanism stiffness ‎and displacement amplification ratio (λ). ‎In the optimization ‎procedure, a single ‎objective constrained ‎particle swarm optimization ‎‎(SOCPSO) algorithm ‎acts on the metamodel to ‎maximize the resonant ‎frequency and ‎provide the minimum ‎acceptable working range. The proposed ‎optimization guideline is ‎‎established for seven ‎different desired working ranges and succeeded in ‎predicting the ‎objective function ‎with ‎an error of less than 3%. The findings provide insights into the ‎‎design and geometric ‎optimization of the ‎mechanical structures. Furthermore, it will be ‎employed as a ‎guideline ‎for implementing DNN for ‎metamodeling in other engineering ‎problems.‎