<p>Rapid start-stop pitching maneuvers in furnace-front operation robots frequently induce severe vertical vibrations in the manipulator arm, compromising structural integrity and operational stability. To mitigate these vibrations, we propose a novel inverse design method for tuned mass dampers (TMDs) utilizing a tandem artificial neural network architecture. A specialized TMD structure comprising a rubber element and an iron casing is designed via theoretical analysis, featuring ease of installation and frequency tunability through mass adjustment. To ensure high-fidelity training data for the artificial neural network, a small-scale prototype is constructed and validated using a multi-physics approach combining theoretical calculation, finite element simulation, and frequency response testing. Subsequently, a tandem neural network model, integrating a forward frequency prediction network and an inverse structural parameter prediction network, is established to accurately map target frequencies to optimal structural dimensions. Guided by on-site measurements of the dominant vibration frequency of the robot, a full-scale TMD is fabricated and installed for engineering verification. Field test results demonstrate that the proposed device significantly reduces the peak vibration acceleration in the pitching direction from 17 m/s² to 3 m/s², with negligible impact on non-principal vibration axes. This study provides an effective, intelligent solution for vibration control in heavy-duty robotics and offers a robust framework for the inverse parameter design of complex mechanical structures.</p>