A multi-objective optimization strategy for a four-DOF robotic arm based on an improved artificial lemming algorithm and BP neural network
Abstract
This study presents a multi-objective optimization framework for the highly nonlinear structural design of a four-axis robotic arm. It integrates Finite Element Analysis (FEA), an improved Artificial Lemming Algorithm (TGCALA), Backpropagation Neural Network (BPNN), and Multi-Objective Exponential Distribution Optimizer (MOEDO). TGCALA incorporates Tent Map chaotic initialization, Golden Sine time-varying parameters, and crisscross optimization, achieving faster convergence and greater stability than existing methods. For further lightweight design, a nonlinear multi-objective model is built. The structural optimization is organized in two stages: a first stage performs single-objective weighted-sum sizing of five macro-dimensional variables ( L 1 , L 2 , h 1 , h 2 , b ) to fix the overall arm geometry, and a second stage performs Pareto-based multi-objective optimization of four cross-sectional parameters ( R , W , L , T ) of the hollowed arc-shaped links. The model takes four cross-sectional geometric parameters of the robotic arm as design variables. Its goals are to minimize total mass, maximum equivalent stress and maximum deformation. A TGCALA-optimized BPNN surrogate model, trained with Central Composite Design (CCD), Latin Hypercube Sampling (LHS), and Gaussian White Noise (GWN) augmentation, yields RMSE of 0.003880 (training) and 0.004744 (testing). This surrogate is coupled with MOEDO to generate the Pareto front. Compared with the initial design, the optimized structure achieves a 9.37%–11.81% reduction in maximum equivalent stress, a 3.30%–6.84% reduction in maximum deformation, and a 2.48%–2.52% reduction in total mass. Vision-based repeatability tests confirmed that the optimized links maintain end-effector positioning errors within ± 0 . 35 mm ( 3 σ ) over 50 cycles, satisfying ISO 9283 Class 2 standards and demonstrating practical industrial applicability.