A shape optimization method for resistance reduction of local piping components with multiscale validation

Reducing the flow resistance of local components in building transmission and distribution systems is a key pathway to building energy conservation. Here we propose a novel low-resistance optimization method applied to U-bend shape design, with the minor axis a, major axis b, and offset c defined as shape features. The sample size is 150, and the parameter ranges are as follows: [50 mm, 150 mm] for a, [50 mm, 150 mm] for b, and [-30 mm, 30 mm] for c. On this basis, five representative machine learning regression modeling paradigms, including ridge regression, support vector regression, random forest, multilayer perceptron, and Gaussian process regression, are systematically compared, and the model with the best predictive performance is selected as the surrogate model for U-bend optimization. The proposed method is validated through full-scale experiments, numerical simulations, and turbulent energy dissipation. The results show that within a Reynolds number range of 1.0 × 10⁵ to 2.4 × 10⁵, the optimized U-bend achieves a resistance reduction rate of 13-24% relative to the traditional circular U-bend. This study provides a reference for the low-resistance design and energy-saving optimization of building transmission and distribution systems.