Design optimization of a four-wheeled robot chassis frame based on artificial neural network

Optimizing robotic structures to endure random fatigue conditions represents a critical challenge in robotic design. This paper introduces an advanced optimization strategy specifically for robotic structures subjected to random base excitations (RBE). The approach aims to enhance structural performance and reliability under unpredictable conditions. The proposed methodology features an efficient problem formulation and employs a meta-model-based optimization (MBO) technique, validated through the analysis of a robotic structure's fatigue life under RBE. Structural vibration responses to RBE are assessed using finite element method (FEM), which includes the computation of participation factors to quantify each mode's contribution to the overall structural response. A frequency transfer function is then developed to link input excitations with the resulting structural behavior. Experimental validation in the laboratory confirms the accuracy of the FEM predictions for structures under RBE. The MBO approach is applied to a robot chassis, aiming to minimize its total mass while respecting constraints on maximum equivalent stress and fatigue life. Random vibrations caused by road roughness are simulated to evaluate the chassis response in terms of power spectral densities (PSD) of Von Mises equivalent stress, calculated for different chassis elements. A database is generated through co-simulation with ANSYS and MATLAB, computing fatigue life using the Dirlik model for various design variables. This database supports the development of a radial basis function (RBF) based neural network meta-model for constraint evaluation in the optimization process. Two MBO strategies, sequential and adaptive, are tested. The sequential approach required a significantly large dataset yet failed to achieve target accuracy, showing an uncertainty of over 43%, rendering it unsuitable. In contrast, the adaptive approach achieved the desired accuracy with an uncertainty of less than 0.22%, using only 2% of the dataset required by the sequential approach. This adaptive strategy enabled a substantial reduction in chassis mass while maintaining fatigue life. Moreover, optimization results showed no significant changes when different optimization techniques are applied.