Notes that incomplete or biased training data and added sensor complexity increase costs and cybersecurity demands when implementing autonomous digital twins.
Agentic AI-enabled digital twins for 3D-printed rehabilitation equipment can autonomously perceive, reason, and execute corrective actions, transforming non-destructive testing (NDT) into an intelligent, predictive, and autonomous decision-support system.
However, implementation faces significant hurdles: Data Limitations: Incomplete, biased, or fragmented training datasets lead to inaccurate predictions and limited model generalizability across varied manufacturing processes and patient-specific geometries. Operational Complexity: Integrating embedded sensors for real-time synchronization increases system complexity, implementation costs, and cybersecurity requirements, including risks of data interception and sensor manipulation. * Safety and Validation: Overreliance on simulation may miss rare defects, requiring complementary physical inspection; furthermore, issues like AI hallucinations, limited interpretability, and regulatory uncertainty necessitate ongoing human oversight and rigorous clinical validation.
The material presents a narrative review of how agentic AI can be integrated with digital twins to support intelligent non-destructive testing (NDT) of 3D-printed rehabilitation equipment. It situates this work within the broader challenge of quality assurance for additively manufactured medical devices, where complex geometries, material and process variability, and safety-critical performance make conventional inspection methods less sufficient. The review frames agentic AI as a layer of autonomy in which AI agents can coordinate sensing, interpret inspection data, update digital-twin models, and potentially recommend or execute corrective actions, thereby moving digital twins from passive visualization tools toward active decision-support systems.
A central contribution is its emphasis on the practical barriers that limit autonomous digital-twin deployment. The review highlights that incomplete or biased training data can undermine defect detection, reduce generalizability, and create false confidence in automated assessments. It also stresses that adding more sensors to improve observability increases system complexity, integration effort, and cost, while expanding the cybersecurity attack surface. These considerations are especially important in rehabilitation equipment, where reliability, traceability, and regulatory compliance are not merely engineering preferences but safety requirements.
The work matters because it balances the technical promise of agentic AI-enabled digital twins with the operational and governance challenges of bringing them into real manufacturing and clinical-quality workflows. For a technically literate audience, its value lies in framing the field not only around model capability, but also around data quality, sensor architecture, explainability, validation, and secure system design. In doing so, it provides a useful reference point for researchers and practitioners assessing when autonomous digital twins are likely to add value, what prerequisites must be met, and where further standardization and empirical validation are needed.