On the use of the differential evolution algorithm for truss-type structures optimization

In recent decades, bio-inspired numerical algorithms have emerged as an alternative for optimizing the structural design of trusses. The differential evolution algorithm (DEA) has demonstrated both good performance and ease of implementation. However, unlocking the full potential of DEA to address engineering problems poses a significant challenge, necessitating a strategic and informed definition of each component of the algorithm. This research systematically evaluates the influence of defining DEA components on improving the reliability of truss optimization. The algorithm structure of DEA was configured for five aspects: (I) the mutation operator, (II) inclusion of multi-modal techniques, (III) inclusion of parameter control techniques, (IV) definition of the initial population, and (V) local search heuristics. A comprehensive evaluation is conducted to assess the performance of 23 DEA configurations in optimizing eight planar and spatial trusses, varying in size from 10 to 163 elements. Assessment is based on key criteria such as optimal weight, robustness, and computational cost, providing a thorough basis for comparison. The results showed that no tested DEA configuration is the best for all trusses. Instead, the study revealed the presence of recommendable configurations, each tailored to the specific complexities and scales inherent in various truss structures. The integration of multimodal and local search techniques proves particularly advantageous for larger trusses, amplifying the algorithm’s exploratory capabilities to effectively navigate and uncover optimal regions. In contrast, using parameter control technique was more effective in optimizing smaller trusses, capitalizing on the rapid exploration of potential optimal areas in a smaller search space. There was a mutation operator that produced the best results for large trusses and good results for smaller structures. This operator uses the target vector as the base vector and guides its movement from a best-based difference vector, achieving a balance between the exploratory stage and user-defined constraints that promote the exploitation of potentially optimal areas. No discernible impact was observed when the initial population heuristic proposed was used. Finally, this study underscores the feasibility of DEA configurations for optimizing trusses of varying complexities, as proved by a comparison with results from the literature.