Perpustakaan Digital ITB

In Indonesia, the rising prevalence of upper-limb amputations, driven by traffic accidents, industrial injuries, and diabetes, creates an urgent need for prosthetic wrists that are affordable, locally manufacturable, and biomechanically effective. Conventional devices depend on bulky, rigid linkages and rarely replicate natural wrist kinematics. This study introduces a compliant, two-degree-of-freedom (2-DoF) prosthetic wrist that uses elastic deformation to deliver smooth, lightweight, and durable motion. A fully parametric CAD model was scripted in SolidWorks, enabling automated generation of thousands of design variants. Each variant was then analysed in a lights-out ANSYS Mechanical pipeline, where ACT scripts applied meshing, material assignments, boundary conditions, and post-processing without user intervention. Key mechanical outputs, directional stiffness and range of motion (RoM), were extracted to train machine-learning surrogate models. These surrogates were fine-tuned with Optuna hyper-parameter optimisation and further enhanced via stacked ensemble learning. Embedded in a Unified Non-dominated Sorting Genetic Algorithm III (UNSGA-III) multi-objective framework, the surrogates enabled high-throughput optimisation, with design selection guided by hyper-volume contribution to balance stiffness and RoM. Five optimal candidates spanning paediatric to adult wrist sizes, representing user-specific customization across a broad anthropometric spectrum, were fabricated in PETG and validated experimentally. High-fidelity simulations and bench tests confirmed full directional RoM (? 85° flexion, 80° extension, 35° radial, 40° ulnar) with von Mises stresses safely below 35 MPa. Average ANSYS–experiment deviations were ? 3 % in most metrics; larger errors in flexion-final and radial-final moments highlighted areas for future surrogate retraining. The resulting wrists are fully 3D-printable, low-cost, and adaptable across users, with stiffness and size parameters customizable for individual needs. This demonstrates that a simulation-driven, data-centric workflow can deliver advanced prosthetic solutions in resource-constrained settings. Beyond prosthetics, the integrated CAD-FEA-ML-optimisation pipeline is transferable to compliant mechanisms in aerospace, automotive, and general industrial design.