Research on intelligent prediction of vibration performance and multi-objective optimization using CGWOA-RBF enhanced neural network and MOCGWOA-based optimization algorithm

Advanced vibration analysis and optimization have been integrated into parametric mechanical design to mitigate imbalance, misalignment, and resonance. Traditional finite-element analysis (FEA) entails extensive repetitive setup, leading to inefficiency and suboptimal outcomes. To streamline the workflow, an enhanced Whale Optimization Algorithm (WOA), augmented by chaotic initialization (CI) and genetic operators (GA), was developed. Building on this algorithm, a CGWOA-RBF model and a multi-objective WOA (MOCGWOA) were developed for vibration-performance prediction and structural-parameter optimization. The methods were validated through simulation and prototype testing on the U-frame of a 2D airborne optoelectronic rotary table. With a fixed finite-element (FE) evaluation budget, CGWOA exhibited superior accuracy, stability, robustness, and multi-objective adaptability. The approach reduces three-dimensional FEA cycles to efficient one-dimensional iterative searches and enables rapid closed-loop design. Simulations yielded a mean absolute error (MAE) of 0.0005 for U-frame mass, 0.1173 for the First-order intrinsic frequency, and 0.0021 for Y -direction deformation. Relative to a tuned RBF baseline, the errors were reduced by 61.54%, 31.12%, and 4.65%, respectively. The optimized mass was 8.2837 kg, a reduction of 8.38%; the Y -direction deformation was 0.0471 mm, a reduction of 47.02%; and the First-order intrinsic frequency was 186.51 Hz, a reduction of 5.79%. Experiments demonstrated a 59.1% reduction in overall root-mean-square (RMS) response, an 86.5% decrease in total variance, and a 75.7% reduction in peak power spectral density (PSD), while hazardous speed zones were avoided. Peak gain was reduced, broadband energy was compressed, equivalent damping was increased, and the overall response remained low across the entire speed range. The method enables rapid iteration and comprehensive testing, thereby improving product-level competitiveness. Future work will deepen AI integration to advance model-based forward design and enable fully intelligent, transformative mechanical-product development.