The welding and construction processes for H-type thick-plate bridge steel involve complex multi-pass welding processes, which make it difficult to ensure its welding performance. Accordingly, it is crucial to explore the inherent correlations between the welding process parameters and welding quality, and apply them to welding robots, eliminating the instability in manual welding. In order to improve welding quality, the GMAW (gas metal arc welding) welding process parameters are simulated, using the Q345qD bridge steel flat joint model. Four welds with X-shaped grooves are designed to optimize the parameters of the welding current, welding voltage, and welding speed. The optimal welding process parameters are investigated through thermal–elastic–plastic simulation analysis and experimental verification. The results indicate that, when the welding current is set to 230 A, the welding voltage to 32 V, and the welding speed to 0.003 m/s, the maximum deformation of the welded plate is 0.52 mm, with a maximum welding residual stress of 345 MPa. Both the simulation results of multi-pass welding, and the experimental tests meet the welding requirements, as they show no excessive stress or strain. These parameters can be applied to building large steel-frame bridges using welding robots, improving the quality of welded joints.

The feed system, as a core component of machine tools, has modal characteristics that directly affect machining accuracy and operational stability. However, existing research on modal optimization of feed systems generally suffers from two limitations: the optimization variables rarely involve joint stiffness, and the optimization methods are predominantly based on orthogonal experiments, which are inefficient in handling the complex nonlinear relationship between joint stiffness and natural frequencies. To address these issues, this study approaches the problem from the perspective of joint stiffness and proposes a novel optimization method to improve the modal characteristics of the feed system, thereby reducing the resonance risk caused by an insufficient low-order natural frequency. First, a finite element model of a typical feed system is established and validated through modal experiments, yielding the first five natural frequencies and corresponding mode shapes. Subsequent operating-condition analysis reveals that the maximum screw rotational speed corresponds to an excitation frequency of 88 Hz, while the first-order natural frequency of the current design is only 100.49 Hz. Considering factors such as load fluctuations and environmental disturbances, this narrow safety margin is clearly inadequate and indicates a potential resonance risk. To address this, an improved particle swarm optimization algorithm (IPSO) is proposed. Building upon the standard particle swarm optimization (PSO) framework, IPSO incorporates a Tent map and random solution selection mechanism, nonlinear variation strategies for the inertia weight and learning factors, as well as an orthogonal opposition-based learning mechanism based on lagging individuals and random leaders. The algorithm is compared with classical algorithms (PSO, Genetic algorithm) and recent single-objective optimization algorithms (real-coded genetic algorithm and particle swarm optimization, sine-cosine intelligence algorithm) on seven benchmark functions, and the results demonstrate that IPSO achieves the best convergence performance. On this basis, a neural network model is constructed to capture the complex nonlinear mapping between the key joint stiffness parameters—namely the screw-nut joint, guide rail-slider joint, and bearing joint—and the first-order natural frequency. The IPSO algorithm is then coupled with this neural network surrogate model to perform system parameter optimization. The optimization results show that the first-order natural frequency increases from 100.49 Hz to 114.24 Hz, successfully exceeding the safety threshold and effectively enhancing the resonance resistance of the feed system.