In particle therapy for cancer treatment, the radiation dose to tissues around the tumour can be reduced by employing a rotating gantry—a mechanical structure allowing the delivery of the particle beam to the patient from various angles. Gantries for ion therapy can benefit from the integration of superconducting magnets to minimize the size and weight of the machine. One significant challenge associated with the supporting system of superconducting elements is related to the management of their accuracy during both alignment and operational phases. Concurrently, heat flow from room temperature to cryogenic temperatures through the supporting system must be restricted as the ratio of power needed to operate the cooling system is around 1000 times the power extracted at cryogenic temperature. The design of the supports must consider the variability of the load during operation, i.e. guarantee accuracy of the cold-mass pose (position and rotation) under its own weight during a $${360\,\mathrm{ ^{\circ }}}$$
360
∘
gantry rotation. A literature review had been done highlighting the possible application of a novel exact-constrained solution for the support of superconducting magnets. Within the framework of the European project HITRIplus (heavy Ion Therapy Research Integration), this study proposes the design and optimization of a support system based on a $$6$$
6
degrees of freedom parallel mechanism (exactly constrained kinematics). A mathematical model is proposed, referred as “error model”, to estimate the accuracy of the pose of the cold-mass due to major unrecoverable errors. The error model estimates the contribution of main error sources: the elasticity of the supports, the elasticity of the vacuum vessel enclosure and the influence of backlash in the joints. An optimization genetic algorithm has been developed and employed to find the optimal configuration of supports that simultaneously increases the accuracy and minimizes heat-loads to the cold-mass. The error model has been validated by finite element analysis, showing its validity for the optimization process. The optimised solution has been compared to solutions that were proposed initially based on common sense, intuition and had been manually refined: the optimized solution shows considerable improvements in the overall accuracy of the system and a substantial reduction of the heat-loads. The optimized solution also implements a pre-load system to eliminate backlash in the joints: this considerably improves the accuracy of the system. The error model presented allows computationally cheaper optimizations and variations of the designed architecture (i.e. variation in cross sections, change in material, change in geometry, implementation of pre-load etc...) with respect to a classic approach based only on finite element analysis. Furthermore, thanks to the kinematics characteristic of the proposed architecture, the 6-legs design clears the way for a more reliable implementation of an automated positioning system with respect to classical over-constrained architectures.