Accurate AI-design and additive manufacturing aspects of dental prostheses

Abstract

The development and integration of artificial intelligence (AI) and additive manufacturing (AM) hold transformative potential for advancing dental prostheses design and fabrication in digital dentistry. This thesis addresses critical challenges in traditional computer-aided design (CAD) systems, which rely heavily on manual expertise, and subtractive manufacturing methods, which suffer from material waste and geometric limitations. By synergizing AI-driven design innovations with precision AM techniques, this research establishes an intelligent, streamlined workflow for personalized, high-performance dental restorations.

The study first evaluates commercial AI software against conventional CAD tools, revealing that while AI reduces design time by 40–60%, its morphological accuracy, particularly in occlusal surfaces, lags experienced technicians. To address this, a novel generative adversarial network, ToothGAN, is developed. By incorporating anatomical feature extraction and second-order derivative loss, ToothGAN generates crowns with enhanced pit/fissure accuracy, surface smoothness, and biomechanical integrity. Training on hybrid datasets (natural teeth and technician-designed prostheses) improves feature generalizability but highlights trade-offs between clinical practicality and biological fidelity. Thus, iterative clinician feedback for occlusal validation would be necessary.

In another part of this study, digital light processing (DLP) 3D printing was utilized to fabricate zirconia dental prostheses. Ultra-thin (0.1mm) zirconia prostheses were made by using a high-solid-loading (80 wt%) zirconia slurry with optimized DLP parameters. Thus, sufficient green-body 3-point flexural strength (>16 MPa) and post-sintering biaxial flexural strength (687–1040 MPa) were achieved. Further, this study also evaluated the process control during the DLP to reveal critical factors that could affect accuracy: (1) light scattering limits sub-300 µm features, (2) support structures require >65° overhang angles to minimize deformation, and (3) asymmetric sintering shrinkage (1.323–1.345) for outer and inner XY plane would be essential to compensate the design factor. Another technology – Radiation-assisted sintering (RAS) – was also developed which could reduce densification time to 35–50 minutes while retaining near-nano grains (100–200 nm), though phase stability challenges emerge with reduced fracture toughness (3.01–5.30 MPa·m¹/²).

Clinical benchmarking demonstrates DLP-printed veneers achieve root mean square (RMS) deviations of 48±9 µm, comparable to milled lithium disilicate controls (47±8 µm). Marginal adaptation meets clinical standards, despite light scattering was broadened marginally at edges. Moreover, the developed DLP-printed thin to ultrathin zirconia exhibited superior low-temperature degradation resistance (40.78% monoclinic phase vs. 72.51% in commercial zirconia) and transparency increases rapidly as the thickness decreases, which is critical for aesthetic applications.

Ultimately, this research bridges AI and AM, offering scalable solutions for efficient, patient-specific dental prostheses. Key contributions include the ToothGAN framework, DLP process optimization protocols, and RAS integration, collectively advancing digital dentistry toward closed-loop, automated workflows. Future directions include multi-material printing, real-time clinical data integration, and AI-driven shade matching, paving the way for next-generation dental care.